visualdynamics.core.data

data

Data arrays: time histories, spectra, FRFs, PSDs.

Shared structure: one common abscissa (time or frequency, SI), an ordinate matrix of shape (records, samples), a response DOF string per record ('101X+'), an optional reference DOF per record (FRFs, cross spectra), and per-record unit information.

Units are per record, because one measurement set routinely mixes quantities — a test's time data may hold accelerations, forces, voltages, strains and temperatures side by side. Each record carries:

• ordinate_unit[i]: the unit its values were declared in, or None when the source did not say. reference_unit[i] holds an FRF's denominator unit.
• ordinate_dim[i]: the dimension tag driving display ('acceleration', 'acceleration/force', ..., or 'unknown').
• dimension_hint[i]: what kind of quantity the source said this was, kept only for records whose units are undefined.

A record whose unit is unknown keeps the file's raw numbers untouched; define_units() converts it to SI once and remembers the unit, so a wrong guess can be corrected later by reinterpreting rather than reimporting.

Knowing what a quantity is and knowing what scale its numbers are on are separate things, and a file can tell us one without the other: a UNV with no dataset 164 still says a channel is an acceleration, but leaves no way to tell g from m/s**2. Such a record imports undefined — raw values, dimension 'unknown', nothing converted — while dimension_hint remembers the claim. It is advisory only: nothing scales by it. It narrows the choices offered when units are defined by hand, labels a plot axis, and survives a round trip back out to a format that can name a quantity without sizing it.

Function type codes follow UFF dataset 58 (also used by sdynpy): 1 time response, 4 FRF, 6 coherence, 9 PSD, 12 spectrum, 24 shock response spectrum, 26 multiple coherence. A specification is a PSD and a shock specification is an SRS — each is the ordinary thing with limits on it, and rides a file as its own type.

Classes:

Name	Description
DataArray	Base class; use a concrete subclass (TimeHistory, Spectrum, Frf, Psd).
TimeHistory	A measurement against time: the record as it was acquired.
Spectrum	A linear spectrum: amplitude and phase at each frequency line.
Frf	Frequency response function: response per unit reference.
Coherence	Ordinary coherence: one response against one reference.
MultipleCoherence	How much of a response all the references together account for.
Psd	Power spectral density: the declared unit is the engineering unit
Srs	A shock response spectrum: the peak an oscillator reached.
Bounded	Limit curves held beside an ordinate, in the same units as it.
Specification	What a random vibration test was controlled to: a PSD, and its band.
ShockSpecification	What a shock test was controlled to: an SRS, and its band.
TransientSpecification	What a transient test was controlled to: a target time history.

Functions:

Name	Description
direction_code	'Z+' -> 3. The inverse of the table dof_string reads.
parse_dof	'101RX+' -> (101, 'RX+'). The inverse of dof_string().
has_phase	Is there imaginary content here, or only the dust of computing it?
channel_quantities	(response, reference) quantities a record's dimension names.
frequency_axis	Read or set how frequency axes are drawn: 'log', 'linear', or
paired_channels	The channels two densities share, as (signal row, floor row,
density_ratio	One density over another, channel by channel, line by line.

Classes

DataArray

DataArray(
    abscissa: ArrayLike,
    ordinate: ArrayLike,
    response_dof: str | Sequence[str],
    reference_dof: str | Sequence[str] | None = None,
    ordinate_dim: str | Sequence[str] | None = None,
    comment: str | Sequence[str] | None = None,
    ordinate_unit: str | Sequence[str | None] | None = None,
    reference_unit: str
    | Sequence[str | None]
    | None = None,
    dimension_hint: str
    | Sequence[str | None]
    | None = None,
    block: str | Sequence[str] | None = None,
)

Base class; use a concrete subclass (TimeHistory, Spectrum, Frf, Psd).

One object holds many records — 36 accelerometer channels, or the 2 592 FRFs of a 36-by-72 matrix — sharing one abscissa. Everything that varies between records is a list of that length, so record i is ordinate[i] measured at response_dof[i], and there is no per-record object to go stale.

Values are stored in SI once their units are known. A record whose units were never declared keeps the file's raw numbers and reports ordinate_dim == 'unknown'; what the file said it was, without saying its scale, is kept beside it in dimension_hint.

Attributes: abscissa: The x axis, shared by every record — seconds for a time history, hertz for anything in the frequency domain. Stored as it arrived: uneven spacing and out-of-order samples are both allowed, because both are real and refusing them at the door would refuse real data. What needs an even step — anything with an FFT under it — asks for one at the point of use and says so when it cannot have it. ordinate: (records, len(abscissa)). Complex where the subclass says so (complex_ordinate), real otherwise. response_dof: What each record was measured at, as a DOF string ('101X+'). One per record. reference_dof: What each record was measured against, for the types that need one — the shaker on an FRF, the other channel of a cross spectrum. None where the type has no reference. block: Which repeat of the same measurement each record is: an average, a run, a shock. A short label, not a time. ordinate_dim: The quantity each record measures ('acceleration', 'force'), or 'unknown' while its units are undeclared. ordinate_unit: The unit its values are in — always the SI one while the dimension is known, since that is how they are stored. None means undeclared. reference_unit: The same for the reference of a ratio, so an FRF record knows both halves of m/s²/N. dimension_hint: What the file claimed a record measures without saying at what scale. Nothing is ever scaled by a hint: it narrows the units offered, labels an axis, and survives export. comment: Free text per record, as the source file carried it.

Methods:

Name	Description
known_dim	What quantity record i holds, whether or not its unit is known.
rename_dof	Give a channel's coordinate a new name, in place.
delete_records	Remove records in place — by index, by DOF, or by capture.
define_units	Declare what the ordinate values are in, converting them to SI.
undefine_units	Take a declaration back, restoring the file's raw values.
column_keys	What tells one record from another besides its response.
record_label	A short label for one record, for a legend or an axis.
record_pair	The (response, reference) record i is between.
log_scaled	Whether this object's magnitude reads on a log axis.
display_abscissa	The abscissa converted into a unit system's own units.
display_ordinate	Ordinate in display units; undefined records pass through as-is.
save	Write this object to a .vdyn file of its own.
plot	Draw every record on one set of axes.
save_plot	Draw the records and write the figure to path.
plot_waterfall	The records spread along a depth axis, coloured by level —

Attributes:

Name	Type	Description
num_records	int	How many records this object holds. One measurement per
units_defined	bool	Whether every record knows what it measures. False while
undefined_records	list[int]	Which records still have no declared dimension — the ones

Source code in src/visualdynamics/core/data.py

def __init__(self, abscissa: ArrayLike, ordinate: ArrayLike,
             response_dof: str | Sequence[str],
             reference_dof: str | Sequence[str] | None = None,
             ordinate_dim: str | Sequence[str] | None = None,
             comment: str | Sequence[str] | None = None,
             ordinate_unit: str | Sequence[str | None] | None = None,
             reference_unit: str | Sequence[str | None] | None = None,
             dimension_hint: str | Sequence[str | None] | None = None,
             block: str | Sequence[str] | None = None) -> None:
    self.abscissa: np.ndarray = np.asarray(abscissa, dtype=np.float64)
    ordinate = np.atleast_2d(np.asarray(
        ordinate, dtype=np.complex128 if self.complex_ordinate else np.float64))
    self.ordinate: np.ndarray = ordinate
    n = ordinate.shape[0]
    if self.abscissa.ndim != 1 or ordinate.shape[1] != len(self.abscissa):
        raise ValueError(
            f'ordinate shape {ordinate.shape} does not match '
            f'abscissa length {len(self.abscissa)}')
    # a channel whose node or direction was never recorded imports
    # as a DOF of '' — see `validate.dofs`
    self.response_dof: list[str] = _dofs(
        self._str_list(response_dof, n, 'response_dof'), 'response DOF',
        allow_unknown=True)
    self.reference_dof: list[str] | None
    if reference_dof is None:
        if self.needs_reference:
            raise ValueError(f'{type(self).__name__} requires reference_dof')
        self.reference_dof = None
    else:
        self.reference_dof = _dofs(
            self._str_list(reference_dof, n, 'reference_dof'),
            'reference DOF', allow_unknown=True)
    # A record is its response plus, sometimes, one more thing. For a
    # measurement against something else that is the reference DOF; for
    # the same measurement repeated it is which repeat — an average, a
    # run, a temperature. `block` holds the short label of that second
    # kind, and nothing about it is time-specific.
    self.block: list[str] | None = (None if block is None
                                    else self._str_list(block, n, 'block'))
    self.ordinate_dim: list[str] = self._str_list(
        UNKNOWN if ordinate_dim is None else ordinate_dim, n, 'ordinate_dim')
    for dim in set(self.ordinate_dim):
        parse_dimension(dim)  # validates
    self.ordinate_unit: list[str | None] = self._opt_list(
        ordinate_unit, n, 'ordinate_unit')
    self.reference_unit: list[str | None] = self._opt_list(
        reference_unit, n, 'reference_unit')
    self.comment: list[str] = self._str_list(
        comment if comment is not None else '', n, 'comment')
    self.dimension_hint: list[str | None] = self._opt_list(
        dimension_hint, n, 'dimension_hint')
    for hint in set(self.dimension_hint):
        if hint is not None:
            parse_dimension(hint)  # validates
    self._fill_si_units()
    self._drop_stale_hints()

Attributes

num_records property

num_records: int

How many records this object holds. One measurement per record, all sharing the object's single abscissa.

units_defined property

units_defined: bool

Whether every record knows what it measures. False while any record is still in the file's own unconverted numbers.

undefined_records property

undefined_records: list[int]

Which records still have no declared dimension — the ones holding the file's raw numbers, awaiting define_units.

Methods:

known_dim

known_dim(i: int) -> str

What quantity record i holds, whether or not its unit is known.

The dimension when the units are defined, otherwise the source's claim, otherwise 'unknown'. Never use this to scale anything — a hinted record's values are raw, and the scale is exactly what is missing.

Parameters:

Name	Type	Description	Default
i	int	Which record.	required

Returns:

Type	Description
str	The quantity the record holds, falling back to its dimension hint when the unit is undeclared.

Source code in src/visualdynamics/core/data.py

def known_dim(self, i: int) -> str:
    """What quantity record `i` holds, whether or not its unit is known.

    The dimension when the units are defined, otherwise the source's
    claim, otherwise 'unknown'. Never use this to scale anything — a
    hinted record's values are raw, and the scale is exactly what is
    missing.

    Parameters
    ----------
    i : int
        Which record.

    Returns
    -------
    str
        The quantity the record holds, falling back to its
        dimension hint when the unit is undeclared.
    """
    if self.ordinate_dim[i] != UNKNOWN:
        return self.ordinate_dim[i]
    return self.dimension_hint[i] or UNKNOWN

rename_dof

rename_dof(
    old: str, new: str, quantity: str | None = None
) -> int

Give a channel's coordinate a new name, in place.

A channel is a coordinate and a quantity — a drive point carries a load cell and an accelerometer at one DOF — and the rename is the channel's: a force labelled at the wrong node moves without taking the accelerometer at that node with it, and the other is changed explicitly if it should be (Brandon, 2026-09-06). The channel moves wherever a record wears it, as a response and as a reference alike: a CPSD's accelerometer is on both sides of its cross terms and is one sensor. A rename that would give two records one identity — two accelerometers at one point — is refused, where it would have made two rows of the grid into one and hidden a record.

Parameters:

Name	Type	Description	Default
old	str	The coordinate as it is, '101Z+'.	required
new	str	The coordinate to give it; normalised the way every DOF is ('101Z' is '101Z+'), and refused when it is not one.	required
quantity	str	Which channel at old: 'acceleration', 'force' … as channel_quantities names a record's factors. Every quantity at the coordinate when omitted — the point rather than the channel, said explicitly.	None

Returns:

Type	Description
int	How many records changed, counting a response and a reference on one record separately.

Source code in src/visualdynamics/core/data.py

def rename_dof(self, old: str, new: str,
               quantity: str | None = None) -> int:
    """Give a channel's coordinate a new name, in place.

    A channel is a coordinate *and* a quantity — a drive point
    carries a load cell and an accelerometer at one DOF — and the
    rename is the channel's: a force labelled at the wrong node
    moves without taking the accelerometer at that node with it,
    and the other is changed explicitly if it should be (Brandon,
    2026-09-06). The channel moves wherever a record wears it, as
    a response and as a reference alike: a CPSD's accelerometer is
    on both sides of its cross terms and is one sensor. A rename
    that would give two records one identity — two accelerometers
    at one point — is refused, where it would have made two rows of
    the grid into one and hidden a record.

    Parameters
    ----------
    old : str
        The coordinate as it is, '101Z+'.
    new : str
        The coordinate to give it; normalised the way every DOF is
        ('101Z' is '101Z+'), and refused when it is not one.
    quantity : str, optional
        Which channel at `old`: 'acceleration', 'force' … as
        `channel_quantities` names a record's factors. Every
        quantity at the coordinate when omitted — the point rather
        than the channel, said explicitly.

    Returns
    -------
    int
        How many records changed, counting a response and a
        reference on one record separately.
    """
    old = str(old).strip()
    (new,) = _dofs([str(new)], 'DOF')
    if new == old:
        return 0
    factors = [channel_quantities(self.known_dim(i))
               for i in range(self.num_records)]
    rows = [i for i, dof in enumerate(self.response_dof)
            if dof == old and quantity in (None, factors[i][0])]
    columns = [i for i, dof in enumerate(self.reference_dof or [])
               if dof == old and quantity in (None, factors[i][1])]
    if not rows and not columns:
        raise ValueError(f'no {quantity + " " if quantity else ""}record '
                         f'at {old!r}')
    moving = {factors[i][0] for i in rows}
    for i, dof in enumerate(self.response_dof):
        if dof == new and factors[i][0] in moving:
            raise ValueError(
                f'{new} already has a {factors[i][0]} record — two '
                'channels cannot share a coordinate and a quantity')
    for i in rows:
        self.response_dof[i] = new
    for i in columns:
        self.reference_dof[i] = new
    return len(rows) + len(columns)

delete_records

delete_records(
    indices: Sequence[int] | None = None,
    *,
    dof: str | Sequence[str] | None = None,
    dim: str | Sequence[str] | None = None,
    reference: str | Sequence[str] | None = None,
    capture: int | Sequence[int] | None = None,
) -> None

Remove records in place — by index, by DOF, or by capture.

Everything a record owns goes with it: its row of the ordinate, its DOFs, units, comment, hint, block — and, on a specification, its limit curves, which would otherwise silently belong to the wrong channels. Removing the last record is refused: an empty data array is not a state anything else here can show.

dof and capture are the selectors a person means — "drop channel 101Z+", "drop the third run" — where indices are the machine's (Brandon, 2026-08-30, reading thirteen indices in the journal where one capture number would have said it). They combine as an intersection, and either combines with explicit indices as a union.

Parameters:

Name	Type	Description	Default
indices	sequence of int	Which records to remove, by position.	None
dof	str or sequence of str	Remove every record at these response DOFs. A drive point carries two records at one DOF — a force and an acceleration — and the DOF alone takes both; dim is how the finer thing is said.	None
dim	str or sequence of str	Restrict to these quantities ('force', 'acceleration', …) — the other half of a channel's identity.	None
reference	str or sequence of str	Remove every record at these reference DOFs — a column of an FRF matrix, where dof is a row.	None
capture	int or sequence of int	Remove these captures — each channel's n-th playing, the numbering capture_indices gives. Time histories only.	None

Returns:

Type	Description
None

Source code in src/visualdynamics/core/data.py

def delete_records(self, indices: Sequence[int] | None = None, *,
                   dof: str | Sequence[str] | None = None,
                   dim: str | Sequence[str] | None = None,
                   reference: str | Sequence[str] | None = None,
                   capture: int | Sequence[int] | None = None) -> None:
    """Remove records in place — by index, by DOF, or by capture.

    Everything a record owns goes with it: its row of the ordinate, its
    DOFs, units, comment, hint, block — and, on a specification, its
    limit curves, which would otherwise silently belong to the wrong
    channels. Removing the last record is refused: an empty data array
    is not a state anything else here can show.

    `dof` and `capture` are the selectors a person means — "drop
    channel 101Z+", "drop the third run" — where indices are the
    machine's (Brandon, 2026-08-30, reading thirteen indices in the
    journal where one capture number would have said it). They
    combine as an intersection, and either combines with explicit
    indices as a union.

    Parameters
    ----------
    indices : sequence of int, optional
        Which records to remove, by position.
    dof : str or sequence of str, optional
        Remove every record at these response DOFs. A drive point
        carries two records at one DOF — a force and an
        acceleration — and the DOF alone takes both; `dim` is how
        the finer thing is said.
    dim : str or sequence of str, optional
        Restrict to these quantities ('force', 'acceleration', …)
        — the other half of a channel's identity.
    reference : str or sequence of str, optional
        Remove every record at these reference DOFs — a column of
        an FRF matrix, where `dof` is a row.
    capture : int or sequence of int, optional
        Remove these captures — each channel's n-th playing, the
        numbering `capture_indices` gives. Time histories only.

    Returns
    -------
    None
    """
    doomed = {int(i) for i in indices} if indices is not None else set()
    if dof is not None or dim is not None or reference is not None \
            or capture is not None:
        wanted_dofs = ({dof} if isinstance(dof, str)
                       else None if dof is None else {str(d) for d in dof})
        wanted_dims = ({dim} if isinstance(dim, str)
                       else None if dim is None else {str(d) for d in dim})
        wanted_references = None
        if reference is not None:
            if self.reference_dof is None:
                raise ValueError(
                    'references are a matrix\'s; this '
                    f'{type(self).__name__} has none')
            wanted_references = ({reference}
                                 if isinstance(reference, str)
                                 else {str(r) for r in reference})
        wanted_captures = None
        if capture is not None:
            ordinals = getattr(self, 'capture_indices', None)
            if ordinals is None:
                raise ValueError(
                    'captures are a time history\'s; this is a '
                    f'{type(self).__name__}')
            ordinals = ordinals()
            wanted_captures = ({int(capture)}
                               if isinstance(capture, (int, np.integer))
                               else {int(c) for c in capture})
        chosen = []
        for i in range(self.num_records):
            if wanted_dofs is not None \
                    and self.response_dof[i] not in wanted_dofs:
                continue
            if wanted_dims is not None \
                    and self.ordinate_dim[i] not in wanted_dims:
                continue
            if wanted_references is not None \
                    and self.reference_dof[i] not in wanted_references:
                continue
            if wanted_captures is not None \
                    and ordinals[i] not in wanted_captures:
                continue
            chosen.append(i)
        if not chosen:
            raise ValueError('nothing matches that selection')
        doomed |= set(chosen)
    bad = [i for i in doomed if not 0 <= i < self.num_records]
    if bad:
        raise IndexError(f'no such records: {sorted(bad)}')
    keep = [i for i in range(self.num_records) if i not in doomed]
    if not keep:
        raise ValueError('cannot delete every record; delete the '
                         'object instead')
    self.ordinate = self.ordinate[keep]
    for name in ('response_dof', 'reference_dof', 'block',
                 'ordinate_dim', 'ordinate_unit', 'reference_unit',
                 'comment', 'dimension_hint'):
        values = getattr(self, name)
        if values is not None:
            setattr(self, name, [values[i] for i in keep])
    for name, values in getattr(self, 'limits', {}).items():
        self.limits[name] = values[keep]
    if self.block is not None:
        # repeated-measurement labels renumber to contiguous —
        # 'avg 1, avg 3' after deleting the second capture is a
        # gap the tree's columns would faithfully show (Brandon,
        # 2026-08-30) — by the merge rule, which is the one
        # implementation of what a block label is: bookkeeping,
        # not identity. Named blocks (an exodus 'KE') survive.
        from .merge import _merged_blocks

        self.block = _merged_blocks([self])

define_units

define_units(
    units: str | Sequence[str | None],
    reference_units: str
    | Sequence[str | None]
    | None = None,
) -> DataArray

Declare what the ordinate values are in, converting them to SI.

units is a single unit applied to every record, a sequence with one entry per record, or a {record index: unit} mapping to set only some. Entries of None leave a record's units undefined. Records that already have units are reinterpreted, not re-scaled twice.

Parameters:

Name	Type	Description	Default
units	str or sequence of str	The unit each record's values are in; one string applies to every record.	required
reference_units	str or sequence of str	The denominator unit, for records that have one.	None

Returns:

Type	Description
DataArray	Self, converted to SI in place.

Source code in src/visualdynamics/core/data.py

def define_units(self, units: str | Sequence[str | None],
                 reference_units: str | Sequence[str | None] | None
                 = None) -> DataArray:
    """Declare what the ordinate values are in, converting them to SI.

    `units` is a single unit applied to every record, a sequence with one
    entry per record, or a {record index: unit} mapping to set only some.
    Entries of None leave a record's units undefined. Records that already
    have units are reinterpreted, not re-scaled twice.

    Parameters
    ----------
    units : str or sequence of str
        The unit each record's values are in; one string applies to
        every record.
    reference_units : str or sequence of str, optional
        The denominator unit, for records that have one.

    Returns
    -------
    DataArray
        Self, converted to SI in place.
    """
    specs = self._expand(units, 'units')
    refs = self._expand(reference_units, 'reference_units')
    for i, unit in enumerate(specs):
        if unit is None:
            continue
        reference = refs[i]
        scale, offset, dim = self._conversion(unit, reference)
        old_scale, old_offset = self._current_transform(i)
        for values in self._value_arrays():
            raw = (values[i] - old_offset) / old_scale
            values[i] = raw * scale + offset
        self.ordinate_dim[i] = dim
        self.ordinate_unit[i] = unit
        self.reference_unit[i] = reference
    self._drop_stale_hints()
    return self

undefine_units

undefine_units(
    records: Sequence[int] | None = None,
) -> DataArray

Take a declaration back, restoring the file's raw values.

The inverse of define_units. A wrong guess should be correctable without reimporting, and that means being able to withdraw one, not only to replace it — there is no unit string meaning 'I no longer know'.

Parameters:

Name	Type	Description	Default
records	sequence of int	Which records to revert. All of them when omitted.	None

Returns:

Type	Description
DataArray	Self, with the file's raw values restored.

Source code in src/visualdynamics/core/data.py

def undefine_units(self, records: Sequence[int] | None = None) -> DataArray:
    """Take a declaration back, restoring the file's raw values.

    The inverse of `define_units`. A wrong guess should be correctable
    without reimporting, and that means being able to withdraw one, not
    only to replace it — there is no unit string meaning 'I no longer
    know'.

    Parameters
    ----------
    records : sequence of int, optional
        Which records to revert. All of them when omitted.

    Returns
    -------
    DataArray
        Self, with the file's raw values restored.
    """
    for i in (range(self.num_records) if records is None
              else [int(r) for r in records]):
        if self.ordinate_unit[i] is None:
            continue
        old_scale, old_offset = self._current_transform(i)
        for values in self._value_arrays():
            values[i] = (values[i] - old_offset) / old_scale
        self.ordinate_dim[i] = UNKNOWN
        self.ordinate_unit[i] = None
        self.reference_unit[i] = None
    return self

column_keys

column_keys() -> list[str] | None

What tells one record from another besides its response.

The reference DOF when records are a matrix of measurements, the block when they are the same measurement repeated, None when the response alone is the whole identity. This is what decides whether an object expands into a grid.

Source code in src/visualdynamics/core/data.py

def column_keys(self) -> list[str] | None:
    """What tells one record from another besides its response.

    The reference DOF when records are a matrix of measurements, the
    block when they are the same measurement repeated, None when the
    response alone is the whole identity. This is what decides whether
    an object expands into a grid.
    """
    return self.reference_dof if self.reference_dof is not None else self.block

record_label

record_label(i: int) -> str

A short label for one record, for a legend or an axis.

Parameters:

Name	Type	Description	Default
i	int	Which record.	required

Returns:

Type	Description
str	The record's DOF, plus whatever tells it from its neighbours.

Source code in src/visualdynamics/core/data.py

def record_label(self, i: int) -> str:
    """A short label for one record, for a legend or an axis.

    Parameters
    ----------
    i : int
        Which record.

    Returns
    -------
    str
        The record's DOF, plus whatever tells it from its neighbours.
    """
    if self.reference_dof is not None:
        return f'{self.response_dof[i]}/{self.reference_dof[i]}'
    if self.block is not None:
        # a space, not a slash: this is not a ratio of two DOFs —
        # stripped, because either half may be empty: an exodus
        # global variable is all block and no DOF, and ' KE' with a
        # ghost space is not a label anyone typed
        return f'{self.response_dof[i]} {self.block[i]}'.strip()
    return self.response_dof[i]

record_pair

record_pair(i: int) -> tuple[str, str]

The (response, reference) record i is between.

A record with no reference is an autospectrum — a channel against itself — so it pairs with the diagonal, which is what a specification bounds. The one reading of that rule: the plot, the table beside it and the report all ask here, so their labels cannot disagree.

Parameters:

Name	Type	Description	Default
i	int	The record.	required

Returns:

Type	Description
tuple of str	Response DOF, reference DOF.

Source code in src/visualdynamics/core/data.py

def record_pair(self, i: int) -> tuple[str, str]:
    """The (response, reference) record `i` is between.

    A record with no reference is an autospectrum — a channel
    against itself — so it pairs with the diagonal, which is what
    a specification bounds. The one reading of that rule: the
    plot, the table beside it and the report all ask here, so
    their labels cannot disagree.

    Parameters
    ----------
    i : int
        The record.

    Returns
    -------
    tuple of str
        Response DOF, reference DOF.
    """
    response = self.response_dof[i]
    reference = (self.reference_dof[i]
                 if self.reference_dof is not None else response)
    return response, reference

log_scaled

log_scaled() -> bool

Whether this object's magnitude reads on a log axis.

Logarithmic for frequency-domain data unless the class pins it (log_ordinate — a coherence is a 0..1 ratio and says nothing on a log axis). The object answers so the 2-D plot and the 3-D waterfall read one rule and cannot disagree about its axis.

Source code in src/visualdynamics/core/data.py

def log_scaled(self) -> bool:
    """Whether this object's magnitude reads on a log axis.

    Logarithmic for frequency-domain data unless the class pins it
    (`log_ordinate` — a coherence is a 0..1 ratio and says nothing on
    a log axis). The object answers so the 2-D plot and the 3-D
    waterfall read one rule and cannot disagree about its axis.
    """
    return (self.abscissa_dim == 'frequency' if self.log_ordinate is None
            else bool(self.log_ordinate))

display_abscissa

display_abscissa(unit_system: UnitSystem) -> ndarray

The abscissa converted into a unit system's own units.

Parameters:

Name	Type	Description	Default
unit_system	UnitSystem	The units to present in.	required

Returns:

Type	Description
ndarray	The abscissa in display units.

Source code in src/visualdynamics/core/data.py

def display_abscissa(self, unit_system: UnitSystem) -> np.ndarray:
    """The abscissa converted into a unit system's own units.

    Parameters
    ----------
    unit_system : UnitSystem
        The units to present in.

    Returns
    -------
    numpy.ndarray
        The abscissa in display units.
    """
    return unit_system.from_si(self.abscissa, self.abscissa_dim)

display_ordinate

display_ordinate(
    unit_system: UnitSystem,
    records: Iterable[int] | None = None,
) -> ndarray

Ordinate in display units; undefined records pass through as-is.

records limits the work to the ones asked for. A 1356-record FRF has one or two distinct dimensions in it, so the conversion is gathered per dimension and applied to a whole block at once — converting row by row meant a unit lookup per record, which is how drawing a single curve came to cost 2713 trips through pint.

Parameters:

Name	Type	Description	Default
unit_system	UnitSystem	The units to present the values in.	required
records	iterable of int	Which records to convert. All of them when omitted.	None

Returns:

Type	Description
ndarray	The values in display units. Records with undefined units pass through untouched.

Source code in src/visualdynamics/core/data.py

def display_ordinate(self, unit_system: UnitSystem,
                     records: Iterable[int] | None = None
                     ) -> np.ndarray:
    """Ordinate in display units; undefined records pass through as-is.

    `records` limits the work to the ones asked for. A 1356-record FRF
    has one or two distinct dimensions in it, so the conversion is
    gathered per dimension and applied to a whole block at once —
    converting row by row meant a unit lookup per record, which is how
    drawing a single curve came to cost 2713 trips through pint.

    Parameters
    ----------
    unit_system : UnitSystem
        The units to present the values in.
    records : iterable of int, optional
        Which records to convert. All of them when omitted.

    Returns
    -------
    numpy.ndarray
        The values in display units. Records with undefined
        units pass through untouched.
    """
    rows = range(self.num_records) if records is None else list(records)
    out = np.empty((len(rows), self.ordinate.shape[1]),
                   dtype=self.ordinate.dtype)
    by_dimension = {}
    for position, record in enumerate(rows):
        by_dimension.setdefault(self.ordinate_dim[record], []).append(
            (position, record))
    for dim, pairs in by_dimension.items():
        positions = [position for position, _ in pairs]
        source = self.ordinate[[record for _, record in pairs]]
        out[positions] = (source if dim == UNKNOWN
                          else unit_system.from_si(source, dim))
    return out

save

save(path: str | PathLike) -> None

Write this object to a .vdyn file of its own.

Parameters:

Name	Type	Description	Default
path	str or PathLike	Where to write it.	required

Returns:

Type	Description
None

Source code in src/visualdynamics/core/data.py

def save(self, path: str | os.PathLike) -> None:
    """Write this object to a `.vdyn` file of its own.

    Parameters
    ----------
    path : str or os.PathLike
        Where to write it.

    Returns
    -------
    None
    """
    from ..io import native
    native.save(self, path)

plot

plot(
    unit_system: UnitSystem | None = None, **kwargs: Any
) -> Any

Draw every record on one set of axes.

Parameters:

Name	Type	Description	Default
unit_system	UnitSystem	Units to draw in.	None
**kwargs	Any	Passed through to the plotting layer.	{}

Returns:

Type	Description
object	The plot widget or plotter.

Source code in src/visualdynamics/core/data.py

def plot(self, unit_system: UnitSystem | None = None,
         **kwargs: Any) -> Any:
    """Draw every record on one set of axes.

    Parameters
    ----------
    unit_system : UnitSystem, optional
        Units to draw in.
    **kwargs
        Passed through to the plotting layer.

    Returns
    -------
    object
        The plot widget or plotter.
    """
    from ..plot import plot_data
    return plot_data(self, unit_system=unit_system, **kwargs)

save_plot

save_plot(
    path: str | PathLike,
    unit_system: UnitSystem | None = None,
    **kwargs: Any,
) -> Any

Draw the records and write the figure to path.

Parameters:

Name	Type	Description	Default
path	str or PathLike	Where to write the image.	required
unit_system	UnitSystem	Units to draw in.	None
**kwargs	Any	Passed through to the plotting layer.	{}

Returns:

Type	Description
object	The plot widget or plotter.

Source code in src/visualdynamics/core/data.py

def save_plot(self, path: str | os.PathLike,
              unit_system: UnitSystem | None = None,
              **kwargs: Any) -> Any:
    """Draw the records and write the figure to `path`.

    Parameters
    ----------
    path : str or os.PathLike
        Where to write the image.
    unit_system : UnitSystem, optional
        Units to draw in.
    **kwargs
        Passed through to the plotting layer.

    Returns
    -------
    object
        The plot widget or plotter.
    """
    from ..plot import save_plot
    return save_plot(self, path, unit_system=unit_system, **kwargs)

plot_waterfall

plot_waterfall(
    records: Sequence[int] | None = None, **kwargs: Any
) -> Any

The records spread along a depth axis, coloured by level — the plot bar's 3-D reading, scripted. screenshot= renders headless to a file; without it a window of the app's own 3-D pane opens.

Parameters:

Name	Type	Description	Default
records	sequence of int	Which records to stage. All of them when omitted.	None
**kwargs	Any	Passed through to the scene.	{}

Returns:

Type	Description
object	The plot widget or plotter.

Source code in src/visualdynamics/core/data.py

def plot_waterfall(self, records: Sequence[int] | None = None,
                   **kwargs: Any) -> Any:
    """The records spread along a depth axis, coloured by level —
    the plot bar's 3-D reading, scripted. `screenshot=` renders
    headless to a file; without it a window of the app's own 3-D
    pane opens.

    Parameters
    ----------
    records : sequence of int, optional
        Which records to stage. All of them when omitted.
    **kwargs
        Passed through to the scene.

    Returns
    -------
    object
        The plot widget or plotter.
    """
    from ..viz.waterfall import plot_waterfall
    return plot_waterfall(self, records, **kwargs)

TimeHistory

TimeHistory(
    abscissa: ArrayLike,
    ordinate: ArrayLike,
    response_dof: str | Sequence[str],
    reference_dof: str | Sequence[str] | None = None,
    ordinate_dim: str | Sequence[str] | None = None,
    comment: str | Sequence[str] | None = None,
    ordinate_unit: str | Sequence[str | None] | None = None,
    reference_unit: str
    | Sequence[str | None]
    | None = None,
    dimension_hint: str
    | Sequence[str | None]
    | None = None,
    block: str | Sequence[str] | None = None,
)

Bases: DataArray

A measurement against time: the record as it was acquired.

Everything else in a random or shock test is derived from one of these, and the deriving is here — spectra, PSDs, the full CPSD matrix, multiple coherence, shock response spectra. Two things ride along that say how to read it: averaging, the frames a spectrum is averaged over, and shocks, the events an SRS is computed from. Both are the app's two views of a trace, and both are stored on the history rather than passed at the call, so a PSD and the coherence beside it cannot describe different measurements.

Methods:

Name	Description
compute_spectra	The averaged spectrum of every channel — sdynpy's convention.
channel_key	What makes record i the same channel as another.
suggest_averaging	Averaging parameters worked out from the record itself.
capture_indices	Which playing each record is: 0 for a channel's first
suggest_truncation	The whole record — the only neutral span.
truncate	This record cut to a span (core.truncate.truncate), using
suggest_filtering	A starting low-pass: a tenth of the sample rate, order 4.
filter	This record through its filter (core.filters.filtered) —
integrate	One integration — acceleration to velocity, velocity to
differentiate	One differentiation — displacement to velocity, velocity to
srs_windows	The stretches an SRS of this record would read, settled.
srs_band	(low, high) in Hz: the band these windows can support.
compute_srs	A shock response spectrum for every channel of every shock.
compute_psds	One-sided auto-power spectral density per channel, averaged
psd_type	What a PSD of this history is.
srs_type	What an SRS of this history is — see psd_type.
compute_cpsds	The full cross-spectral density matrix, averaged across the
drive_dofs	The DOFs this history looks like it was driven at.
compute_frfs	The frequency response functions, one per response/drive pair.
compute_multiple_coherence	How much of each response the drives together account for.
to_sep005	This record as SEP 005 timeseries — the sdypy ecosystem's

Attributes:

Name	Type	Description
sample_rate	float	Samples per second, from the abscissa — which must be even.
records_per_channel	dict[tuple[str, str, str | None], int]	{channel key: how many records carry it}.
split_into_frames	bool	Whether the records are already the averages.
average_counts	tuple[int, int]	(fewest, most) frames any one channel will be averaged over.

Source code in src/visualdynamics/core/data.py

def __init__(self, abscissa: ArrayLike, ordinate: ArrayLike,
             response_dof: str | Sequence[str],
             reference_dof: str | Sequence[str] | None = None,
             ordinate_dim: str | Sequence[str] | None = None,
             comment: str | Sequence[str] | None = None,
             ordinate_unit: str | Sequence[str | None] | None = None,
             reference_unit: str | Sequence[str | None] | None = None,
             dimension_hint: str | Sequence[str | None] | None = None,
             block: str | Sequence[str] | None = None) -> None:
    self.abscissa: np.ndarray = np.asarray(abscissa, dtype=np.float64)
    ordinate = np.atleast_2d(np.asarray(
        ordinate, dtype=np.complex128 if self.complex_ordinate else np.float64))
    self.ordinate: np.ndarray = ordinate
    n = ordinate.shape[0]
    if self.abscissa.ndim != 1 or ordinate.shape[1] != len(self.abscissa):
        raise ValueError(
            f'ordinate shape {ordinate.shape} does not match '
            f'abscissa length {len(self.abscissa)}')
    # a channel whose node or direction was never recorded imports
    # as a DOF of '' — see `validate.dofs`
    self.response_dof: list[str] = _dofs(
        self._str_list(response_dof, n, 'response_dof'), 'response DOF',
        allow_unknown=True)
    self.reference_dof: list[str] | None
    if reference_dof is None:
        if self.needs_reference:
            raise ValueError(f'{type(self).__name__} requires reference_dof')
        self.reference_dof = None
    else:
        self.reference_dof = _dofs(
            self._str_list(reference_dof, n, 'reference_dof'),
            'reference DOF', allow_unknown=True)
    # A record is its response plus, sometimes, one more thing. For a
    # measurement against something else that is the reference DOF; for
    # the same measurement repeated it is which repeat — an average, a
    # run, a temperature. `block` holds the short label of that second
    # kind, and nothing about it is time-specific.
    self.block: list[str] | None = (None if block is None
                                    else self._str_list(block, n, 'block'))
    self.ordinate_dim: list[str] = self._str_list(
        UNKNOWN if ordinate_dim is None else ordinate_dim, n, 'ordinate_dim')
    for dim in set(self.ordinate_dim):
        parse_dimension(dim)  # validates
    self.ordinate_unit: list[str | None] = self._opt_list(
        ordinate_unit, n, 'ordinate_unit')
    self.reference_unit: list[str | None] = self._opt_list(
        reference_unit, n, 'reference_unit')
    self.comment: list[str] = self._str_list(
        comment if comment is not None else '', n, 'comment')
    self.dimension_hint: list[str | None] = self._opt_list(
        dimension_hint, n, 'dimension_hint')
    for hint in set(self.dimension_hint):
        if hint is not None:
            parse_dimension(hint)  # validates
    self._fill_si_units()
    self._drop_stale_hints()

Attributes

sample_rate property

sample_rate: float

Samples per second, from the abscissa — which must be even.

records_per_channel property

records_per_channel: dict[tuple[str, str, str | None], int]

{channel key: how many records carry it}.

Usually one. A capture a controller saved frame by frame holds one record per average.

split_into_frames property

split_into_frames: bool

Whether the records are already the averages.

A controller that saves its spectral captures writes each frame as its own record. There is then nothing to slice and nothing to overlap: the frame length and the count are settled by the file, and the only parameter left to choose is the window.

average_counts property

average_counts: tuple[int, int]

(fewest, most) frames any one channel will be averaged over.

The two agree for anything a controller wrote, which saves every channel the same number of times. They part only for a history assembled by hand out of unequal captures, and then the table has to say so rather than quote a number that is true of some channels and not others.

Methods:

compute_spectra

compute_spectra() -> Spectrum

The averaged spectrum of every channel — sdynpy's convention.

A channel's frames — its records across the averages — are each FFT'd single-sided with no amplitude scaling (numpy's rfft, norm='backward': a sine of amplitude A on a bin reads AN/2), rectangular window, and averaged as the complex mean*, exactly what sdynpy's TimeHistoryArray.fft does with frames. Phase is preserved; content whose phase is random frame to frame — a burst random excitation's response — averages toward zero, which is that convention's documented behavior. Returns a Spectrum with one record per channel, carrying the channel's own quantity and units.

Source code in src/visualdynamics/core/data.py

def compute_spectra(self) -> Spectrum:
    """The averaged spectrum of every channel — sdynpy's convention.

    A channel's frames — its records across the averages — are each
    FFT'd single-sided with no amplitude scaling (numpy's rfft,
    norm='backward': a sine of amplitude A on a bin reads A*N/2),
    rectangular window, and averaged as the *complex mean*, exactly
    what sdynpy's `TimeHistoryArray.fft` does with frames. Phase is
    preserved; content whose phase is random frame to frame — a
    burst random excitation's response — averages toward zero,
    which is that convention's documented behavior. Returns a
    Spectrum with one record per channel, carrying the channel's
    own quantity and units.
    """
    if len(self.abscissa) < 2:
        raise ValueError('Spectra need at least two time samples')
    step = _even_steps(self.abscissa, 'Spectra')
    groups = {}
    for i, dof in enumerate(self.response_dof):
        key = (dof, self.ordinate_dim[i], self.ordinate_unit[i],
               self.dimension_hint[i])
        groups.setdefault(key, []).append(i)
    frequencies = np.fft.rfftfreq(len(self.abscissa), step)
    rows, dofs, dims, units, hints = [], [], [], [], []
    for (dof, dim, unit, hint), indices in groups.items():
        frames = np.fft.rfft(self.ordinate[indices], axis=1)
        rows.append(np.mean(frames, axis=0))
        dofs.append(dof)
        dims.append(dim)
        units.append(unit)
        hints.append(hint)
    return Spectrum(frequencies,
                    np.asarray(rows, dtype=np.complex128),
                    response_dof=dofs, ordinate_dim=dims,
                    ordinate_unit=units, dimension_hint=hints,
                    comment='averaged spectrum')

channel_key

channel_key(i: int) -> tuple[str, str, str | None]

What makes record i the same channel as another.

The DOF and what is measured there, never the DOF alone: a drive point carries a force record and an acceleration record at the same DOF, and those two are not each other's averages. This is the key the framing groups on, so counting channels and averaging them cannot disagree about what a channel is.

Parameters:

Name	Type	Description	Default
i	int	Which record.	required

Returns:

Type	Description
tuple of (str, str, str or None)	The DOF, quantity and unit that identify the channel — records sharing this key are the same channel.

Source code in src/visualdynamics/core/data.py

def channel_key(self, i: int) -> tuple[str, str, str | None]:
    """What makes record `i` the same channel as another.

    The DOF and what is measured there, never the DOF alone: a drive
    point carries a force record and an acceleration record at the
    same DOF, and those two are not each other's averages. This is
    the key the framing groups on, so counting channels and
    averaging them cannot disagree about what a channel is.

    Parameters
    ----------
    i : int
        Which record.

    Returns
    -------
    tuple of (str, str, str or None)
        The DOF, quantity and unit that identify the channel —
        records sharing this key are the same channel.
    """
    return (self.response_dof[i], self.ordinate_dim[i],
            self.dimension_hint[i])

suggest_averaging

suggest_averaging(**kwargs: Any) -> Averaging

Averaging parameters worked out from the record itself.

A run holds more than the test — the shaker coming up, a reduced-level check, whatever was still recording afterwards — and this finds the settled stretch worth averaging and as many frames as it will carry. See visualdynamics.core.detect.

Parameters:

Name	Type	Description	Default
**kwargs	Any	Overrides for individual parameters, such as window.	{}

Returns:

Type	Description
Averaging	Parameters worked out from the record itself.

Source code in src/visualdynamics/core/data.py

def suggest_averaging(self, **kwargs: Any) -> Averaging:
    """Averaging parameters worked out from the record itself.

    A run holds more than the test — the shaker coming up, a
    reduced-level check, whatever was still recording afterwards —
    and this finds the settled stretch worth averaging and as many
    frames as it will carry. See `visualdynamics.core.detect`.

    Parameters
    ----------
    **kwargs
        Overrides for individual parameters, such as `window`.

    Returns
    -------
    Averaging
        Parameters worked out from the record itself.
    """
    from .detect import suggest

    return suggest(self, **kwargs)

capture_indices

capture_indices() -> list[int]

Which playing each record is: 0 for a channel's first record, 1 for its second, and so on — in exactly the order _spectral_frame pools them, so the playing the averaging view pages to is one of the playings the average adds up. A channel is a channel_key group, the same grouping the pooling uses.

Source code in src/visualdynamics/core/data.py

def capture_indices(self) -> list[int]:
    """Which playing each record is: 0 for a channel's first
    record, 1 for its second, and so on — in exactly the order
    `_spectral_frame` pools them, so the playing the averaging
    view pages to is one of the playings the average adds up. A
    channel is a `channel_key` group, the same grouping the
    pooling uses.
    """
    seen: dict = {}
    out = []
    for i in range(len(self.response_dof)):
        key = self.channel_key(i)
        out.append(seen.get(key, 0))
        seen[key] = seen.get(key, 0) + 1
    return out

suggest_truncation

suggest_truncation() -> Any

The whole record — the only neutral span.

A starting point for the truncate view's handles, never a default the act adopts: keeping everything is not an act, so Truncate Data refuses until a real span is set.

Source code in src/visualdynamics/core/data.py

def suggest_truncation(self) -> Any:
    """The whole record — the only neutral span.

    A starting point for the truncate view's handles, never a
    default the act adopts: keeping everything is not an
    act, so Truncate Data refuses until a real span is set.
    """
    from .truncate import Truncation

    abscissa = np.asarray(self.abscissa, dtype=float)
    return Truncation(float(abscissa[0]), float(abscissa[-1]))

truncate

truncate(truncation: Any = None) -> TimeHistory

This record cut to a span (core.truncate.truncate), using the history's own truncation unless one is passed.

Parameters:

Name	Type	Description	Default
truncation	Truncation	The start and stop, in seconds on the record's own clock. Defaults to the record's own truncation.	None

Returns:

Type	Description
TimeHistory	The samples inside the span, every channel, the clock kept.

Source code in src/visualdynamics/core/data.py

def truncate(self, truncation: Any = None) -> TimeHistory:
    """This record cut to a span (`core.truncate.truncate`), using
    the history's own `truncation` unless one is passed.

    Parameters
    ----------
    truncation : Truncation, optional
        The start and stop, in seconds on the record's own clock.
        Defaults to the record's own `truncation`.

    Returns
    -------
    TimeHistory
        The samples inside the span, every channel, the clock
        kept.
    """
    from .truncate import truncate

    chosen = self.truncation if truncation is None else truncation
    if chosen is None:
        raise ValueError('no span to keep: set `truncation` first, '
                         'or pass one')
    return truncate(self, chosen)

suggest_filtering

suggest_filtering() -> Any

A starting low-pass: a tenth of the sample rate, order 4.

A low-pass rather than any other kind, because cutting noise above the content is the reach-for-first case; the filter view offers high- and band-pass beside it.

A judgement, not a detection — nothing in the record says where its content stops being signal. A tenth of the rate is where the integrate/differentiate round trip was measured at ~2% RMS (core.filters), and it sits below the mounted-resonance range a shock accelerometer pollutes. The filter view exists precisely so this number gets looked at rather than trusted.

Source code in src/visualdynamics/core/data.py

def suggest_filtering(self) -> Any:
    """A starting low-pass: a tenth of the sample rate, order 4.

    A low-pass rather than any other kind, because cutting noise
    above the content is the reach-for-first case; the filter view
    offers high- and band-pass beside it.

    A judgement, not a detection — nothing in the record says where
    its content stops being signal. A tenth of the rate is where
    the integrate/differentiate round trip was measured at ~2% RMS
    (core.filters), and it sits below the mounted-resonance range a
    shock accelerometer pollutes. The filter view exists precisely
    so this number gets looked at rather than trusted.
    """
    from .filters import Filtering

    return Filtering(high=self.sample_rate / 10.0)

filter

filter(filtering: Any = None) -> TimeHistory

This record through its filter (core.filters.filtered) — low-, high- or band-pass, whichever the filtering describes — using the history's own filtering unless one is passed.

Parameters:

Name	Type	Description	Default
filtering	Filtering	The pass-band edges and order. Defaults to the record's own filtering.	None

Returns:

Type	Description
TimeHistory	Every channel through the filter, zero phase.

Source code in src/visualdynamics/core/data.py

def filter(self, filtering: Any = None) -> TimeHistory:
    """This record through its filter (`core.filters.filtered`) —
    low-, high- or band-pass, whichever the filtering describes —
    using the history's own `filtering` unless one is passed.

    Parameters
    ----------
    filtering : Filtering, optional
        The pass-band edges and order. Defaults to the record's
        own `filtering`.

    Returns
    -------
    TimeHistory
        Every channel through the filter, zero phase.
    """
    from .filters import filtered

    chosen = self.filtering if filtering is None else filtering
    if chosen is None:
        raise ValueError('no filter to apply: set `filtering` first, '
                         'or pass one')
    return filtered(self, chosen)

integrate

integrate(drift_corner: Any = ...) -> TimeHistory

One integration — acceleration to velocity, velocity to displacement (core.filters.integrate). ... takes the default drift corner; None integrates raw, drift and all.

Parameters:

Name	Type	Description	Default
drift_corner	float or None	High-pass corner in Hz applied after integrating, so a sensor bias cannot become a ramp. ... takes the default; None integrates raw.	...

Returns:

Type	Description
TimeHistory	Acceleration becomes velocity, velocity becomes displacement. Other quantities are left out.

Source code in src/visualdynamics/core/data.py

def integrate(self, drift_corner: Any = ...) -> TimeHistory:
    """One integration — acceleration to velocity, velocity to
    displacement (`core.filters.integrate`). ``...`` takes the
    default drift corner; ``None`` integrates raw, drift and all.

    Parameters
    ----------
    drift_corner : float or None, optional
        High-pass corner in Hz applied after integrating, so a sensor
        bias cannot become a ramp. `...` takes the default; `None`
        integrates raw.

    Returns
    -------
    TimeHistory
        Acceleration becomes velocity, velocity becomes
        displacement. Other quantities are left out.
    """
    from .filters import DRIFT_CORNER, integrate

    return integrate(self, DRIFT_CORNER if drift_corner is ...
                     else drift_corner)

differentiate

differentiate() -> TimeHistory

One differentiation — displacement to velocity, velocity to acceleration (core.filters.differentiate).

Source code in src/visualdynamics/core/data.py

def differentiate(self) -> TimeHistory:
    """One differentiation — displacement to velocity, velocity to
    acceleration (`core.filters.differentiate`)."""
    from .filters import differentiate

    return differentiate(self)

srs_windows

srs_windows(
    shocks: Sequence[Any] | None = None,
) -> list[Any]

The stretches an SRS of this record would read, settled.

The fallback chain compute_srs has always used, extracted so the shock panel's derived rows and the spectrum itself cannot disagree about it (one implementation): the shocks the record carries; failing those, the averaging frames when the record is being read as frames; failing everything, the whole record as one window. Clipped to the record either way.

Parameters:

Name	Type	Description	Default
shocks	sequence of Shock	Override windows; the record's own when omitted.	None

Returns:

Type	Description
list of Shock	The settled windows, clipped to the record.

Source code in src/visualdynamics/core/data.py

def srs_windows(self, shocks: Sequence[Any] | None = None
                ) -> list[Any]:
    """The stretches an SRS of this record would read, settled.

    The fallback chain `compute_srs` has always used, extracted so
    the shock panel's derived rows and the spectrum itself cannot
    disagree about it (one implementation): the shocks the record
    carries; failing those, the averaging frames when the record
    is being read as frames; failing everything, the whole record
    as one window. Clipped to the record either way.

    Parameters
    ----------
    shocks : sequence of Shock, optional
        Override windows; the record's own when omitted.

    Returns
    -------
    list of Shock
        The settled windows, clipped to the record.
    """
    from .shocks import Shock

    rate = self.sample_rate
    samples = self.ordinate.shape[1]
    windows = self.shocks if shocks is None else shocks
    if not windows and self.averaging is not None:
        # the frames, when the record is being read as frames. A
        # transient run's playings are its averaging and nothing
        # else — the detector, asked at the same record, hunts for
        # events it was never meant to find here and answers with
        # its own idea of how many there are. Two readings of one
        # record is exactly what the two buttons refuse to show at
        # once, and the numbers should not disagree either.
        windows = tuple(Shock(start, stop) for start, stop
                        in self.averaging.frame_bounds(rate))
    if not windows:
        windows = (Shock(0.0, samples / rate),)
    return [w.clipped(samples, rate) for w in windows]

srs_band

srs_band(
    shocks: Sequence[Any] | None = None,
) -> tuple[float, float]

(low, high) in Hz: the band these windows can support.

From a frequency low enough that the shortest window still holds a cycle of it — a curve is one grid across every event, so the shortest is what the grid has to fit — up to a fifth of the sample rate, above which the ramp-invariant filter is being asked about frequencies the record cannot resolve. What compute_srs uses when no band is given, and what the shock panel states beside its settings.

Parameters:

Name	Type	Description	Default
shocks	sequence of Shock	Override windows; the record's own when omitted.	None

Returns:

Type	Description
tuple of float	(low, high) in Hz.

Source code in src/visualdynamics/core/data.py

def srs_band(self, shocks: Sequence[Any] | None = None
             ) -> tuple[float, float]:
    """(low, high) in Hz: the band these windows can support.

    From a frequency low enough that the *shortest* window still
    holds a cycle of it — a curve is one grid across every event,
    so the shortest is what the grid has to fit — up to a fifth
    of the sample rate, above which the ramp-invariant filter is
    being asked about frequencies the record cannot resolve.
    What `compute_srs` uses when no band is given, and what the
    shock panel states beside its settings.

    Parameters
    ----------
    shocks : sequence of Shock, optional
        Override windows; the record's own when omitted.

    Returns
    -------
    tuple of float
        (low, high) in Hz.
    """
    rate = self.sample_rate
    windows = self.srs_windows(shocks)
    shortest = min(w.samples(rate) for w in windows) / rate
    return max(1.0 / shortest, 1.0), rate / 5.0

compute_srs

compute_srs(
    shocks: Sequence[int] | None = None,
    low: float | None = None,
    high: float | None = None,
    per_octave: int | None = None,
    q: float | None = None,
    kind: str = "maximax",
) -> Srs

A shock response spectrum for every channel of every shock.

The parameters live on the history, the way averaging does: pass shocks to override, or leave it and the windows already on the object are used. With none anywhere, the whole record is one window, which is what a history holding a single trimmed transient is.

The windows are the shocks the record carries, or — when it is being read as frames rather than events — the frames. A specification carries neither and is one window: it is a single playing of a waveform, whole.

One curve per channel per event, never averaged across events. A shock test is judged on the worst shock, and the mean of four of them describes none of them. Which event a curve came from is in block, exactly as which average a frame came from is.

The band defaults to what the windows can support: from a frequency low enough that the shortest window still holds a cycle of it — a curve is one grid across every event, so the shortest is what the grid has to fit — up to a fifth of the sample rate, above which the ramp-invariant filter is being asked about frequencies the record cannot resolve.

Parameters:

Name	Type	Description	Default
shocks	sequence of int	Which shock windows to use. All of them when omitted.	None
low	float	The natural-frequency band, in Hz.	None
high	float	The natural-frequency band, in Hz.	None
per_octave	int	Frequency lines per octave.	None
q	float	The oscillator amplification.	None
kind	str	Which peak to keep: 'maximax' (largest magnitude of either sign), 'positive' or 'negative'.	'maximax'

Returns:

Type	Description
Srs	One curve per channel per shock.

Source code in src/visualdynamics/core/data.py

def compute_srs(self, shocks: Sequence[int] | None = None,
                low: float | None = None, high: float | None = None,
                per_octave: int | None = None, q: float | None = None,
                kind: str = 'maximax') -> Srs:
    """A shock response spectrum for every channel of every shock.

    The parameters live on the history, the way averaging does: pass
    `shocks` to override, or leave it and the windows already on the
    object are used. With none anywhere, the whole record is one
    window, which is what a history holding a single trimmed
    transient is.

    The windows are the shocks the record carries, or — when it is
    being read as frames rather than events — the frames. A
    specification carries neither and is one window: it is a single
    playing of a waveform, whole.

    One curve per channel *per event*, never averaged across events.
    A shock test is judged on the worst shock, and the mean of four
    of them describes none of them. Which event a curve came from is
    in `block`, exactly as which average a frame came from is.

    The band defaults to what the windows can support: from a
    frequency low enough that the *shortest* window still holds a
    cycle of it — a curve is one grid across every event, so the
    shortest is what the grid has to fit — up to a fifth of the
    sample rate, above which the ramp-invariant filter is being
    asked about frequencies the record cannot resolve.

    Parameters
    ----------
    shocks : sequence of int, optional
        Which shock windows to use. All of them when omitted.
    low, high : float, optional
        The natural-frequency band, in Hz.
    per_octave : int, optional
        Frequency lines per octave.
    q : float, optional
        The oscillator amplification.
    kind : str, default 'maximax'
        Which peak to keep: 'maximax' (largest magnitude of either
        sign), 'positive' or 'negative'.

    Returns
    -------
    Srs
        One curve per channel per shock.
    """
    from .srs import DEFAULT_Q, PER_OCTAVE, maximax, octave_frequencies
    from .srs import peaks as srs_peaks

    rate = self.sample_rate
    windows = self.srs_windows(shocks)
    named = len(windows) > 1
    band_low, band_high = self.srs_band(shocks)
    low = band_low if low is None else float(low)
    high = band_high if high is None else float(high)
    frequencies = octave_frequencies(
        low, high, PER_OCTAVE if per_octave is None else per_octave)
    q = DEFAULT_Q if q is None else float(q)

    rows, dofs, dims, hints, blocks = [], [], [], [], []
    for i, record in enumerate(self.ordinate):
        for k, window in enumerate(windows):
            first, last = window.bounds(rate)
            cut = record.real[first:last]
            if kind == 'maximax':
                rows.append(maximax(cut, frequencies, rate, q=q))
            else:
                highest, lowest = srs_peaks(cut, frequencies, rate, q=q)
                rows.append(highest if kind == 'positive' else -lowest)
            dofs.append(self.response_dof[i])
            dims.append(self.ordinate_dim[i])
            hints.append(self.dimension_hint[i])
            blocks.append(f'shock {k + 1}' if named
                          else (None if self.block is None
                                else self.block[i]))
    return self.srs_type()(
        frequencies, np.asarray(rows, dtype=np.float64),
        response_dof=dofs, ordinate_dim=dims,
        dimension_hint=hints,
        block=blocks if named or self.block is not None else None,
        q=q, kind=kind,
        comment=f'{kind} SRS at Q={q:g}')

compute_psds

compute_psds(averaging: Averaging | None = None) -> Psd

One-sided auto-power spectral density per channel, averaged across the frames.

Welch's method where each average is already its own frame: rectangular window, no overlap, Gxx = 2|X|²/(fs·N) with DC and Nyquist unhalved, the frames' powers averaged. Power is phase-insensitive, so burst random's random phase costs nothing here. Values land in (SI unit)²/Hz with the Psd dimension convention ('acceleration**2/frequency'); a channel with undefined units stays undefined, its hint squared along.

Parameters:

Name	Type	Description	Default
averaging	Averaging	How to cut the record into frames. Defaults to the record's own averaging, or one frame per record.	None

Returns:

Type	Description
Psd	One auto-power spectral density per channel.

Source code in src/visualdynamics/core/data.py

def compute_psds(self, averaging: Averaging | None = None) -> Psd:
    """One-sided auto-power spectral density per channel, averaged
    across the frames.

    Welch's method where each average is already its own frame:
    rectangular window, no overlap, Gxx = 2|X|²/(fs·N) with DC and
    Nyquist unhalved, the frames' powers averaged. Power is
    phase-insensitive, so burst random's random phase costs
    nothing here. Values land in (SI unit)²/Hz with the Psd
    dimension convention ('acceleration**2/frequency'); a channel
    with undefined units stays undefined, its hint squared along.

    Parameters
    ----------
    averaging : Averaging, optional
        How to cut the record into frames. Defaults to the record's
        own `averaging`, or one frame per record.

    Returns
    -------
    Psd
        One auto-power spectral density per channel.
    """
    frequencies, scale, groups = self._spectral_frame(averaging)
    rows, dofs, dims, hints = [], [], [], []
    for (dof, dim, hint), windowed in groups.items():
        frames = np.fft.rfft(windowed, axis=1)
        rows.append(np.mean(np.abs(frames) ** 2, axis=0) * scale)
        dofs.append(dof)
        dims.append(f'{dim}**2/frequency' if dim != UNKNOWN
                    else UNKNOWN)
        hints.append(f'{hint}**2/frequency' if hint else None)
    out = self.psd_type()(
        frequencies, np.asarray(rows, dtype=np.float64),
        response_dof=dofs, ordinate_dim=dims,
        dimension_hint=hints, comment='averaged PSD')
    # computed from a record, so a density per bin whatever class
    # it came back as — the PSD of a target is a Specification, and
    # a Specification is otherwise a power law through breakpoints
    out.interpolation = 'bin'
    return out

psd_type

psd_type() -> type[Psd]

What a PSD of this history is.

A plain record's spectra are plain spectra. A target's are still a target: the PSD of a waveform the article was required to see is the spectrum it was required to see, and losing that on the way through an FFT would leave two objects of the same class with nothing but a name to say which was the requirement. Overridden in TransientSpecification rather than decided by the caller, so a script and the app cannot disagree.

Source code in src/visualdynamics/core/data.py

def psd_type(self) -> type[Psd]:
    """What a PSD of this history is.

    A plain record's spectra are plain spectra. A **target's** are
    still a target: the PSD of a waveform the article was required
    to see is the spectrum it was required to see, and losing that
    on the way through an FFT would leave two objects of the same
    class with nothing but a name to say which was the requirement.
    Overridden in `TransientSpecification` rather than decided by
    the caller, so a script and the app cannot disagree.
    """
    return Psd

srs_type

srs_type() -> type[Srs]

What an SRS of this history is — see psd_type.

Source code in src/visualdynamics/core/data.py

def srs_type(self) -> type[Srs]:
    """What an SRS of this history is — see `psd_type`."""
    return Srs

compute_cpsds

compute_cpsds(averaging: Averaging | None = None) -> Psd

The full cross-spectral density matrix, averaged across the frames — every channel against every channel, not just each against itself.

Gxy = 2·conj(X)·Y/(fs·N) with DC and Nyquist unhalved, which is compute_psds on the diagonal, where conj(X)·X is |X|². The cross terms are how two channels move together, which is most of what a CPSD is for, and they are what a PSD throws away.

Laid out as the importer lays an imported matrix out: one record per (response, reference) pair, row by row, so an n-channel history gives n² records that read as a grid. A cross term's dimension is the product of the two, a*b/frequency, against a**2/frequency down the diagonal.

Parameters:

Name	Type	Description	Default
averaging	Averaging	How to cut the record into frames. Defaults to the record's own averaging, or one frame per record.	None

Returns:

Type	Description
Psd	The full cross-spectral matrix, every channel against every channel.

Source code in src/visualdynamics/core/data.py

def compute_cpsds(self, averaging: Averaging | None = None) -> Psd:
    """The full cross-spectral density matrix, averaged across the
    frames — every channel against every channel, not just each
    against itself.

    Gxy = 2·conj(X)·Y/(fs·N) with DC and Nyquist unhalved, which is
    `compute_psds` on the diagonal, where conj(X)·X is |X|². The
    cross terms are how two channels move together, which is most of
    what a CPSD is for, and they are what a PSD throws away.

    Laid out as the importer lays an imported matrix out: one record
    per (response, reference) pair, row by row, so an n-channel
    history gives n² records that read as a grid. A cross term's
    dimension is the product of the two, `a*b/frequency`, against
    `a**2/frequency` down the diagonal.

    Parameters
    ----------
    averaging : Averaging, optional
        How to cut the record into frames. Defaults to the record's
        own `averaging`, or one frame per record.

    Returns
    -------
    Psd
        The full cross-spectral matrix, every channel against
        every channel.
    """
    frequencies, scale, groups = self._spectral_frame(averaging)
    keys = list(groups)
    frames = {key: np.fft.rfft(windowed, axis=1)
              for key, windowed in groups.items()}
    records, responses, references = [], [], []
    dims, hints = [], []
    for row, (dof_i, dim_i, hint_i) in enumerate(keys):
        for column, (dof_j, dim_j, hint_j) in enumerate(keys):
            cross = np.mean(np.conj(frames[keys[row]])
                            * frames[keys[column]], axis=0) * scale
            records.append(cross)
            responses.append(dof_i)
            references.append(dof_j)
            known = UNKNOWN not in (dim_i, dim_j)
            dims.append(
                (f'{dim_i}**2/frequency' if dim_i == dim_j
                 else f'{dim_i}*{dim_j}/frequency') if known
                else UNKNOWN)
            hints.append(
                (f'{hint_i}**2/frequency' if hint_i == hint_j
                 else f'{hint_i}*{hint_j}/frequency')
                if hint_i and hint_j else None)
    return Psd(frequencies, np.asarray(records, dtype=np.complex128),
               response_dof=responses, reference_dof=references,
               ordinate_dim=dims, dimension_hint=hints,
               comment='averaged CPSD')

drive_dofs

drive_dofs() -> list[str]

The DOFs this history looks like it was driven at.

A guess from the quantities alone, and it is only ever a default. What actually makes a channel a drive is that the controller had a feedback device on it, which the channel table records and a time history does not.

Source code in src/visualdynamics/core/data.py

def drive_dofs(self) -> list[str]:
    """The DOFs this history looks like it was driven at.

    A guess from the quantities alone, and it is only ever a
    default. What actually makes a channel a drive is that the
    controller had a feedback device on it, which the channel table
    records and a time history does not.
    """
    return list(dict.fromkeys(
        dof for dof, dim in zip(self.response_dof, self.ordinate_dim)
        if dim in self.EXCITATION_DIMS))

compute_frfs

compute_frfs(
    references: Sequence[str] | None = None,
    averaging: Averaging | None = None,
    method: str = "Hv",
) -> Frf

The frequency response functions, one per response/drive pair.

The three estimators differ in one assumption — where the noise is — and agree wherever there is little of it. They part company exactly where a measurement is worst, which is why the choice matters and why it is a choice.

H1 assumes the noise is on the response. It biases low at resonance, where the response is large and the force small, and it is what a controller computes and a modal fit expects.

Gfx = Gff H,   so   H = Gff^-1 Gfx

With one reference that is the textbook Gfx / Gff. With several it is the MIMO estimate, and the matrix inverse is the whole point: two shakers driving one article are correlated, and dividing each response by each drive separately would credit both with the same motion.

H2 assumes the noise is on the reference, and biases high at anti-resonance for the mirror-image reason. With one reference it is Gxx / Gxf, one response at a time. With several references it needs as many equations as unknowns, and there are exactly enough only when the system is square — as many responses as references — where it becomes the classical coupled form Gxx * Gfx^-1 (Rocklin, Crowley and Vold, 1985), computed here exactly as sdynpy computes it (matched by decision, Brandon 2026-08-28, and pinned against its numbers). The coupling is worth knowing about: every response feeds one matrix inverse, so a channel's H2 depends on which other channels are in the set — measured at 0.2% on the oracle signals, growing with noise — where H1 and Hv rows never do. A non-square multi-reference set is refused; Hv answers the same noise-on-both question per response, uncoupled.

Hv (the default) assumes noise on both and asks for neither: it is the total-least-squares fit, the null direction of

[[Gff, Gfx], [Gxf, Gxx]]

taken as the eigenvector of its smallest eigenvalue, per response and per line. It falls between H1 and H2 — strictly between, wherever the coherence is under one — and needs no claim about which instrument is the better one. It is the default because that claim is the one a test least often gets to make honestly: an accelerometer out on a structure and a force cell in the load path are both imperfect, in different places.

Note what a total-least-squares fit means with units in play: it weighs a unit of error on the force against a unit of error on the acceleration, and those are not the same thing. That is baked into the estimator and is the received formulation; it is why Hv is a middle reading and not a better one.

A pseudo-inverse rather than a solve for H1, because two shakers can be very nearly the same drive and Gff is then close to singular — where a solve raises or returns nonsense, a pseudo-inverse gives the least-squares answer the estimate is asking for anyway.

Built on the same frames compute_psds averages and the same references compute_multiple_coherence uses, so the coherence beside an FRF is that FRF's coherence.

Parameters:

Name	Type	Description	Default
references	sequence of str	The drive DOFs. Detected from the record when omitted.	None
averaging	Averaging	How to cut the record into frames. Defaults to the record's own averaging, or one frame per record.	None
method	str	The estimator: 'Hv', 'H1' or 'H2'.	'Hv'

Returns:

Type	Description
Frf	One record per response and drive pair.

Source code in src/visualdynamics/core/data.py

def compute_frfs(self, references: Sequence[str] | None = None,
                 averaging: Averaging | None = None,
                 method: str = 'Hv') -> Frf:
    """The frequency response functions, one per response/drive pair.

    The three estimators differ in one assumption — where the noise
    is — and agree wherever there is little of it. They part company
    exactly where a measurement is worst, which is why the choice
    matters and why it is a choice.

    **H1** assumes the noise is on the *response*. It biases low at
    resonance, where the response is large and the force small, and
    it is what a controller computes and a modal fit expects.

        Gfx = Gff H,   so   H = Gff^-1 Gfx

    With one reference that is the textbook `Gfx / Gff`. With
    several it is the MIMO estimate, and the matrix inverse is the
    whole point: two shakers driving one article are correlated, and
    dividing each response by each drive separately would credit
    both with the same motion.

    **H2** assumes the noise is on the *reference*, and biases high
    at anti-resonance for the mirror-image reason. With one
    reference it is `Gxx / Gxf`, one response at a time. With
    several references it needs as many equations as unknowns, and
    there are exactly enough only when the system is *square* — as
    many responses as references — where it becomes the classical
    coupled form `Gxx * Gfx^-1` (Rocklin, Crowley and Vold, 1985),
    computed here exactly as sdynpy computes it (matched by
    decision, Brandon 2026-08-28, and pinned against its numbers).
    The coupling is worth knowing about: every response feeds one
    matrix inverse, so a channel's H2 depends on which other
    channels are in the set — measured at 0.2% on the oracle
    signals, growing with noise — where H1 and Hv rows never do. A
    non-square multi-reference set is refused; Hv answers the same
    noise-on-both question per response, uncoupled.

    **Hv** (the default) assumes noise on both and asks for neither:
    it is the total-least-squares fit, the null direction of

        [[Gff, Gfx], [Gxf, Gxx]]

    taken as the eigenvector of its smallest eigenvalue, per response
    and per line. It falls between H1 and H2 — strictly between,
    wherever the coherence is under one — and needs no claim about
    which instrument is the better one. It is the default because
    that claim is the one a test least often gets to make honestly:
    an accelerometer out on a structure and a force cell in the load
    path are both imperfect, in different places.

    Note what a total-least-squares fit means with units in play: it
    weighs a unit of error on the force against a unit of error on
    the acceleration, and those are not the same thing. That is
    baked into the estimator and is the received formulation; it is
    why Hv is a middle reading and not a better one.

    A pseudo-inverse rather than a solve for H1, because two shakers
    can be very nearly the same drive and `Gff` is then close to
    singular — where a solve raises or returns nonsense, a
    pseudo-inverse gives the least-squares answer the estimate is
    asking for anyway.

    Built on the same frames `compute_psds` averages and the same
    references `compute_multiple_coherence` uses, so the coherence
    beside an FRF is that FRF's coherence.

    Parameters
    ----------
    references : sequence of str, optional
        The drive DOFs. Detected from the record when omitted.
    averaging : Averaging, optional
        How to cut the record into frames. Defaults to the record's
        own `averaging`, or one frame per record.
    method : str, default 'Hv'
        The estimator: 'Hv', 'H1' or 'H2'.

    Returns
    -------
    Frf
        One record per response and drive pair.
    """
    if method not in self.FRF_METHODS:
        raise ValueError(
            f'{method!r} is not an FRF estimator: '
            + ', '.join(self.FRF_METHODS))
    frequencies, keys, drives, responses, cross, averages = \
        self._cross_spectral_frame('FRFs', references, averaging)
    # Multi-reference H2 exists only where the system is square —
    # as many responses as references — and there it is the
    # classical coupled form, Gxx * Gfx^-1 (Rocklin, Crowley and
    # Vold, 1985), adopted to match sdynpy (Brandon, 2026-08-28;
    # this file refused it for a day first, and the reasoning both
    # ways is worth keeping). The coupling is real: every response
    # feeds the one inverse, so a channel's H2 depends on which
    # other channels are in the set — measured at 0.2% on the
    # oracle signals with 5% noise, growing with noise, where H1
    # is bit-identical under the same swap. That property argued
    # for refusing; matching the tool the audience already trusts
    # argued for computing; Brandon chose compatibility, and the
    # property is documented instead of avoided. Non-square
    # multi-reference stays refused — there the equations genuinely
    # cannot balance — and Hv answers per response, uncoupled.
    coupled = None
    if method == 'H2' and len(drives) > 1:
        if len(responses) != len(drives):
            raise ValueError(
                f'H2 with {len(drives)} references needs exactly '
                f'{len(drives)} responses (the coupled square form) '
                f'and this history has {len(responses)}. '
                f'Hv is the estimator that answers the same question '
                f'for any shape.')
        cxx = np.stack([np.stack([cross(i, j) for j in responses],
                                 axis=-1)
                        for i in responses], axis=-2)
        cfx = np.stack([np.stack([cross(i, j) for j in responses],
                                 axis=-1)
                        for i in drives], axis=-2)
        # H solves H * Gfx = Gxx, taken as solve(Gfx^T, Gxx^T)^T per
        # line. The outer conj translates conventions: this file's
        # cross() is Bendat & Piersol's conj(X)*Y, the classical
        # form is written in X*conj(Y), and the two matrices are
        # elementwise conjugates — verified against sdynpy's own
        # square H2 at machine precision rather than trusted.
        coupled = np.conj(np.swapaxes(
            np.linalg.solve(np.swapaxes(cfx, -2, -1),
                            np.swapaxes(cxx, -2, -1)), -2, -1))

    estimator = getattr(self, f'_frf_{method.lower()}')
    gff = np.stack([np.stack([cross(i, j) for j in drives], axis=-1)
                    for i in drives], axis=-2)

    rows, response_dof, reference_dof, dims, hints = [], [], [], [], []
    for position, x in enumerate(responses):
        if coupled is not None:
            estimate = coupled[:, position, :]
        else:
            gfx = np.stack([cross(i, x) for i in drives], axis=-1)
            estimate = estimator(gff, gfx, np.real(cross(x, x)))
        # response-major, so a pair reads the way it is named:
        # every drive for one response, then the next response
        for column, i in enumerate(drives):
            rows.append(estimate[:, column])
            response_dof.append(keys[x][0])
            reference_dof.append(keys[i][0])
            dims.append(self._ratio(keys[x][1], keys[i][1]))
            hints.append(self._ratio(keys[x][2], keys[i][2])
                         if keys[x][2] and keys[i][2] else None)
    return Frf(frequencies, np.asarray(rows), response_dof=response_dof,
               reference_dof=reference_dof, ordinate_dim=dims,
               dimension_hint=hints,
               comment=f'{method} from {len(drives)} '
                       f'reference{"s" * (len(drives) != 1)}, '
                       f'{averages} averages')

compute_multiple_coherence

compute_multiple_coherence(
    references: Sequence[str] | None = None,
    averaging: Averaging | None = None,
) -> MultipleCoherence

How much of each response the drives together account for.

Ordinary coherence asks what one reference explains. Multiple coherence asks what a whole set of them explains at once, which is the only useful question in a MIMO test: two shakers driving one article are correlated with each other, so a response can look poorly coherent with either one alone while being fully accounted for by the pair.

For a response x and references r,

gamma^2 = (Grx^H Grr^-1 Grx) / Gxx

— the power of the best linear prediction of x from all the references at once, over the power actually measured. With one reference it collapses to the ordinary coherence, which is the cheapest check that the algebra is right.

Built on the same frames compute_psds averages, so it covers the stretch of record the averaging view has set and no other: a coherence worked out over the whole file would describe a different measurement from the PSD beside it.

references and the framing are _cross_spectral_frame's, the same ones compute_frfs uses — so the coherence beside an FRF is that FRF's coherence and not a differently-framed one.

A pseudo-inverse rather than a solve, because two shakers driving one article can be very nearly the same drive and the reference matrix is then close to singular — where a solve raises or returns nonsense, a pseudo-inverse gives the least-squares answer the estimate is asking for anyway.

Parameters:

Name	Type	Description	Default
references	sequence of str	The drive DOFs. Detected from the record when omitted.	None
averaging	Averaging	How to cut the record into frames. Defaults to the record's own averaging, or one frame per record.	None

Returns:

Type	Description
MultipleCoherence	One curve per response channel.

Source code in src/visualdynamics/core/data.py

def compute_multiple_coherence(
        self, references: Sequence[str] | None = None,
        averaging: Averaging | None = None) -> MultipleCoherence:
    """How much of each response the drives together account for.

    Ordinary coherence asks what one reference explains. Multiple
    coherence asks what a whole set of them explains at once, which
    is the only useful question in a MIMO test: two shakers driving
    one article are correlated with each other, so a response can
    look poorly coherent with either one alone while being fully
    accounted for by the pair.

    For a response x and references r,

        gamma^2 = (Grx^H Grr^-1 Grx) / Gxx

    — the power of the best linear prediction of x from all the
    references at once, over the power actually measured. With one
    reference it collapses to the ordinary coherence, which is the
    cheapest check that the algebra is right.

    Built on the same frames `compute_psds` averages, so it covers
    the stretch of record the averaging view has set and no other:
    a coherence worked out over the whole file would describe a
    different measurement from the PSD beside it.

    `references` and the framing are `_cross_spectral_frame`'s, the
    same ones `compute_frfs` uses — so the coherence beside an FRF
    is that FRF's coherence and not a differently-framed one.

    A pseudo-inverse rather than a solve, because two shakers
    driving one article can be very nearly the same drive and the
    reference matrix is then close to singular — where a solve
    raises or returns nonsense, a pseudo-inverse gives the
    least-squares answer the estimate is asking for anyway.

    Parameters
    ----------
    references : sequence of str, optional
        The drive DOFs. Detected from the record when omitted.
    averaging : Averaging, optional
        How to cut the record into frames. Defaults to the record's
        own `averaging`, or one frame per record.

    Returns
    -------
    MultipleCoherence
        One curve per response channel.
    """
    frequencies, keys, drives, responses, cross, averages = \
        self._cross_spectral_frame('multiple coherence', references,
                                   averaging)

    # (lines, drives, drives), Hermitian and positive semi-definite
    grr = np.stack([np.stack([cross(i, j) for j in drives], axis=-1)
                    for i in drives], axis=-2)
    inverse = np.linalg.pinv(grr)

    rows = []
    for i in responses:
        grx = np.stack([cross(k, i) for k in drives], axis=-1)[..., None]
        explained = np.real(
            np.conj(np.swapaxes(grx, -1, -2)) @ inverse @ grx)[..., 0, 0]
        measured = np.real(cross(i, i))
        with np.errstate(divide='ignore', invalid='ignore'):
            value = np.where(measured > 0.0, explained / measured, 0.0)
        # a ratio of powers cannot exceed one; a few parts in 1e12
        # over is the arithmetic, and a real overshoot means too few
        # averages for the number of references, which is a property
        # of the analysis and not of the article
        rows.append(np.clip(value, 0.0, 1.0))
    return MultipleCoherence(
        frequencies, np.asarray(rows, dtype=np.float64),
        response_dof=[keys[i][0] for i in responses],
        comment=f'multiple coherence against {len(drives)} references, '
                f'{averages} averages')

to_sep005

to_sep005(name: str | None = None) -> list[dict[str, Any]]

This record as SEP 005 timeseries — the sdypy ecosystem's interchange form (io.sep005 reads them back).

timeseries = history.to_sep005('run 4')

Returns the standard's list form: usually one dict, and one per unit where channels mix. The sdypy validator holds a series' unit_str to a single string, so accelerometers beside a force gauge cannot be one compliant series — the list of series is exactly what the standard provides for that, and a split series wears the unit in its name so two of them stay distinguishable.

Values go exactly as they are held: SI where units are defined, with unit_str naming the SI unit, and the file's raw numbers where they are not, with unit_str empty — the standard allows an empty unit, and inventing one would claim a scale nobody declared. fs says the sampling when it is even; an uneven record sends its time vector, which the standard equally accepts. quantity rides where the standard has a letter for what a series measures.

name defaults to the comment when the record carries one, because a SEP 005 series must be named and the comment is the nearest thing to a name an object holds — the project knows what it called this record, the record does not.

Parameters:

Name	Type	Description	Default
name	str	A name for the series.	None

Returns:

Type	Description
list of dict	One SEP 005 timeseries mapping per channel.

Source code in src/visualdynamics/core/data.py

def to_sep005(self, name: str | None = None) -> list[dict[str, Any]]:
    """This record as SEP 005 timeseries — the sdypy ecosystem's
    interchange form (`io.sep005` reads them back).

        timeseries = history.to_sep005('run 4')

    Returns the standard's **list** form: usually one dict, and one
    per unit where channels mix. The sdypy validator holds a
    series' ``unit_str`` to a single string, so accelerometers
    beside a force gauge cannot be one compliant series — the list
    of series is exactly what the standard provides for that, and a
    split series wears the unit in its name so two of them stay
    distinguishable.

    Values go exactly as they are held: SI where units are defined,
    with ``unit_str`` naming the SI unit, and the file's raw
    numbers where they are not, with ``unit_str`` empty — the
    standard allows an empty unit, and inventing one would claim a
    scale nobody declared. ``fs`` says the sampling when it is
    even; an uneven record sends its ``time`` vector, which the
    standard equally accepts. ``quantity`` rides where the
    standard has a letter for what a series measures.

    `name` defaults to the comment when the record carries one,
    because a SEP 005 series must be named and the comment is the
    nearest thing to a name an object holds — the project knows
    what it called this record, the record does not.

    Parameters
    ----------
    name : str, optional
        A name for the series.

    Returns
    -------
    list of dict
        One SEP 005 timeseries mapping per channel.
    """
    from ..io.sep005 import DIMENSION_TO_QUANTITY
    from ..units import SI

    name = str(name or next((c for c in self.comment if c),
                            'time history'))
    groups: dict[tuple[str, str | None], list[int]] = {}
    for i in range(self.num_records):
        unit = (SI.label(self.ordinate_dim[i])
                if self.ordinate_unit[i] is not None else '')
        letter = DIMENSION_TO_QUANTITY.get(self.known_dim(i))
        groups.setdefault((unit, letter), []).append(i)
    data = np.asarray(np.real(self.ordinate), dtype=float)
    out = []
    for (unit, letter), rows in groups.items():
        series: dict[str, Any] = {
            # (n,) for a lone channel — the standard's preferred
            # single-channel shape, and the one the sdypy validator
            # measures a `time` vector against correctly
            'data': data[rows] if len(rows) > 1 else data[rows[0]],
            'name': (name if len(groups) == 1
                     else f'{name} [{unit or letter or "raw"}]'),
            'channel_name': [str(self.response_dof[i]) for i in rows],
            'unit_str': unit,
        }
        try:
            series['fs'] = float(self.sample_rate)
        except ValueError:
            series['time'] = np.asarray(self.abscissa, dtype=float)
        if letter is not None:
            series['quantity'] = letter
        out.append(series)
    return out

Spectrum

Spectrum(
    abscissa: ArrayLike,
    ordinate: ArrayLike,
    response_dof: str | Sequence[str],
    reference_dof: str | Sequence[str] | None = None,
    ordinate_dim: str | Sequence[str] | None = None,
    comment: str | Sequence[str] | None = None,
    ordinate_unit: str | Sequence[str | None] | None = None,
    reference_unit: str
    | Sequence[str | None]
    | None = None,
    dimension_hint: str
    | Sequence[str | None]
    | None = None,
    block: str | Sequence[str] | None = None,
)

Bases: DataArray

A linear spectrum: amplitude and phase at each frequency line.

The complex average of a record's frames, not a power average — so content whose phase is random frame to frame averages toward zero, which is what makes this the wrong reading for burst random and the right one for a deterministic signal. A Psd is the power average and does not have that property.

Methods:

Name	Description
animate	The operating deflection shape at one frequency line, moving

Source code in src/visualdynamics/core/data.py

def __init__(self, abscissa: ArrayLike, ordinate: ArrayLike,
             response_dof: str | Sequence[str],
             reference_dof: str | Sequence[str] | None = None,
             ordinate_dim: str | Sequence[str] | None = None,
             comment: str | Sequence[str] | None = None,
             ordinate_unit: str | Sequence[str | None] | None = None,
             reference_unit: str | Sequence[str | None] | None = None,
             dimension_hint: str | Sequence[str | None] | None = None,
             block: str | Sequence[str] | None = None) -> None:
    self.abscissa: np.ndarray = np.asarray(abscissa, dtype=np.float64)
    ordinate = np.atleast_2d(np.asarray(
        ordinate, dtype=np.complex128 if self.complex_ordinate else np.float64))
    self.ordinate: np.ndarray = ordinate
    n = ordinate.shape[0]
    if self.abscissa.ndim != 1 or ordinate.shape[1] != len(self.abscissa):
        raise ValueError(
            f'ordinate shape {ordinate.shape} does not match '
            f'abscissa length {len(self.abscissa)}')
    # a channel whose node or direction was never recorded imports
    # as a DOF of '' — see `validate.dofs`
    self.response_dof: list[str] = _dofs(
        self._str_list(response_dof, n, 'response_dof'), 'response DOF',
        allow_unknown=True)
    self.reference_dof: list[str] | None
    if reference_dof is None:
        if self.needs_reference:
            raise ValueError(f'{type(self).__name__} requires reference_dof')
        self.reference_dof = None
    else:
        self.reference_dof = _dofs(
            self._str_list(reference_dof, n, 'reference_dof'),
            'reference DOF', allow_unknown=True)
    # A record is its response plus, sometimes, one more thing. For a
    # measurement against something else that is the reference DOF; for
    # the same measurement repeated it is which repeat — an average, a
    # run, a temperature. `block` holds the short label of that second
    # kind, and nothing about it is time-specific.
    self.block: list[str] | None = (None if block is None
                                    else self._str_list(block, n, 'block'))
    self.ordinate_dim: list[str] = self._str_list(
        UNKNOWN if ordinate_dim is None else ordinate_dim, n, 'ordinate_dim')
    for dim in set(self.ordinate_dim):
        parse_dimension(dim)  # validates
    self.ordinate_unit: list[str | None] = self._opt_list(
        ordinate_unit, n, 'ordinate_unit')
    self.reference_unit: list[str | None] = self._opt_list(
        reference_unit, n, 'reference_unit')
    self.comment: list[str] = self._str_list(
        comment if comment is not None else '', n, 'comment')
    self.dimension_hint: list[str | None] = self._opt_list(
        dimension_hint, n, 'dimension_hint')
    for hint in set(self.dimension_hint):
        if hint is not None:
            parse_dimension(hint)  # validates
    self._fill_si_units()
    self._drop_stale_hints()

Methods:

animate

animate(
    geometry: Any,
    frequency: float | None = None,
    **kwargs: Any,
) -> Any

The operating deflection shape at one frequency line, moving on a geometry as the GUI animates it. Defaults to the strongest line; frequency picks another.

Parameters:

Name	Type	Description	Default
geometry	Geometry	The geometry to move.	required
frequency	float	Which frequency line. The strongest when omitted.	None
**kwargs	Any	Passed through to the scene.	{}

Returns:

Type	Description
object	The plot widget or plotter.

Source code in src/visualdynamics/core/data.py

def animate(self, geometry: Any, frequency: float | None = None,
            **kwargs: Any) -> Any:
    """The operating deflection shape at one frequency line, moving
    on a geometry as the GUI animates it. Defaults to the strongest
    line; `frequency` picks another.

    Parameters
    ----------
    geometry : Geometry
        The geometry to move.
    frequency : float, optional
        Which frequency line. The strongest when omitted.
    **kwargs
        Passed through to the scene.

    Returns
    -------
    object
        The plot widget or plotter.
    """
    from ..viz.animate import animate_ods
    return animate_ods(geometry, self, frequency=frequency, **kwargs)

Frf

Frf(
    abscissa: ArrayLike,
    ordinate: ArrayLike,
    response_dof: str | Sequence[str],
    reference_dof: str | Sequence[str] | None = None,
    ordinate_dim: str | Sequence[str] | None = None,
    comment: str | Sequence[str] | None = None,
    ordinate_unit: str | Sequence[str | None] | None = None,
    reference_unit: str
    | Sequence[str | None]
    | None = None,
    dimension_hint: str
    | Sequence[str | None]
    | None = None,
    block: str | Sequence[str] | None = None,
)

Bases: DataArray

Frequency response function: response per unit reference.

Methods:

Name	Description
plot_cmif	The CMIF the fitting screen draws; with shapes the modal
animate	The operating deflection shape at one frequency line, moving

Source code in src/visualdynamics/core/data.py

def __init__(self, abscissa: ArrayLike, ordinate: ArrayLike,
             response_dof: str | Sequence[str],
             reference_dof: str | Sequence[str] | None = None,
             ordinate_dim: str | Sequence[str] | None = None,
             comment: str | Sequence[str] | None = None,
             ordinate_unit: str | Sequence[str | None] | None = None,
             reference_unit: str | Sequence[str | None] | None = None,
             dimension_hint: str | Sequence[str | None] | None = None,
             block: str | Sequence[str] | None = None) -> None:
    self.abscissa: np.ndarray = np.asarray(abscissa, dtype=np.float64)
    ordinate = np.atleast_2d(np.asarray(
        ordinate, dtype=np.complex128 if self.complex_ordinate else np.float64))
    self.ordinate: np.ndarray = ordinate
    n = ordinate.shape[0]
    if self.abscissa.ndim != 1 or ordinate.shape[1] != len(self.abscissa):
        raise ValueError(
            f'ordinate shape {ordinate.shape} does not match '
            f'abscissa length {len(self.abscissa)}')
    # a channel whose node or direction was never recorded imports
    # as a DOF of '' — see `validate.dofs`
    self.response_dof: list[str] = _dofs(
        self._str_list(response_dof, n, 'response_dof'), 'response DOF',
        allow_unknown=True)
    self.reference_dof: list[str] | None
    if reference_dof is None:
        if self.needs_reference:
            raise ValueError(f'{type(self).__name__} requires reference_dof')
        self.reference_dof = None
    else:
        self.reference_dof = _dofs(
            self._str_list(reference_dof, n, 'reference_dof'),
            'reference DOF', allow_unknown=True)
    # A record is its response plus, sometimes, one more thing. For a
    # measurement against something else that is the reference DOF; for
    # the same measurement repeated it is which repeat — an average, a
    # run, a temperature. `block` holds the short label of that second
    # kind, and nothing about it is time-specific.
    self.block: list[str] | None = (None if block is None
                                    else self._str_list(block, n, 'block'))
    self.ordinate_dim: list[str] = self._str_list(
        UNKNOWN if ordinate_dim is None else ordinate_dim, n, 'ordinate_dim')
    for dim in set(self.ordinate_dim):
        parse_dimension(dim)  # validates
    self.ordinate_unit: list[str | None] = self._opt_list(
        ordinate_unit, n, 'ordinate_unit')
    self.reference_unit: list[str | None] = self._opt_list(
        reference_unit, n, 'reference_unit')
    self.comment: list[str] = self._str_list(
        comment if comment is not None else '', n, 'comment')
    self.dimension_hint: list[str | None] = self._opt_list(
        dimension_hint, n, 'dimension_hint')
    for hint in set(self.dimension_hint):
        if hint is not None:
            parse_dimension(hint)  # validates
    self._fill_si_units()
    self._drop_stale_hints()

Methods:

plot_cmif

plot_cmif(shapes: Any = None, **kwargs: Any) -> Any

The CMIF the fitting screen draws; with shapes the modal model's synthesis is drawn dashed over the measurement.

Parameters:

Name	Type	Description	Default
shapes	ShapeSet	A modal fit, drawn as the synthesised CMIF over the measured one.	None
**kwargs	Any	Passed through to the plotting layer.	{}

Returns:

Type	Description
object	The plot widget or plotter.

Source code in src/visualdynamics/core/data.py

def plot_cmif(self, shapes: Any = None, **kwargs: Any) -> Any:
    """The CMIF the fitting screen draws; with `shapes` the modal
    model's synthesis is drawn dashed over the measurement.

    Parameters
    ----------
    shapes : ShapeSet, optional
        A modal fit, drawn as the synthesised CMIF over the measured one.
    **kwargs
        Passed through to the plotting layer.

    Returns
    -------
    object
        The plot widget or plotter.
    """
    from ..plot import plot_cmif
    return plot_cmif(self, shapes, **kwargs)

animate

animate(
    geometry: Any,
    frequency: float | None = None,
    **kwargs: Any,
) -> Any

The operating deflection shape at one frequency line, moving on a geometry as the GUI animates it. Defaults to the strongest line; frequency picks another.

Parameters:

Name	Type	Description	Default
geometry	Geometry	The geometry to move.	required
frequency	float	Which frequency line. The strongest when omitted.	None
**kwargs	Any	Passed through to the scene.	{}

Returns:

Type	Description
object	The plot widget or plotter.

Source code in src/visualdynamics/core/data.py

def animate(self, geometry: Any, frequency: float | None = None,
            **kwargs: Any) -> Any:
    """The operating deflection shape at one frequency line, moving
    on a geometry as the GUI animates it. Defaults to the strongest
    line; `frequency` picks another.

    Parameters
    ----------
    geometry : Geometry
        The geometry to move.
    frequency : float, optional
        Which frequency line. The strongest when omitted.
    **kwargs
        Passed through to the scene.

    Returns
    -------
    object
        The plot widget or plotter.
    """
    from ..viz.animate import animate_ods
    return animate_ods(geometry, self, frequency=frequency, **kwargs)

Coherence

Coherence(*args: Any, **kwargs: Any)

Bases: _CoherenceBase

Ordinary coherence: one response against one reference.

UFF dataset 58 calls this function type 6, and gives multiple coherence a code of its own — see MultipleCoherence. They are not the same object with an optional reference: one is a matrix of pairs and the other is one curve per response, and a file that says 6 is promising a reference DOF per record.

Source code in src/visualdynamics/core/data.py

def __init__(self, *args: Any, **kwargs: Any) -> None:
    # whatever a caller passes, a coherence is dimensionless
    kwargs['ordinate_dim'] = 'dimensionless'
    kwargs.setdefault('ordinate_unit', 'dimensionless')
    super().__init__(*args, **kwargs)

MultipleCoherence

MultipleCoherence(*args: Any, **kwargs: Any)

Bases: _CoherenceBase

How much of a response all the references together account for.

One curve per response channel, with no reference of its own — the references are summed over, which is the whole point of it. This is what a Rattlesnake modal survey reports, and what its random environment reports against its control channels.

UFF dataset 58 function type 26. visualdynamics wrote 6 here for a while, which claimed ordinary coherence and a reference DOF that was never there.

Source code in src/visualdynamics/core/data.py

def __init__(self, *args: Any, **kwargs: Any) -> None:
    # whatever a caller passes, a coherence is dimensionless
    kwargs['ordinate_dim'] = 'dimensionless'
    kwargs.setdefault('ordinate_unit', 'dimensionless')
    super().__init__(*args, **kwargs)

Psd

Psd(
    *args: Any,
    bandwidth: ArrayLike | None = None,
    **kwargs: Any,
)

Bases: DataArray

Power spectral density: the declared unit is the engineering unit whose square-per-Hz the values are in (declare 'g' for g^2/Hz).

Held complex because a cross spectrum is complex — the phase between two channels is most of what a CPSD is for. An autospectrum is not: a channel against itself is a magnitude squared, real by construction. So the type allows complex and the object stores what it actually has, which for a specification or a set of ASDs is a real array of half the size.

Methods:

Name	Description
principal_shapes	The dominant shape of the cross-spectral matrix at each line,
animate	This set on a geometry, as the GUI shows it.
area	The area under one record, over a band or over all of it.
to_octave	This spectrum integrated onto proportional bands.
bin_widths	The width of every line's own bin.
bin_bounds	(left, right) of every line's own bin — see octave.bin_bounds.

Source code in src/visualdynamics/core/data.py

def __init__(self, *args: Any, bandwidth: ArrayLike | None = None,
             **kwargs: Any) -> None:
    super().__init__(*args, **kwargs)
    if not has_phase(self.ordinate):
        self.ordinate: np.ndarray = np.ascontiguousarray(self.ordinate.real)
    if bandwidth is not None:
        bandwidth = np.asarray(bandwidth, dtype=np.float64)
        if bandwidth.shape != self.abscissa.shape:
            raise ValueError(
                f'bandwidth has shape {bandwidth.shape}, expected '
                f'{self.abscissa.shape} to match the abscissa')
        self.bandwidth = bandwidth

Methods:

principal_shapes

principal_shapes(
    quantity: str | None = None,
) -> tuple[list[str], ndarray, str]

The dominant shape of the cross-spectral matrix at each line, as (dofs, shapes (channels x lines), quantity).

The channels of one quantity form a square Hermitian matrix per line; its largest eigenvalue's eigenvector, scaled by the square root of that eigenvalue, is the principal operating deflection shape — the direction of the output spectra's own CMIF, with each channel's phase relative to the others and no reference to choose. quantity picks which channels (the commonest when not told). Refuses a set with no cross records — an autospectrum set has no phase and its reading is the envelope — and an incomplete block, whose eigenvectors would be shapes of holes.

Parameters:

Name	Type	Description	Default
quantity	str	Which quantity, for a mixed object.	None

Returns:

Type	Description
tuple of (list of str, numpy.ndarray, str)	The DOF labels, the dominant shape at each line, and the quantity they are in.

Source code in src/visualdynamics/core/data.py

def principal_shapes(self, quantity: str | None = None
                     ) -> tuple[list[str], np.ndarray, str]:
    """The dominant shape of the cross-spectral matrix at each line,
    as (dofs, shapes (channels x lines), quantity).

    The channels of one quantity form a square Hermitian matrix per
    line; its largest eigenvalue's eigenvector, scaled by the square
    root of that eigenvalue, is the principal operating deflection
    shape — the direction of the output spectra's own CMIF, with
    each channel's phase relative to the others and no reference to
    choose. `quantity` picks which channels (the commonest when not
    told). Refuses a set with no cross records — an autospectrum set
    has no phase and its reading is the envelope — and an incomplete
    block, whose eigenvectors would be shapes of holes.

    Parameters
    ----------
    quantity : str, optional
        Which quantity, for a mixed object.

    Returns
    -------
    tuple of (list of str, numpy.ndarray, str)
        The DOF labels, the dominant shape at each line, and the
        quantity they are in.
    """
    if self.reference_dof is None:
        raise ValueError('no cross records — an autospectrum set '
                         'has no phase')
    factors = [channel_quantities(self.known_dim(i))
               for i in range(self.num_records)]
    kinds = [response for response, _reference in factors]
    if quantity is None:
        kinds_present = [k for k in kinds if k != UNKNOWN]
        if not kinds_present:
            raise ValueError('no record names a quantity')
        quantity = max(set(kinds_present), key=kinds_present.count)
    channels: list[str] = []
    for i, dof in enumerate(self.response_dof):
        if kinds[i] == quantity and dof not in channels:
            channels.append(dof)
    n = len(channels)
    if n < 2:
        raise ValueError(f'fewer than two {quantity} channels')
    index = {dof: k for k, dof in enumerate(channels)}
    matrix = np.zeros((len(self.abscissa), n, n), dtype=np.complex128)
    filled = np.zeros((n, n), dtype=bool)
    for i in range(self.num_records):
        row = index.get(self.response_dof[i])
        column = index.get(self.reference_dof[i])
        if (row is None or column is None
                or (kinds[i], factors[i][1]) != (quantity, quantity)):
            continue
        matrix[:, row, column] = self.ordinate[i]
        filled[row, column] = True
    if not filled.all():
        raise ValueError(f'the {quantity} cross-spectral block is '
                         'incomplete')
    values, vectors = np.linalg.eigh(matrix)
    principal = (vectors[:, :, -1]
                 * np.sqrt(np.maximum(values[:, -1:], 0.0)))
    return channels, principal.T.copy(), quantity

animate

animate(
    geometry: Any,
    frequency: float | None = None,
    quantity: str | None = None,
    **kwargs: Any,
) -> Any

This set on a geometry, as the GUI shows it.

A CPSD — cross records present — animates its principal operating deflection shape: the dominant eigenvector of the cross-spectral matrix per line, each channel's phase relative to the others. An autospectrum set has no phase, so it shows the envelope instead: two copies deflected ±sqrt(PSD), colour reading dB below the loudest node at any line. Defaults to the strongest line; frequency picks another, quantity which measurement deflects.

Parameters:

Name	Type	Description	Default
geometry	Geometry	The geometry to move.	required
frequency	float	Which frequency line.	None
quantity	str	Which quantity, for a mixed object.	None
**kwargs	Any	Passed through to the scene.	{}

Returns:

Type	Description
object	The plot widget or plotter.

Source code in src/visualdynamics/core/data.py

def animate(self, geometry: Any, frequency: float | None = None,
            quantity: str | None = None, **kwargs: Any) -> Any:
    """This set on a geometry, as the GUI shows it.

    A CPSD — cross records present — animates its principal
    operating deflection shape: the dominant eigenvector of the
    cross-spectral matrix per line, each channel's phase relative
    to the others. An autospectrum set has no phase, so it shows
    the envelope instead: two copies deflected ±sqrt(PSD), colour
    reading dB below the loudest node at any line. Defaults to the
    strongest line; `frequency` picks another, `quantity` which
    measurement deflects.

    Parameters
    ----------
    geometry : Geometry
        The geometry to move.
    frequency : float, optional
        Which frequency line.
    quantity : str, optional
        Which quantity, for a mixed object.
    **kwargs
        Passed through to the scene.

    Returns
    -------
    object
        The plot widget or plotter.
    """
    try:
        dofs, shapes, _used = self.principal_shapes(quantity)
    except ValueError:
        from ..viz.animate import animate_envelope
        return animate_envelope(geometry, self, frequency=frequency,
                                quantity=quantity, **kwargs)
    return Spectrum(self.abscissa, shapes,
                    response_dof=dofs).animate(
        geometry, frequency, **kwargs)

area

area(
    record: int = 0,
    low: float | None = None,
    high: float | None = None,
) -> float

The area under one record, over a band or over all of it.

The one integral. Whichever way this spectrum is read, it is read the same way here as it is drawn — that is what the field above is for, and why nothing outside this method chooses.

Units are the ordinate's times frequency, so the square root of it is an RMS for a PSD.

Parameters:

Name	Type	Description	Default
record	int	Which record.	0
low	float	The band to integrate over.	None
high	float	The band to integrate over.	None

Returns:

Type	Description
float	The area beneath the curve — the mean square, whose root is the RMS.

Source code in src/visualdynamics/core/data.py

def area(self, record: int = 0, low: float | None = None,
         high: float | None = None) -> float:
    """The area under one record, over a band or over all of it.

    The one integral. Whichever way this spectrum is read, it is
    read the same way here as it is drawn — that is what the field
    above is for, and why nothing outside this method chooses.

    Units are the ordinate's times frequency, so the square root of
    it is an RMS for a PSD.

    Parameters
    ----------
    record : int, default 0
        Which record.
    low, high : float, optional
        The band to integrate over.

    Returns
    -------
    float
        The area beneath the curve — the mean square, whose root
        is the RMS.
    """
    from .compliance import log_log_area

    values = np.real(np.asarray(self.ordinate[record], dtype=float))
    if self.interpolation == 'log_log':
        return log_log_area(self.abscissa, values, low, high)
    left, right = self.bin_bounds()
    if low is not None:
        left, right = np.maximum(left, low), np.maximum(right, low)
    if high is not None:
        left, right = np.minimum(left, high), np.minimum(right, high)
    width = np.maximum(right - left, 0.0)
    good = np.isfinite(values) & (values >= 0.0) & np.isfinite(width)
    if not good.any():
        return float('nan')
    return float(np.sum(values[good] * width[good]))

to_octave

to_octave(
    per_octave: int | None = None,
    low: float | None = None,
    high: float | None = None,
) -> Psd

This spectrum integrated onto proportional bands.

An integration, not a resampling: each band takes the mean-square content that falls in it, divided by its own width, so the area under the spectrum — and therefore the RMS it carries — is unchanged. Reading the narrowband curve at each band centre would throw away everything between the centres.

The band grid is absolute (see visualdynamics.core.octave), so this needs no specification to be told about: the frequency range only chooses which bands of the one fixed grid come back, and two runs banded the same way land on the same bands whatever their ranges were.

Cross terms come through complex, which is what makes this work for a CPSD as well as a PSD.

Parameters:

Name	Type	Description	Default
per_octave	int	Bands per octave.	None
low	float	The band to cover.	None
high	float	The band to cover.	None

Returns:

Type	Description
Psd	The same power, arranged on proportional bands.

Source code in src/visualdynamics/core/data.py

def to_octave(self, per_octave: int | None = None,
              low: float | None = None,
              high: float | None = None) -> Psd:
    """This spectrum integrated onto proportional bands.

    An integration, not a resampling: each band takes the
    mean-square content that falls in it, divided by its own width,
    so the area under the spectrum — and therefore the RMS it
    carries — is unchanged. Reading the narrowband curve at each
    band centre would throw away everything between the centres.

    The band grid is absolute (see `visualdynamics.core.octave`), so this
    needs no specification to be told about: the frequency range
    only chooses which bands of the one fixed grid come back, and
    two runs banded the same way land on the same bands whatever
    their ranges were.

    Cross terms come through complex, which is what makes this work
    for a CPSD as well as a PSD.

    Parameters
    ----------
    per_octave : int, optional
        Bands per octave.
    low, high : float, optional
        The band to cover.

    Returns
    -------
    Psd
        The same power, arranged on proportional bands.
    """
    from .octave import PER_OCTAVE, bands, resample

    frequencies = np.asarray(self.abscissa, dtype=float)
    positive = frequencies[frequencies > 0.0]
    if positive.size < 2:
        raise ValueError(
            'octave bands need a spectrum with frequency lines above '
            'zero; this one has ' + str(positive.size))
    # the range the bands have to span is the range the *bins*
    # cover, not the range the centres do. A line at 0.5 Hz stands
    # for a bin reaching down to 0.25, and a grid that started at
    # the line would leave that half outside every band — which is
    # a small loss of area and an unnecessary one.
    reach, beyond = self.bin_bounds()
    inside = reach[reach > 0.0]
    low = (float(inside.min()) if inside.size else float(positive.min())) \
        if low is None else float(low)
    high = float(beyond.max()) if high is None else float(high)
    per_octave = PER_OCTAVE if per_octave is None else int(per_octave)
    centres, widths, bounds = bands(low, high, per_octave)
    banded = resample(frequencies, self.ordinate, bounds, widths,
                      source=self.bin_bounds())
    out = Psd(centres, banded,
              response_dof=list(self.response_dof),
              reference_dof=(None if self.reference_dof is None
                             else list(self.reference_dof)),
              ordinate_dim=list(self.ordinate_dim),
              ordinate_unit=list(self.ordinate_unit),
              reference_unit=list(self.reference_unit),
              dimension_hint=list(self.dimension_hint),
              block=None if self.block is None else list(self.block),
              bandwidth=widths,
              comment=f'1/{per_octave} octave bands')
    # banding conserves the area and changes nothing about what the
    # object is — a held comparison scale included, so the report's
    # own banding compares what the narrowband comparison compared
    out.scale_db = self.scale_db
    return out

bin_widths

bin_widths() -> ndarray

The width of every line's own bin.

Its own when it has one, the midpoints between neighbours when it does not — so a caller integrating a spectrum never has to ask which kind it is holding.

Source code in src/visualdynamics/core/data.py

def bin_widths(self) -> np.ndarray:
    """The width of every line's own bin.

    Its own when it has one, the midpoints between neighbours when
    it does not — so a caller integrating a spectrum never has to
    ask which kind it is holding.
    """
    if self.bandwidth is not None:
        return self.bandwidth
    return np.gradient(np.asarray(self.abscissa, dtype=float))

bin_bounds

bin_bounds() -> tuple[ndarray, ndarray]

(left, right) of every line's own bin — see octave.bin_bounds.

Source code in src/visualdynamics/core/data.py

def bin_bounds(self) -> tuple[np.ndarray, np.ndarray]:
    """(left, right) of every line's own bin — see `octave.bin_bounds`."""
    from .octave import bin_bounds

    return bin_bounds(np.asarray(self.abscissa, dtype=float),
                      self.bandwidth)

Srs

Srs(
    *args: Any,
    q: float | None = None,
    kind: str = "maximax",
    **kwargs: Any,
)

Bases: DataArray

A shock response spectrum: the peak an oscillator reached.

Not a spectrum of the shock. Every point is the largest response a single-degree-of-freedom oscillator of that natural frequency ever reached while its base was shaken by the measured transient — one number out of one whole run of one filter. Two shocks with the same SRS can look nothing alike, and an SRS cannot be turned back into a time history, because the phase that produced each peak is gone.

The abscissa is the oscillator's natural frequency, not a frequency in the shock — laid out in decades (log_abscissa): an SRS is specified at octave-spaced natural frequencies, and drawn linear the bottom five octaves crush into the left margin. The ordinate is in the quantity the base was measured in — an acceleration transient gives an acceleration SRS.

q and kind change what the curve means, so they belong to the object rather than to whoever happened to compute it: an SRS at Q = 10 and the same shock at Q = 50 are different curves, and a maximax reading is not a positive one. See visualdynamics.core.srs.

Attributes:

Name	Type	Description
damping	float	The damping ratio the amplification factor means.

Source code in src/visualdynamics/core/data.py

def __init__(self, *args: Any, q: float | None = None,
             kind: str = 'maximax', **kwargs: Any) -> None:
    from .srs import DEFAULT_Q

    super().__init__(*args, **kwargs)
    self.q: float = DEFAULT_Q if q is None else float(q)
    self.kind: str = str(kind)

Attributes

damping property

damping: float

The damping ratio the amplification factor means.

Bounded

Bounded(
    *args: Any,
    warning_lower: ArrayLike | None = None,
    warning_upper: ArrayLike | None = None,
    abort_lower: ArrayLike | None = None,
    abort_upper: ArrayLike | None = None,
    **kwargs: Any,
)

Limit curves held beside an ordinate, in the same units as it.

What a test was controlled to is a target and the room around it: how far the response may stray before the controller warns, and how far before it stops. That is true of a random vibration specification and of a shock one alike, and it is the same four curves either way, so it is written once here and mixed into both.

The limits are arrays beside the ordinate rather than four more data objects. They share the abscissa, the DOFs, the dimensions and the units by construction, which is the point: nothing can drift out of step, and _value_arrays carries them through every unit conversion the ordinate sees. limit() hands one back as a plain object of the underlying kind when a caller wants one, to plot or export as such.

Limits are per control channel, so a record that is not one has none: those rows are NaN, which plots as a gap rather than as a line at zero.

Attributes: warning_lower: The curve below which the controller warns, per record, in the ordinate's own units. NaN where a record is not a control channel, or where no such limit was written. warning_upper: The same, above the target. abort_lower: The curve below which the controller stops the test. abort_upper: The same, above the target.

Methods:

Name	Description
limit	One limit in its own right, or None if it has none.
display_limit	A limit in display units; undefined records pass through as-is.

Source code in src/visualdynamics/core/data.py

def __init__(self, *args: Any, warning_lower: ArrayLike | None = None,
             warning_upper: ArrayLike | None = None,
             abort_lower: ArrayLike | None = None,
             abort_upper: ArrayLike | None = None,
             **kwargs: Any) -> None:
    super().__init__(*args, **kwargs)
    self.limits: dict[str, np.ndarray] = {}
    given = {'warning_lower': warning_lower,
             'warning_upper': warning_upper,
             'abort_lower': abort_lower,
             'abort_upper': abort_upper}
    for name, values in given.items():
        if values is None:
            continue
        values = np.atleast_2d(np.asarray(values, dtype=np.float64))
        if values.shape != self.ordinate.shape:
            raise ValueError(
                f'{name} has shape {values.shape}, expected '
                f'{self.ordinate.shape} to match the ordinate')
        self.limits[name] = values

Methods:

limit

limit(name: str) -> DataArray | None

One limit in its own right, or None if it has none.

Source code in src/visualdynamics/core/data.py

def limit(self, name: str) -> DataArray | None:
    """One limit in its own right, or None if it has none."""
    if name not in self.limits:
        return None
    out = self.PLAIN(
        abscissa=self.abscissa, ordinate=self.limits[name],
        response_dof=self.response_dof,
        reference_dof=self.reference_dof, block=self.block,
        ordinate_dim=self.ordinate_dim,
        ordinate_unit=self.ordinate_unit,
        reference_unit=self.reference_unit,
        comment=[f'{c} {name}'.strip() for c in self.comment],
        **self._plain_kwargs())
    # a limit is written the way its specification is, so it is read
    # and integrated the same way. Left to its class default, the
    # limits of a computed specification would come back claiming to
    # be breakpoints of a power law.
    reading = getattr(self, 'interpolation', None)
    if reading is not None and hasattr(out, 'interpolation'):
        out.interpolation = reading
    return out

display_limit

display_limit(
    name: str, unit_system: UnitSystem
) -> ndarray | None

A limit in display units; undefined records pass through as-is.

Source code in src/visualdynamics/core/data.py

def display_limit(self, name: str,
                  unit_system: UnitSystem) -> np.ndarray | None:
    """A limit in display units; undefined records pass through as-is."""
    values = self.limits.get(name)
    if values is None:
        return None
    out = np.empty_like(values)
    for i, dim in enumerate(self.ordinate_dim):
        out[i] = (values[i] if dim == UNKNOWN
                  else unit_system.from_si(values[i], dim))
    return out

Specification

Specification(
    *args: Any,
    warning_lower: ArrayLike | None = None,
    warning_upper: ArrayLike | None = None,
    abort_lower: ArrayLike | None = None,
    abort_upper: ArrayLike | None = None,
    **kwargs: Any,
)

Bases: Bounded, Psd

What a random vibration test was controlled to: a PSD, and its band.

A specification is a PSD in every respect — same abscissa, same quantity, same conversions — with the four limit curves Bounded carries. So it inherits both, rather than reimplementing either.

Source code in src/visualdynamics/core/data.py

def __init__(self, *args: Any, warning_lower: ArrayLike | None = None,
             warning_upper: ArrayLike | None = None,
             abort_lower: ArrayLike | None = None,
             abort_upper: ArrayLike | None = None,
             **kwargs: Any) -> None:
    super().__init__(*args, **kwargs)
    self.limits: dict[str, np.ndarray] = {}
    given = {'warning_lower': warning_lower,
             'warning_upper': warning_upper,
             'abort_lower': abort_lower,
             'abort_upper': abort_upper}
    for name, values in given.items():
        if values is None:
            continue
        values = np.atleast_2d(np.asarray(values, dtype=np.float64))
        if values.shape != self.ordinate.shape:
            raise ValueError(
                f'{name} has shape {values.shape}, expected '
                f'{self.ordinate.shape} to match the ordinate')
        self.limits[name] = values

ShockSpecification

ShockSpecification(
    *args: Any,
    warning_lower: ArrayLike | None = None,
    warning_upper: ArrayLike | None = None,
    abort_lower: ArrayLike | None = None,
    abort_upper: ArrayLike | None = None,
    **kwargs: Any,
)

Bases: Bounded, Srs

What a shock test was controlled to: an SRS, and its band.

The same relationship a Specification has to a Psd. A shock target is written as a required SRS with tolerance either side of it — conventionally +6 dB and -3 dB, which is a factor of 2 up and 0.707 down — and the limits are those curves.

Written at one Q, and read at that Q: a target quoted at Q = 10 says nothing about what the same shock does to a Q = 50 oscillator, so the amplification travels with the target the way it travels with a measurement.

Source code in src/visualdynamics/core/data.py

def __init__(self, *args: Any, warning_lower: ArrayLike | None = None,
             warning_upper: ArrayLike | None = None,
             abort_lower: ArrayLike | None = None,
             abort_upper: ArrayLike | None = None,
             **kwargs: Any) -> None:
    super().__init__(*args, **kwargs)
    self.limits: dict[str, np.ndarray] = {}
    given = {'warning_lower': warning_lower,
             'warning_upper': warning_upper,
             'abort_lower': abort_lower,
             'abort_upper': abort_upper}
    for name, values in given.items():
        if values is None:
            continue
        values = np.atleast_2d(np.asarray(values, dtype=np.float64))
        if values.shape != self.ordinate.shape:
            raise ValueError(
                f'{name} has shape {values.shape}, expected '
                f'{self.ordinate.shape} to match the ordinate')
        self.limits[name] = values

TransientSpecification

TransientSpecification(
    abscissa: ArrayLike,
    ordinate: ArrayLike,
    response_dof: str | Sequence[str],
    reference_dof: str | Sequence[str] | None = None,
    ordinate_dim: str | Sequence[str] | None = None,
    comment: str | Sequence[str] | None = None,
    ordinate_unit: str | Sequence[str | None] | None = None,
    reference_unit: str
    | Sequence[str | None]
    | None = None,
    dimension_hint: str
    | Sequence[str | None]
    | None = None,
    block: str | Sequence[str] | None = None,
)

Bases: TimeHistory

What a transient test was controlled to: a target time history.

A different thing from a ShockSpecification, and the difference is worth keeping straight because the two get called by each other's names. A shock specification is an SRS: a required response spectrum, and a controller meets it by producing some transient whose spectrum lands inside the band. A transient specification is a waveform: this acceleration, sample by sample, and the controller inverts the structure's transfer function to reproduce it. Rattlesnake can run the second today; the first it cannot.

So this is a time history that happens to be a target, and the comparison it invites is against another time history — what the article actually did — rather than against a band. It carries no limits for that reason: a tolerance on a waveform is not a settled idea the way a tolerance on a spectrum is, and inventing one here would be inventing a convention rather than reading one.

Its derived spectra stay targets. A PSD of this is a Specification and an SRS of it is a ShockSpecification — both without limits, for the reason above — because the spectrum of a waveform the article was required to see is the spectrum it was required to see. Left as plain objects they would be indistinguishable from the response's own spectra but for a name, and the comparison between them would have to be made by hand instead of by type, which is the one thing this whole arrangement exists to avoid.

Methods:

Name	Description
psd_type	What a PSD of this record is: still a specification.
srs_type	What an SRS of this record is: a shock specification, for

Source code in src/visualdynamics/core/data.py

def __init__(self, abscissa: ArrayLike, ordinate: ArrayLike,
             response_dof: str | Sequence[str],
             reference_dof: str | Sequence[str] | None = None,
             ordinate_dim: str | Sequence[str] | None = None,
             comment: str | Sequence[str] | None = None,
             ordinate_unit: str | Sequence[str | None] | None = None,
             reference_unit: str | Sequence[str | None] | None = None,
             dimension_hint: str | Sequence[str | None] | None = None,
             block: str | Sequence[str] | None = None) -> None:
    self.abscissa: np.ndarray = np.asarray(abscissa, dtype=np.float64)
    ordinate = np.atleast_2d(np.asarray(
        ordinate, dtype=np.complex128 if self.complex_ordinate else np.float64))
    self.ordinate: np.ndarray = ordinate
    n = ordinate.shape[0]
    if self.abscissa.ndim != 1 or ordinate.shape[1] != len(self.abscissa):
        raise ValueError(
            f'ordinate shape {ordinate.shape} does not match '
            f'abscissa length {len(self.abscissa)}')
    # a channel whose node or direction was never recorded imports
    # as a DOF of '' — see `validate.dofs`
    self.response_dof: list[str] = _dofs(
        self._str_list(response_dof, n, 'response_dof'), 'response DOF',
        allow_unknown=True)
    self.reference_dof: list[str] | None
    if reference_dof is None:
        if self.needs_reference:
            raise ValueError(f'{type(self).__name__} requires reference_dof')
        self.reference_dof = None
    else:
        self.reference_dof = _dofs(
            self._str_list(reference_dof, n, 'reference_dof'),
            'reference DOF', allow_unknown=True)
    # A record is its response plus, sometimes, one more thing. For a
    # measurement against something else that is the reference DOF; for
    # the same measurement repeated it is which repeat — an average, a
    # run, a temperature. `block` holds the short label of that second
    # kind, and nothing about it is time-specific.
    self.block: list[str] | None = (None if block is None
                                    else self._str_list(block, n, 'block'))
    self.ordinate_dim: list[str] = self._str_list(
        UNKNOWN if ordinate_dim is None else ordinate_dim, n, 'ordinate_dim')
    for dim in set(self.ordinate_dim):
        parse_dimension(dim)  # validates
    self.ordinate_unit: list[str | None] = self._opt_list(
        ordinate_unit, n, 'ordinate_unit')
    self.reference_unit: list[str | None] = self._opt_list(
        reference_unit, n, 'reference_unit')
    self.comment: list[str] = self._str_list(
        comment if comment is not None else '', n, 'comment')
    self.dimension_hint: list[str | None] = self._opt_list(
        dimension_hint, n, 'dimension_hint')
    for hint in set(self.dimension_hint):
        if hint is not None:
            parse_dimension(hint)  # validates
    self._fill_si_units()
    self._drop_stale_hints()

Methods:

psd_type

psd_type() -> type[Psd]

What a PSD of this record is: still a specification.

The spectrum of a required waveform is itself a requirement, so it comes back as Specification rather than a plain Psd.

Source code in src/visualdynamics/core/data.py

def psd_type(self) -> type[Psd]:
    """What a PSD of this record is: still a specification.

    The spectrum of a required waveform is itself a requirement,
    so it comes back as `Specification` rather than a plain `Psd`.
    """
    return Specification

srs_type

srs_type() -> type[Srs]

What an SRS of this record is: a shock specification, for the same reason psd_type gives.

Source code in src/visualdynamics/core/data.py

def srs_type(self) -> type[Srs]:
    """What an SRS of this record is: a shock specification, for
    the same reason `psd_type` gives."""
    return ShockSpecification

Functions:

direction_code

direction_code(direction: str) -> int

'Z+' -> 3. The inverse of the table dof_string reads.

An unsigned direction means the positive one, which is how DOFs are usually written by hand.

Source code in src/visualdynamics/core/data.py

def direction_code(direction: str) -> int:
    """'Z+' -> 3. The inverse of the table `dof_string` reads.

    An unsigned direction means the positive one, which is how DOFs are
    usually written by hand.
    """
    text = str(direction).strip().upper()
    if text in _CODES:
        return _CODES[text]
    if text + '+' in _CODES:
        return _CODES[text + '+']
    raise ValueError(f'unknown direction {direction!r}')

parse_dof

parse_dof(text: str) -> tuple[int | None, str]

'101RX+' -> (101, 'RX+'). The inverse of dof_string().

Returns (None, direction) when the leading node number is missing or unparseable, so callers can report it rather than crash.

Source code in src/visualdynamics/core/data.py

def parse_dof(text: str) -> tuple[int | None, str]:
    """'101RX+' -> (101, 'RX+'). The inverse of dof_string().

    Returns (None, direction) when the leading node number is missing or
    unparseable, so callers can report it rather than crash.
    """
    text = str(text).strip()
    digits = 0
    while digits < len(text) and text[digits].isdigit():
        digits += 1
    if not digits:
        return None, text
    return int(text[:digits]), text[digits:]

has_phase

has_phase(values: ArrayLike) -> bool

Is there imaginary content here, or only the dust of computing it?

Source code in src/visualdynamics/core/data.py

def has_phase(values: ArrayLike) -> bool:
    """Is there imaginary content here, or only the dust of computing it?"""
    values = np.asarray(values)
    if not np.iscomplexobj(values):
        return False
    real = np.abs(values.real).max(initial=0.0)
    return bool(np.abs(values.imag).max(initial=0.0) > REAL_FLOOR * real)

channel_quantities

channel_quantities(dim: str) -> tuple[str, str]

(response, reference) quantities a record's dimension names.

A record's dimension composes its two channels' quantities — an FRF's 'acceleration/force', a CPSD cross term's 'accelerationforce/frequency', a diagonal's 'acceleration2/frequency' — and channel identity* needs the factors, not the compound: keyed by the whole dimension, one accelerometer's records split into a row per thing it was measured against, which is how a 13-channel CPSD came up 26 rows.

Source code in src/visualdynamics/core/data.py

def channel_quantities(dim: str) -> tuple[str, str]:
    """(response, reference) quantities a record's dimension names.

    A record's dimension composes its two channels' quantities — an
    FRF's 'acceleration/force', a CPSD cross term's
    'acceleration*force/frequency', a diagonal's
    'acceleration**2/frequency' — and channel *identity* needs the
    factors, not the compound: keyed by the whole dimension, one
    accelerometer's records split into a row per thing it was measured
    against, which is how a 13-channel CPSD came up 26 rows.
    """
    numerator = dim.partition('/')[0]
    if numerator.endswith('**2'):
        base = numerator.removesuffix('**2')
        return base, base
    response, star, reference = numerator.partition('*')
    if star:
        return response, reference
    denominator = dim.partition('/')[2]
    return numerator, denominator or numerator

frequency_axis

frequency_axis(mode: str | bool | None = ...) -> str

Read or set how frequency axes are drawn: 'log', 'linear', or 'default' (each kind of data by its own convention).

Parameters:

Name	Type	Description	Default
mode	('log', 'linear', 'default')	The choice to make. Omitted, the current one is read back. True and False stand for 'log' and 'linear'; None for 'default'.	'log'

Returns:

Type	Description
str	The choice in force after the call.

Source code in src/visualdynamics/core/data.py

def frequency_axis(mode: str | bool | None = ...) -> str:
    """Read or set how frequency axes are drawn: 'log', 'linear', or
    'default' (each kind of data by its own convention).

    Parameters
    ----------
    mode : {'log', 'linear', 'default'} or bool or None, optional
        The choice to make. Omitted, the current one is read back.
        True and False stand for 'log' and 'linear'; None for
        'default'.

    Returns
    -------
    str
        The choice in force after the call.
    """
    global _FREQUENCY_AXIS
    if mode is not ...:
        if mode in ('default', None):
            _FREQUENCY_AXIS = None
        elif mode in ('log', True):
            _FREQUENCY_AXIS = True
        elif mode in ('linear', False):
            _FREQUENCY_AXIS = False
        else:
            raise ValueError(f"frequency axis is 'log', 'linear' or "
                             f"'default', not {mode!r}")
    return {None: 'default', True: 'log', False: 'linear'}[_FREQUENCY_AXIS]

paired_channels

paired_channels(
    signal: Any, floor: Any
) -> list[tuple[int, int, str, str]]

The channels two densities share, as (signal row, floor row, DOF, quantity).

The pairing rule under every two-density reading (Brandon, 2026-08-25): matched by (DOF, quantity), never by index — a drive point carries an accelerometer and a load cell at one DOF, and keyed by DOF alone the dict collapsed them, silently losing half the drive-point channels from the first ratio drawn. Channels only one side measured are left out; sharing none refuses. Both sides must stand on the same frequency lines — neither a ratio nor an overlay means anything across mismatched bins, so that refuses rather than interpolating an answer nobody measured.

Here rather than inside density_ratio because the overlay of the two densities and the ratio of them are the same pair read two ways: the report draws them one above the other, and a figure that paired its channels differently from the figure below it would be two claims about one measurement.

Source code in src/visualdynamics/core/data.py

def paired_channels(signal: Any, floor: Any) -> list[tuple[int, int, str, str]]:
    """The channels two densities share, as (signal row, floor row,
    DOF, quantity).

    The pairing rule under every two-density reading (Brandon,
    2026-08-25): matched by **(DOF, quantity)**, never by index — a
    drive point carries an accelerometer and a load cell at one DOF,
    and keyed by DOF alone the dict collapsed them, silently losing
    half the drive-point channels from the first ratio drawn.
    Channels only one side measured are left out; sharing none
    refuses. Both sides must stand on the same frequency lines —
    neither a ratio nor an overlay means anything across mismatched
    bins, so that refuses rather than interpolating an answer nobody
    measured.

    Here rather than inside `density_ratio` because the overlay of
    the two densities and the ratio of them are the same pair read
    two ways: the report draws them one above the other, and a figure
    that paired its channels differently from the figure below it
    would be two claims about one measurement.
    """
    if len(signal.abscissa) != len(floor.abscissa) or not np.allclose(
            signal.abscissa, floor.abscissa, rtol=1e-9, atol=1e-12):
        raise ValueError(
            'the two densities stand on different frequency lines; '
            'compute both with the same framing and the ratio exists')

    def keys(density):
        dims = np.atleast_1d(density.ordinate_dim)
        return [(str(dof), str(dims[min(i, len(dims) - 1)]))
                for i, dof in enumerate(density.response_dof)]

    quiet = {key: j for j, key in enumerate(keys(floor))}
    shared = [(i, quiet[key], key[0], key[1])
              for i, key in enumerate(keys(signal)) if key in quiet]
    if not shared:
        raise ValueError('the two densities share no channel; there '
                         'is no ratio to take')
    return shared

density_ratio

density_ratio(signal: Any, floor: Any)

One density over another, channel by channel, line by line.

Channels pair by paired_channels; a zero floor line answers NaN, never infinity — on real hardware zero means below resolution, in a simulation it means the quiet was exactly silent, and neither is an infinite signal-to-noise.

Returns (abscissa, rows, dofs, dims) — the ratio as plain arrays plus each row's quantity, for whoever is reading it: the plot's ratio view, the stage, the report's ratio figure — every ratio divides here, and the quantity rides along so a mixed object's rows can be told apart.

Source code in src/visualdynamics/core/data.py

def density_ratio(signal: Any, floor: Any):
    """One density over another, channel by channel, line by line.

    Channels pair by `paired_channels`; a zero floor line answers
    NaN, never infinity — on real hardware zero means below
    resolution, in a simulation it means the quiet was exactly
    silent, and neither is an infinite signal-to-noise.

    Returns ``(abscissa, rows, dofs, dims)`` — the ratio as plain
    arrays plus each row's quantity, for whoever is reading it: the
    plot's ratio view, the stage, the report's ratio figure — every
    ratio divides here, and the quantity rides along so a mixed
    object's rows can be told apart.
    """
    rows, dofs, dims = [], [], []
    for i, j, dof, dim in paired_channels(signal, floor):
        top = np.real(np.asarray(signal.ordinate[i]))
        bottom = np.real(np.asarray(floor.ordinate[j]))
        with np.errstate(divide='ignore', invalid='ignore'):
            rows.append(np.where(bottom > 0.0, top / bottom, np.nan))
        dofs.append(dof)
        dims.append(dim)
    return np.asarray(signal.abscissa), np.asarray(rows), dofs, dims