Data and results¶

Arrays use time × feature (or time × factor). A valid mask means a query is supported, not that it was directly observed. GaussianMixtureSeries contains marginal distributions; TrajectorySamples contains joint paths. Result dataclass fields are listed below, including inherited fields.

Generated from the checked-out source

Signatures, field defaults, and available docstrings below are generated by scripts/build_api_reference.py. Constructor defaults are not a recommendation for every analysis. Semantic choices are explained in the API guide.

TimeSeries¶

from multimodalsrm import TimeSeries · Source

TimeSeries(values, times, mask=None)

Field	Type	Default
`values`	`np.ndarray`	`required`
`times`	`np.ndarray`	`required`
`mask`	`np.ndarray \| None`	`None`

SeriesResult¶

from multimodalsrm import SeriesResult · Source

SeriesResult(values, times, valid, metadata=dict())

Field	Type	Default
`values`	`np.ndarray`	`required`
`times`	`np.ndarray`	`required`
`valid`	`np.ndarray`	`required`
`metadata`	`dict`	`dict() (new per instance)`

PosteriorSeriesResult¶

from multimodalsrm import PosteriorSeriesResult · Source

PosteriorSeriesResult(values, times, valid, metadata=dict(), std=None)

Field	Type	Default
`values`	`np.ndarray`	`required`
`times`	`np.ndarray`	`required`
`valid`	`np.ndarray`	`required`
`metadata`	`dict`	`dict() (new per instance)`
`std`	`np.ndarray \| None`	`None`

An immutable result whose uncertainty is conditional on fitted parameters.

std matches values and uses the same units. Invalid entries in both arrays are NaN; a valid entry cannot have a negative standard deviation.

KernelEstimate¶

from multimodalsrm import KernelEstimate · Source

KernelEstimate(
    family,
    parameters,
    support,
    normalization,
    reference,
    fixed,
    learned,
    pooling,
    level,
    subject,
    modality,
    times,
    values,
    difference_from_group,
    impulse_mass,
    metadata,
)

Field	Type	Default
`family`	`str`	`required`
`parameters`	`dict`	`required`
`support`	`tuple`	`required`
`normalization`	`str`	`required`
`reference`	`str \| None`	`required`
`fixed`	`dict`	`required`
`learned`	`tuple`	`required`
`pooling`	`str`	`required`
`level`	`str`	`required`
`subject`	`object`	`required`
`modality`	`object`	`required`
`times`	`np.ndarray \| None`	`required`
`values`	`np.ndarray \| None`	`required`
`difference_from_group`	`np.ndarray \| None`	`required`
`impulse_mass`	`float \| None`	`required`
`metadata`	`dict`	`required`

Fitted response and optional sampled curve, with the scale convention.

Identity has impulse_mass=1 and values=None even when query times are supplied: a Dirac impulse is not an ordinary finite-valued density. Nonidentity values and group differences are time-by-one arrays.

KernelEstimate.valid¶

KernelEstimate.valid (property)

Sampled curve validity; an analytic impulse has no sampled values.

GaussianMixtureSeries¶

from multimodalsrm.bayesian import GaussianMixtureSeries · Source

GaussianMixtureSeries(component_means, component_variances, times, valid, metadata)

Field	Type	Default
`component_means`	`np.ndarray`	`required`
`component_variances`	`np.ndarray`	`required`
`times`	`np.ndarray`	`required`
`valid`	`np.ndarray`	`required`
`metadata`	`dict`	`required`

Equal-weight marginal mixtures; component shape is draw/time/feature.

These are marginal distributions, not independent joint trajectory draws. A MAP result is represented by one component. Interval endpoints are actual mixture quantiles, not Gaussian approximations based on the mixture SD.

GaussianMixtureSeries.values¶

GaussianMixtureSeries.values (property)

GaussianMixtureSeries.within_parameter_variance¶

GaussianMixtureSeries.within_parameter_variance (property)

GaussianMixtureSeries.between_parameter_variance¶

GaussianMixtureSeries.between_parameter_variance (property)

GaussianMixtureSeries.variance¶

GaussianMixtureSeries.variance (property)

GaussianMixtureSeries.std¶

GaussianMixtureSeries.std (property)

GaussianMixtureSeries.interval¶

GaussianMixtureSeries.interval(probability=0.95)

GaussianMixtureSeries.log_density¶

GaussianMixtureSeries.log_density(values)

Pointwise log density, excluding invalid query rows from scoring.

Zero-variance components are atoms, for which ordinary continuous density scoring is undefined; such results must not be scored here.

TrajectorySamples¶

from multimodalsrm.bayesian import TrajectorySamples · Source

TrajectorySamples(samples, times, valid, metadata)

Field	Type	Default
`samples`	`np.ndarray`	`required`
`times`	`np.ndarray`	`required`
`valid`	`np.ndarray`	`required`
`metadata`	`dict`	`required`

Joint paths with axes parameter draw / conditional draw / time / factor.

Conditional draws sharing a parameter draw are not additional independent posterior parameter samples. Invalid query rows contain NaN. Corresponding draws across runs from one sampling call share their parameter draw.

ParticipantCalibration¶

from multimodalsrm.bayesian import ParticipantCalibration · Source

ParticipantCalibration()

One new participant's MAP mappings in a frozen group's coordinates.

Created by group.calibrate or load_model. Preprocessing is explicit; transform consumes the units used during calibration. No group refit or parameter uncertainty calibration is implied by this object.

ParticipantCalibration.transform¶

ParticipantCalibration.transform(
    data,
    *,
    times,
    modalities=None,
    data_layout="runs",
    readout="gp",
)

Independently transform this participant on new stimulus runs.

CrossValidationResult¶

from multimodalsrm import CrossValidationResult · Source

CrossValidationResult(scores, summary, models, folds)

Field	Type	Default
`scores`	`list`	`required`
`summary`	`list`	`required`
`models`	`list`	`required`
`folds`	`list`	`required`

Unaggregated score rows, explicit summaries, and independent fold fits.

ModelComparisonResult¶

from multimodalsrm import ModelComparisonResult · Source

ModelComparisonResult(scores, predictions, errors, warnings=list())

Field	Type	Default
`scores`	`list`	`required`
`predictions`	`list`	`required`
`errors`	`list`	`required`
`warnings`	`list`	`list() (new per instance)`

ModelSelectionResult¶

from multimodalsrm import ModelSelectionResult · Source

ModelSelectionResult(
    best_name,
    best_estimator,
    scores,
    summary,
    folds,
    predictions,
    source,
    metric,
    status,
    refit_diagnostics=None,
)

Field	Type	Default
`best_name`	`str \| None`	`required`
`best_estimator`	`object \| None`	`required`
`scores`	`list`	`required`
`summary`	`list`	`required`
`folds`	`list`	`required`
`predictions`	`list`	`required`
`source`	`str`	`required`
`metric`	`str`	`required`
`status`	`str`	`required`
`refit_diagnostics`	`dict \| None`	`None`

NestedCrossValidationResult¶

from multimodalsrm import NestedCrossValidationResult · Source

NestedCrossValidationResult(scores, summary, models, folds, selections, predictions)

Field	Type	Default
`scores`	`list`	`required`
`summary`	`list`	`required`
`models`	`list`	`required`
`folds`	`list`	`required`
`selections`	`list`	`required`
`predictions`	`list`	`required`