Creat help text conversion tool

[teddychao]

Changes proposed in this pull request:

• Create pspm_doc.m

[teddychao]

Hi @dominikbach . I provide a template of the markdown file from help text for you to have a look. I will follow this format once you approve it. Thanks

Template based on pspm_dcm.

pspm_dcm

Description

pspm_dcm sets up a non-linear SCR model, prepares and normalises the data, passes it over to the model inversion routine, and saves both the forward model and its inversion. Non-linear SCR models are required if response timing is not known and has to be estimated from data. A typical example are anticipatory SCR in fear conditioning. These occur at some point between CS and US, but this time point is not known. Both flexible-latency (within a response window) and fixed-latency (evoked after a specified event) responses can be modelled. For fixed responses, delay and dispersion are assumed to be constant (either pre-determined or estimated from the data), while for flexible responses, both are estimated for each individual trial. Flexible responses can for example be anticipatory, decision-related, or evoked with unknown onset. PsPM implements an iterative trial-by-trial algorithm. Different from GLM, response parameters are always estimated per trial, and the algorithm is not informed about the condition. For each session, experimental timing is defined by providing a 1-column vector of event onsets in seconds for each fixed event, and a 2-column matrix for each flexible event. Each event must occur in each trial of a session, i.e. all these vectors and matrices must have the same number of rows. (For example, in fear conditioning where the US occurs only on a subset of trials, each trial includes an event "US onset" even if it does not occur, to avoid bias). A timing file should contain a variable 'events' which is a cell array; each cell should contain either a one-column vector or a 2-column matrix.

Format

[sts, dcm] = pspm_dcm(model, options)

Arguments

• model

  • modelfile: [string/cell array] The name of the model output file.
  • datafile: [string/cell array] A file name (single session) OR a cell array of file names.
  • timing: A file name/cell array of events (single session) OR a cell array of file names/cell arrays. When specifying file names, each file must be a .mat file that contain a cell variable called 'events'. Each cell should contain either one column (fixed response) or two columns (flexible response). All matrices in the array need to have the same number of rows, i.e. the event structure should be the same for every trial. For trials that are not going to be analysed later, it is possible to include dummy events with negative onsets. All event timings must be specified in SECONDS.
  • missing: [optional] Allows to specify missing (e.g. artefact) epochs in the data file. See pspm_get_timing for epoch definition; specify a cell array for multiple input files. This must always be specified in SECONDS. Default: no missing values.
  • lasttrialcutoff: [optional] If there fewer data after the end of the last trial in a session than this cutoff value (in s), then estimated parameters from this trial will be assumed inestimable and set to NaN after the inversion. This value can be set as inf to always retain parameters from the last trial. Default: 7 s, corresponding to the time at which the canonical SCRF has decayed to around 80% of its peak value.
  • substhresh:[optional] Maximum duration (in seconds) of missing data periods allowed within a session (these data points will be ignored). For missing data periods longer than this threshold, the algorithm will split up the data into subsessions which are evaluated independently (excluding NaN values). Default: 2 s.
  • filter: [optional] Filter settings. Modality specific default.
  • channel: [optional] Channel number. Default: last SCR channel
  • norm: [optional] Normalise data. i.e. Data are normalised during inversion but results transformed back into raw data units. Default: 0.
  • constrained: [optional] Constrained model for flexible responses which have fixed dispersion (0.3 s SD) but flexible latency.

• options

  • .crfupdate: [0/1] Re-estimate RF parameters from canonical SCRF, or use pre-estimated RF parameters. This can be used when f_SCR has been changed.
  • .indrf: Estimate the response function from the data. This is only recommended for long inter-trial-intervals and should be used with caution. In reference 2, this option lead to worse quality of the trial-by-trial amplitude estimation (potenetially due to overfitting the data available to estimate the response function). Default: 0.
  • .getrf: Only estimate response function, do not do trial-wise DCM.
  • .rf: Call an external file to provide response function (for use when this is previously estimated by pspm_get_rf).
  • .depth: Number of trials to invert at the same time. The iterative estimation will progress trial-by-trial and consider this number of trials into the future, until the last trial of a session. If this parameter is larger than the number of trials in a session, the entire sessin will be inverted at the same time. In reference 2, this parameter (set to 2 or 3) had no impact on the quality of the estimation. Unpublished data suggest that if a session with 24 trials and two events per trial is estimated in one go, then the quality of the estimation suffers (potentially because in the larger parameter landscape, it is more difficult to find the global minimum). Default: 2.
  • .sfpre: SF-free interval before first event of a trial. Default: 2 s.
  • .sfpost: SF-free interval after last event of a trial. Default: 5 s.
  • .sffreq: Maximum frequency of SF in ITIs. Default: 0.5/s.
  • .sclpre: SCR-change-free interval before first event of a trial. Default: 2 s.
  • .sclpost: SCR-change-free interval after last event of a trial. Default: 5 s.
  • .aSCR_sigma_offset: Minimum dispersion (standard deviation) for flexible responses, in seconds. Default: 0.1 s.
  • .dispwin: [0/1] Display progress plot. Default: display.
  • .dispsmallwin: [0/1] Display intermediate progress plots. Default: no display.
  • .nosave: Don't save dcm structure (e.g. used by pspm_get_rf)
  • .overwrite: [0/1] Define whether to overwrite existing output files or not. Default value: determined by pspm_overwrite.
  • .trlnames: Cell array of names for individual trials. This is only for housekeeping (e.g. condition descriptions), not for model estimation. Default: no trial names.
  • .eventnames: Cell array of names for individual events, in the order they are specified in the model.timing array - to be used for display and export only

    Output

• fn: Name of the model file.

• dcm: Model struct. Output units: all timeunits are in seconds; eSCR and aSCR amplitude are in SN units such that an eSCR SN pulse with 1 unit amplitude causes an eSCR with 1 mcS amplitude.

Developer's Notes

1. pspm_dcm can handle NaN values in data channels. Either by specifying missing epochs manually using model.missing, or by detecting NaN epochs in the data. Missing epochs shorter than model.substhresh will be ignored in the inversion; otherwise the data will be split into subsessions that are inverted independently. The results will be unchanged, and events within missing epochs will simply be set to NaN. NaN periods shorter than model.substhresh are interpolated for averages and principal response components.
2. pspm_dcm calculates the inter-trial intervals as the duration between the end of a trial and the start of the next one. ITI value for the last trial in a session is calculated as the duration between the end of the last trial and the end of the whole session. Since this value may differ significantly from the regular ITI duration values, it is not used when computing the minimum ITI duration of a session.

Minimum of session specific min ITI values is used

1. when computing mean SCR signal
2. when computing the PCA from all the trials in all the sessions.

In case of case (2), after each trial, all the samples in the period with duration equal to the just mentioned overall min ITI value is used as a row of the input matrix. Since this minimum does not use the min ITI value of the last trial in each session, the sample period may be longer than the ITI value of the last trial. In such a case, pspm_dcm is not able to compute the PCA and emits a warning.

The rationale behind this behaviour is that we observed that ITI value of the last trial in a session might be much smaller than the usual ITI values. For example, this can happen when a long missing data section starts very soon after the beginning of a trial. If this very small ITI value is used to define the sample periods after each trial, nearly all the trials use much less than available amount of samples in both case (1) and (2). Instead, we aim to use as much data as possible in (1), and perform (2) only if this edge case is not present.

References

1. Model development: Bach DR, Daunizeau J, Friston KJ, Dolan RJ (2010). Dynamic causal modelling of anticipatory skin conductance changes. Biological Psychology, 85(1), 163-70
2. Model validation and improvement: Staib, M., Castegnetti, G., & Bach, D. R. (2015). Optimising a model-based approach to inferring fear learning from skin conductance responses. Journal of Neuroscience Methods, 255, 131-138.

[teddychao]

Alternative version of Arguments

Variable	Field	Description
model	modelfile	[string/cell array] The name of the model output file.
model	datafile	[string/cell array] A file name (single session) OR a cell array of file names.
model	timing	A file name/cell array of events (single session) OR a cell array of file names/cell arrays. When specifying file names, each file must be a .mat file that contain a cell variable called 'events'. Each cell should contain either one column (fixed response) or two columns (flexible response). All matrices in the array need to have the same number of rows, i.e. the event structure should be the same for every trial. For trials that are not going to be analysed later, it is possible to include dummy events with negative onsets. All event timings must be specified in SECONDS.
model	missing	[optional] Allows to specify missing (e.g. artefact) epochs in the data file. See ppm_get_timing	for epoch definition; specify a cell array for multiple input files. This must always be specified in SECONDS. Default: no missing values.
model	lasttrialcutoff	[optional] If there fewer data after the end of the last trial in a session than this cutoff value (in s), then estimated parameters from this trial will be assumed inestimable and set to NaN after the inversion. This value can be set as inf to always retain parameters from the last trial. Default: 7 s, corresponding to the time at which the canonical SCRF has decayed to around 80% of its peak value.
model	substhresh	[optional] Maximum duration (in seconds) of missing data periods allowed within a session (these data points will be ignored). For missing data periods longer than this threshold, the algorithm will split up the data into subsessions which are evaluated independently (excluding NaN values). Default: 2 s.
model	filter	[optional] Filter settings. Modality specific default.
model	channel	[optional] Channel number. Default: last SCR channel.
model	norm	[optional] Normalise data. i.e. Data are normalised during inversion but results transformed back into raw data units. Default: 0.
model	constrained	[optional] Constrained model for flexible responses, which have fixed dispersion (0.3 s SD) but flexible latency.
options
options	crfupdate	[0/1] Re-estimate RF parameters from canonical SCRF, or use pre-estimated RF parameters. This can be used when f_SCR has been changed.
options	indrf	Estimate the response function from the data. This is only recommended for long inter-trial-intervals and should be used with caution. In reference 2, this option lead to worse quality of the trial-by-trial amplitude estimation (potenetially due to overfitting the data available to estimate the response function). Default: 0.
options	getrf	Only estimate response function, do not do trial-wise DCM.
options	rf	Call an external file to provide response function (for use when this is previously estimated by pspm_get_rf).
options	depth	Number of trials to invert at the same time. The iterative estimation will progress trial-by-trial and consider this number of trials into the future, until the last trial of a session. If this parameter is larger than the number of trials in a session, the entire sessin will be inverted at the same time. In reference 2, this parameter (set to 2 or 3) had no impact on the quality of the estimation. Unpublished data suggest that if a session with 24 trials and two events per trial is estimated in one go, then the quality of the estimation suffers (potentially because in the larger parameter landscape, it is more difficult to find the global minimum). Default: 2.
options	sfpre	SF-free interval before first event of a trial. Default: 2 s.
options	sfpost	SF-free interval after last event of a trial. Default: 5 s.
options	sffreq	Maximum frequency of SF in ITIs. Default: 0.5/s.
options	sclpre	SCR-change-free interval before first event of a trial. Default: 2 s.
options	sclpost	SCR-change-free interval after last event of a trial. Default: 5 s.
options	aSCR_sigma_offset	Minimum dispersion (standard deviation) for flexible responses, in seconds. Default: 0.1 s.
options	dispwin	[0/1] Display progress plot. Default: display.
options	dispsmallwin	[0/1] Display intermediate progress plots. Default: no display.
options	nosave	Don't save dcm structure (e.g. used by pspm_get_rf)
options	overwrite	[0/1] Define whether to overwrite existing output files or not. Default value	determined by pspm_overwrite.
options	trlnames	Cell array of names for individual trials. This is only for housekeeping (e.g. condition descriptions), not for model estimation. Default : no trial names.
options	eventnames	Cell array of names for individual events, in the order they are specified in the model.timing array - to be used for display and export only

[dominikbach]

The first version looks nice to me - but let's get more feedback. In either case, remove "Developer's notes".

[teddychao]

Here are some nice examples

pspm_eye

pspm_dcm

pspm_merge

pspm_glm

[teddychao]

Hi @dominikbach Here are two questions

1. Shall I have a website or something so I can write a template of CSS there and reflect the markdown document to HTML files?
2. Would it be useful to write a batch processing function (similar to pspm_test) to convert all the functions' help text into markdown files?

Thanks

[teddychao]

Function needs to support second level deeper tree of argument description. Can be updated after approving PR #766 .

[teddychao]

The two previously unsupported functions have been supported now

[teddychao]

Multiple lines of format are also well supported now