specparam.SpectralTimeModel

class specparam.SpectralTimeModel(*args, **kwargs)

Model a spectrogram as a combination of aperiodic and periodic components.

WARNING: frequency and power values inputs must be in linear space. Passing in logged frequencies and/or power spectra is not detected, and will silently produce incorrect results.

Parameters:

peak_width_limitstuple of (float, float), optional, default: (0.5, 12.0)

Limits on possible peak width, in Hz, as (lower_bound, upper_bound).

max_n_peaksint, optional, default: inf

Maximum number of peaks to fit.

min_peak_heightfloat, optional, default: 0

Absolute threshold for detecting peaks. This threshold is defined in absolute units of the power spectrum (log power).

peak_thresholdfloat, optional, default: 2.0

Relative threshold for detecting peaks. Threshold is defined in relative units of the power spectrum (standard deviation).

aperiodic_mode{‘fixed’, ‘knee’} or Mode

Which approach to take for fitting the aperiodic component.

periodic_mode{‘gaussian’, ‘skewed_gaussian’, ‘cauchy’} or Mode

Which approach to take for fitting the periodic component.

algorithm{‘spectral_fit’} or Algorithm

The fitting algorithm to use.

algorithm_settingsdict

Setting for the algorithm.

metricsMetrics or list of Metric or list or str

Metrics definition(s) to use to evaluate the model.

bandsBands or dict or int or None, optional

Bands object with band definitions, or definition that can be turned into a Bands object.

debugbool, optional, default: False

Whether to run in debug mode. If in debug, any errors encountered during fitting will raise an error.

verbosebool, optional, default: True

Verbosity mode. If True, prints out warnings and general status updates.

**model_kwargs

Additional model fitting related keyword arguments. These are passed into the algorithm object.

Notes

• Input power spectra must be provided in linear scale. Internally they are stored in log10 scale, as this is what the model operates upon.

• Input power spectra should be smooth, as overly noisy power spectra may lead to bad fits. For example, raw FFT inputs are not appropriate. Where possible and appropriate, use longer time segments for power spectrum calculation to get smoother power spectra, as this will give better model fits.

• The time object inherits from the group model, overwriting the data and results objects with versions for fitting models across time. Temporally organized results are collected into the results.time_results attribute, which may include sub-selecting peaks per band (depending on settings). Note that the results.group_results attribute is also available, which maintains the full model results.

Attributes:

algorithmAlgorithm

Algorithm object with model fitting settings and procedures.

modesModes

Modes object with fit mode definitions.

dataData

Data object with spectral data and metadata.

resultsResults

Results object with model fit results and metrics.

__init__(*args, **kwargs)

Initialize time model object.

Methods

__init__(*args, **kwargs)	Initialize time model object.
add_data(freqs, power_spectra[, freq_range, ...])	Add data (frequencies and power spectrum values) to the current object.
add_modes(aperiodic_mode, periodic_mode)	Add modes definition to the object.
convert_results(bands)	Convert results to be organized across time.
copy()	Return a copy of the current object.
fit([freqs, spectrogram, freq_range, bands, ...])	Fit a spectrogram.
get_data([component, space])	Get a data component.
get_group(inds[, output_type])	Get a new model object with the specified sub-selection of model fits.
get_metrics(category[, measure])	Get requested metric(s) from the object.
get_model([ind, regenerate])	Get a model fit object for a specified index.
get_params(component[, field])	Return model fit parameters for specified feature(s).
load(file_name[, file_path, convert_results])	Load time data from file.
plot([plot_type, save_fig, file_name, file_path])	Plot a figure with subplots visualizing the parameters from a SpectralTimeModel object.
print_report_issue([concise])	Prints instructions on how to report bugs and/or problematic fits.
print_results([print_type, concise])	Print out SpectralTimeModel results.
print_settings([description, concise])	Print out the current settings.
report([freqs, spectrogram, freq_range, ...])	Fit a spectrogram and display a report, with a plot and printed results.
save(file_name[, file_path, append, ...])	Save out results and/or settings from time object.
save_model_report(index, file_name[, ...])	"Save out an individual model report for a specified model fit.
save_report(file_name[, file_path, add_settings])	Generate and save out a PDF report for models of a spectrogram.
to_df([bands])	Convert and extract the model results as a pandas object.

add_data(freqs, power_spectra, freq_range=None, clear_results=True)

Add data (frequencies and power spectrum values) to the current object.

Parameters:

freqs1d array

Frequency values for the power spectra, in linear space.

power_spectra2d array, shape=[n_power_spectra, n_freqs]

Matrix of power values, in linear space.

freq_rangelist of [float, float], optional

Frequency range to restrict power spectra to. If not provided, keeps the entire range.

clear_resultsbool, optional, default: True

Whether to clear prior results, if any are present in the object. This should only be set to False if data for the current results are being re-added.

Notes

If called on an object with existing data and/or results these will be cleared by this method call.

add_modes(aperiodic_mode, periodic_mode)

Add modes definition to the object.

Parameters:

aperiodic_modeMode or str

Mode for aperiodic component, or string specifying which mode to use.

periodic_modeMode or str

Mode for periodic component, or string specifying which mode to use.

convert_results(bands)

Convert results to be organized across time.

Parameters:

bandsBands or dict or int, optional

How to organize peaks into bands. If Bands or dict, uses band definitions. If int, extracts the first ‘n’ peaks. If not provided, uses band definition available in object.

copy()

Return a copy of the current object.

fit(freqs=None, spectrogram=None, freq_range=None, bands=None, n_jobs=1, progress=None, prechecks=True, convert_results=True)

Fit a spectrogram.

Parameters:

freqs1d array, optional

Frequency values for the spectrogram, in linear space.

spectrogram2d array, shape: [n_freqs, n_time_windows], optional

Spectrogram of power spectrum values, in linear space.

freq_rangelist of [float, float], optional

Frequency range to fit the model to. If not provided, fits the entire given range.

bandsBands or dict or int, optional

How to organize peaks into bands. If Bands or dict, uses band definitions. If int, extracts the first ‘n’ peaks.

n_jobsint, optional, default: 1

Number of jobs to run in parallel. 1 is no parallelization. -1 uses all available cores.

progress{None, ‘tqdm’, ‘tqdm.notebook’}, optional

Which kind of progress bar to use. If None, no progress bar is used.

prechecksbool, optional, default: True

Whether to run model fitting pre-checks.

convert_resultsbool, optional, default: True

Whether to convert results per spectrogram window to be organized over time.

Notes

Data is optional, if data has already been added to the object.

get_data(component='full', space='log')

Get a data component.

Parameters:

component{‘full’, ‘aperiodic’, ‘peak’}

Which data component to return.

‘full’ - full power spectrum ‘aperiodic’ - isolated aperiodic data component ‘peak’ - isolated peak data component

space{‘log’, ‘linear’}

Which space to return the data component in.

‘log’ - returns in log10 space. ‘linear’ - returns in linear space.

Returns:

output1d array

Specified data component, in specified spacing.

Notes

The ‘space’ parameter doesn’t just define the spacing of the data component values, but rather defines the space of the additive data definition such that power_spectrum = aperiodic_component + peak_component. With space set as ‘log’, this combination holds in log space. With space set as ‘linear’, this combination holds in linear space.

get_group(inds, output_type='time')

Get a new model object with the specified sub-selection of model fits.

Parameters:

indsarray_like of int or array_like of bool

Indices to extract from the object.

output_type{‘time’, ‘group’}, optional

Type of model object to extract:

‘time’ : SpectralTimeObject ‘group’ : SpectralGroupObject

Returns:

outputSpectralTimeModel or SpectralGroupModel

The requested selection of results data loaded into a new model object.

get_metrics(category, measure=None)

Get requested metric(s) from the object.

Parameters:

Name of the specific measure(s) to return.

Returns:

metricsdict

Dictionary of requested metrics.

get_model(ind=None, regenerate=True)

Get a model fit object for a specified index.

Parameters:

indint, optional

The index of the model from group_results to access. If None, return a Model object with initialized settings, with no data or results.

regeneratebool, optional, default: False

Whether to regenerate the model fits for the requested model.

Returns:

modelSpectralModel

The data and fit results loaded into a model object.

get_params(component, field=None)

Return model fit parameters for specified feature(s).

Parameters:

fieldstr or int, optional

Column name / index to extract from selected data, if requested. If str, should align with a parameter label for the component fit mode.

Returns:

outndarray

Requested data.

Raises:

NoModelError

If there are no model fit results available.

ValueError

If the input for the field input is not understood.

Notes

When extracting peak parameters, an additional column is appended to the returned array, indicating the index that the peak came from.

load(file_name, file_path=None, convert_results=True)

Load time data from file.

Parameters:

file_namestr

File to load data from.

file_pathstr, optional

Path to directory to load from. If None, loads from current directory.

convert_resultsbool, optional, default: True

Whether to convert results to be organized over time.

plot(plot_type='time', save_fig=False, file_name=None, file_path=None, **plot_kwargs)

Plot a figure with subplots visualizing the parameters from a SpectralTimeModel object.

Parameters:

**plot_kwargs

Keyword arguments to apply to the plot.

Raises:

NoModelError

If the model object does not have model fit data available to plot.

static print_report_issue(concise=False)

Prints instructions on how to report bugs and/or problematic fits.

Parameters:

concisebool, optional, default: False

Whether to print the report in a concise mode, or not.

print_results(print_type='time', concise=False)

Print out SpectralTimeModel results.

Parameters:

print_type{‘time’, ‘group’}

Which format to print results out in.

concisebool, optional, default: False

Whether to print the report in a concise mode, or not.

print_settings(description=False, concise=False)

Print out the current settings.

Parameters:

descriptionbool, optional, default: False

Whether to print out a description with current settings.

concisebool, optional, default: False

Whether to print the report in a concise mode, or not.

report(freqs=None, spectrogram=None, freq_range=None, bands=None, report_type='time', n_jobs=1, progress=None)

Fit a spectrogram and display a report, with a plot and printed results.

Parameters:

freqs1d array, optional

Frequency values for the spectrogram, in linear space.

spectrogram2d array, shape: [n_freqs, n_time_windows], optional

Spectrogram of power spectrum values, in linear space.

freq_rangelist of [float, float], optional

Frequency range to fit the model to. If not provided, fits the entire given range.

bandsBands or dict or int, optional

How to organize peaks into bands. If Bands or dict, uses band definitions. If int, extracts the first ‘n’ peaks.

n_jobsint, optional, default: 1

Number of jobs to run in parallel. 1 is no parallelization. -1 uses all available cores.

progress{None, ‘tqdm’, ‘tqdm.notebook’}, optional

Which kind of progress bar to use. If None, no progress bar is used.

Notes

Data is optional, if data has already been added to the object.

save(file_name, file_path=None, append=False, save_results=False, save_settings=False, save_data=False)

Save out results and/or settings from time object. Saves out to a JSON file.

Parameters:

file_namestr or FileObject

File to save data to.

file_pathPath or str, optional

Path to directory to load from. If None, saves to current directory.

appendbool, optional, default: False

Whether to append to an existing file, if available. This option is only valid (and only used) if ‘file_name’ is a str.

save_resultsbool, optional

Whether to save out model fit results.

save_settingsbool, optional

Whether to save out settings.

save_databool, optional

Whether to save out power spectra data.

save_model_report(index, file_name, file_path=None, add_settings=True, **plot_kwargs)

“Save out an individual model report for a specified model fit.

Parameters:

indexint

Index of the model fit to save out.

file_namestr

Name to give the saved out file.

file_pathPath or str, optional

Path to directory to save to. If None, saves to current directory.

add_settingsbool, optional, default: True

Whether to add a print out of the model settings to the end of the report.

plot_kwargskeyword arguments

Keyword arguments to pass into the plot method.

save_report(file_name, file_path=None, add_settings=True)

Generate and save out a PDF report for models of a spectrogram.

Parameters:

file_namestr

Name to give the saved out file.

file_pathstr, optional

Path to directory to save to. If None, saves to current directory.

add_settingsbool, optional, default: True

Whether to add a print out of the model settings to the end of the report.

to_df(bands=None)

Convert and extract the model results as a pandas object.

Parameters:

bandsBands or dict or int, optional

How to organize peaks into bands. If Bands or dict, uses band definitions. If int, extracts the first ‘n’ peaks. If provided, re-extracts peak features; if not, converts from time_results.

Returns:

pd.DataFrame

Model results organized into a pandas object.

Examples using specparam.SpectralTimeModel

09: Fitting Models over Time

09: Fitting Models over Time