syncopy.connectivity.AV_compRoutines.GrangerCausality#

class syncopy.connectivity.AV_compRoutines.GrangerCausality(*argv, **kwargs)[source]#

Compute class that computes pairwise Granger causalities of CrossSpectralData objects.

Sub-class of ComputationalRoutine, see Design Guide: Syncopy Compute Classes for technical details on Syncopy’s compute classes and metafunctions.

See also

syncopy.connectivityanalysis

parent metafunction

__init__(*argv, **kwargs)#

Instantiate a ComputationalRoutine subclass

Parameters:
Returns:

obj – Usable class instance for processing Syncopy data objects.

Return type:

instance of ComputationalRoutine-subclass

Methods

__init__(*argv, **kwargs)

Instantiate a ComputationalRoutine subclass

compute(data, out[, parallel, ...])

Central management and processing method

computeFunction(csd_av_dat[, rtol, nIter, ...])

Given the trial averaged cross spectral densities, calculates the pairwise Granger-Geweke causalities for all (non-symmetric!) channel combinations following the algorithm proposed in [1].

compute_parallel(data, out)

Concurrent computing kernel

compute_sequential(data, out)

Sequential computing kernel

initialize(data, out_stackingdim[, ...])

Perform dry-run of calculation to determine output shape

pre_check()

Make sure we have a trial average, so the input data only consists of 1 trial.

preallocate_output(out[, parallel_store])

Storage allocation and provisioning

process_metadata(data, out)

Meta-information manager

write_log(data, out[, log_dict])

Processing of output log

Attributes

dimord

metadata_keys

The keys available in the info property of the returned data instance.

method

valid_kws

metadata_keys = ('converged', 'max rel. err', 'reg. factor', 'initial cond. num')#

The keys available in the info property of the returned data instance.

‘converged’ : bool, True if the algoritm converged successfully

‘max rel, err’ : float, maximum relative error between the input CSD and the spectral factorization

‘reg. factor’ : float, brute force regularization factor in case the CSD is nearly singular

‘initial cond. num’ : float, condition number of the CSD, regularization kicks in if that is too high

dimord = ['time', 'freq', 'channel_i', 'channel_j']#
static computeFunction(csd_av_dat, rtol=5e-06, nIter=100, cond_max=10000.0, chunkShape=None, noCompute=False)#

Given the trial averaged cross spectral densities, calculates the pairwise Granger-Geweke causalities for all (non-symmetric!) channel combinations following the algorithm proposed in [1].

First the CSD matrix is factorized using Wilson’s algorithm, the resulting transfer functions and noise covariance matrix is then used to calculate Granger causality according to Eq. 8 in [1].

Selection of channels and frequencies of interest can and should be done beforehand when calculating the CSDs.

Critical numerical parameters for Wilson’s algorithm (rtol, nIter, cond_max) have sensitive defaults, which were tested for datasets with up to 5000 samples and 256 channels. Changing them is recommended for expert users only.

Parameters:

  • csd_av_dat ((1, nFreq, N, N) numpy.ndarray) – Cross-spectral densities for N x N channels and nFreq frequencies averaged over trials.

  • rtol (float) – Relative error tolerance for Wilson’s algorithm for spectral matrix factorization. Default should be fine for most cases, handle with care!

  • nIter (int) – Maximum number of iterations for CSD factorization. A result is returned if exhausted also if error tolerance was not met.

  • cond_max (float) – The maximal condition number of the spectral matrix. The CSD matrix can be almost singular in cases of many channels and low sample number. In these cases Wilson’s factorization fails to converge, as it relies on positive definiteness of the CSD matrix. If the condition number is above cond_max, a brute force regularization is performed until the regularized CSD matrix has a condition number below cond_max.

  • noCompute (bool) – Preprocessing flag. If True, do not perform actual calculation but instead return expected shape and numpy.dtype of output array.

Returns:

Granger – Spectral Granger-Geweke causality between all channel combinations. Directionality follows array notation: causality from i -> j is Granger[0,:,i,j], causality from j -> i is Granger[0,:,j,i]

Return type:

(1, nFreq, N, N) numpy.ndarray

Notes

This method is intended to be used as computeFunction() inside a ComputationalRoutine. Thus, input parameters are presumed to be forwarded from a parent metafunction. Consequently, this function does not perform any error checking and operates under the assumption that all inputs have been externally validated and cross-checked.

See also

cross_spectra_cF

cross_spectra_cF() Single trial (Multi-)tapered cross spectral densities. Trial averages can be obtained by calling the respective computational routine with keeptrials=False.

wilson_sf

:func:`~syncopy.connectivity.wilson_sf.wilson_sf Spectral matrix factorization that yields the transfer functions and noise covariances from a cross spectral density.

regularize_csd

:func:`~syncopy.connectivity.wilson_sf.regularize_csd Brute force regularization scheme for the CSD matrix

granger

:func:`~syncopy.connectivity.granger.granger Given the results of the spectral matrix factorization, calculates the granger causalities

method = ''#
valid_kws = ['rtol', 'nIter', 'cond_max', 'chunkShape', 'noCompute']#
pre_check()[source]#

Make sure we have a trial average, so the input data only consists of 1 trial. Can only be performed after initialization!

process_metadata(data, out)[source]#

Meta-information manager

Parameters:

  • data (syncopy data object) – Syncopy data object that has been processed

  • out (syncopy data object) – Syncopy data object holding calculation results

Returns:

Nothing

Return type:

None

Notes

This routine is an abstract method and is thus intended to be overloaded. Consult the developer documentation (Design Guide: Syncopy Compute Classes) for further details.

See also

write_log

Logging of calculation parameters

__init__(*argv, **kwargs)#

Instantiate a ComputationalRoutine subclass

Parameters:
Returns:

obj – Usable class instance for processing Syncopy data objects.

Return type:

instance of ComputationalRoutine-subclass

compute(data, out, parallel=False, parallel_store=None, method=None, mem_thresh=0.5, log_dict=None, parallel_debug=False)#

Central management and processing method

Parameters:

  • data (syncopy data object) – Syncopy data object to be processed (has to be the same object that was used by initialize() in the pre-calculation dry-run).

  • out (syncopy data object) – Empty object for holding results

  • parallel (bool) – If True, processing is performed in parallel (i.e., computeFunction() is executed concurrently across trials). If parallel is False, computeFunction() is executed consecutively trial after trial (i.e., the calculation realized in computeFunction() is performed sequentially).

  • parallel_store (None or bool) – Flag controlling saving mechanism. If None, parallel_store = parallel, i.e., the compute-paradigm dictates the employed writing method. Thus, in case of parallel processing, results are written in a fully concurrent manner (each worker saves its own local result segment on disk as soon as it is done with its part of the computation). If parallel_store is False and parallel is True the processing result is saved sequentially using a mutex. If both parallel and parallel_store are False standard single-process HDF5 writing is employed for saving the result of the (sequential) computation.

  • method (None or str) – If None the predefined methods compute_parallel() or compute_sequential() are used to control the actual computation (specifically, calling computeFunction()) depending on whether parallel is True or False, respectively. If method is a string, it has to specify the name of an alternative (provided) class method (starting with the word “compute_”).

  • mem_thresh (float) – Fraction of available memory required to perform computation. By default, the largest single trial result must not occupy more than 50% (mem_thresh = 0.5) of available single-machine or worker memory (if parallel is False or True, respectively).

  • log_dict (None or dict) – If None, the log properties of out is populated with the employed keyword arguments used in computeFunction(). Otherwise, out’s log properties are filled with items taken from log_dict.

  • parallel_debug (bool) – If True, concurrent processing is performed using a single-threaded scheduler, i.e., all parallel computing task are run in the current Python thread permitting usage of tools like pdb/ipdb, cProfile and the like in computeFunction(). Note that enabling parallel debugging effectively runs the given computation on the calling machine locally thereby requiring sufficient memory and CPU capacity.

Returns:

Nothing – The result of the computation is available in out once compute() terminated successfully.

Return type:

None

Notes

This routine calls several other class methods to perform all necessary pre- and post-processing steps in a fully automatic manner without requiring any user-input. Specifically, the following class methods are invoked consecutively (in the given order):

  1. preallocate_output() allocates a (virtual) HDF5 dataset of appropriate dimension for storing the result

  2. compute_parallel() (or compute_sequential()) performs the actual computation via concurrently (or sequentially) calling computeFunction()

  3. process_metadata() attaches all relevant meta-information to the result out after successful termination of the calculation

  4. write_log() stores employed input arguments in out.cfg and out.log to reproduce all relevant computational steps that generated out.

Parallel processing setup and input sanitization is performed by ACME. Specifically, all necessary information for parallel execution and storage of results is propagated to the ParallelMap context manager.

See also

initialize

pre-calculation preparations

preallocate_output

storage provisioning

compute_parallel

concurrent computation using computeFunction()

compute_sequential

sequential computation using computeFunction()

process_metadata

management of meta-information

write_log

log-entry organization

acme.ParallelMap

concurrent execution of Python callables

compute_parallel(data, out)#

Concurrent computing kernel

Parameters:

  • data (syncopy data object) – Syncopy data object, ignored for parallel case. The input data information is already contained in the workerdicts, which are stored in self.pmap (expanded from the inargs in compute()) and can be accessed from it.

  • out (syncopy data object) – Empty object for holding results

Returns:

Nothing

Return type:

None

Notes

The actual reading of source data and writing of results is managed by the decorator syncopy.shared.kwarg_decorators.process_io().

See also

compute

management routine invoking parallel/sequential compute kernels

compute_sequential

serial processing counterpart of this method

compute_sequential(data, out)#

Sequential computing kernel

Parameters:

  • data (syncopy data object) – Syncopy data object to be processed

  • out (syncopy data object) – Empty object for holding results

Returns:

Nothing

Return type:

None

Notes

This method most closely reflects classic iterative process execution: trials in data are passed sequentially to computeFunction(), results are stored consecutively in a regular HDF5 dataset (that was pre-allocated by preallocate_output()). Since the calculation result is immediately stored on disk, propagation of arrays across routines is avoided and memory usage is kept to a minimum.

See also

compute

management routine invoking parallel/sequential compute kernels

compute_parallel

concurrent processing counterpart of this method

initialize(data, out_stackingdim, chan_per_worker=None, keeptrials=True)#

Perform dry-run of calculation to determine output shape

Parameters:

  • data (syncopy data object) – Syncopy data object to be processed (has to be the same object that is passed to compute() for the actual calculation).

  • out_stackingdim (int) – Index of data dimension for stacking trials in output object

  • chan_per_worker (None or int) – Number of channels to be processed by each worker (only relevant in case of concurrent processing). If chan_per_worker is None (default) by-trial parallelism is used, i.e., each worker processes data corresponding to a full trial. If chan_per_worker > 0, trials are split into channel-groups of size chan_per_worker (+ rest if the number of channels is not divisible by chan_per_worker without remainder) and workers are assigned by-trial channel-groups for processing.

  • keeptrials (bool) – Flag indicating whether to return individual trials or average

Returns:

Nothing

Return type:

None

Notes

This class method has to be called prior to performing the actual computation realized in computeFunction().

See also

compute

core routine performing the actual computation

preallocate_output(out, parallel_store=False)#

Storage allocation and provisioning

Parameters:

  • out (syncopy data object) – Empty object for holding results

  • parallel_store (bool) – If True, a directory for virtual source files is created in Syncopy’s temporary on-disk storage (defined by syncopy.__storage__). Otherwise, a dataset of appropriate type and shape is allocated in a new regular HDF5 file created inside Syncopy’s temporary storage folder.

Returns:

Nothing

Return type:

None

See also

compute

management routine controlling memory pre-allocation

write_log(data, out, log_dict=None)#

Processing of output log

Parameters:

  • data (syncopy data object) – Syncopy data object that has been processed

  • out (syncopy data object) – Syncopy data object holding calculation results

  • log_dict (None or dict) – If None, the log properties of out is populated with the employed keyword arguments used in computeFunction(). Otherwise, out’s log properties are filled with items taken from log_dict.

Returns:

Nothing

Return type:

None

See also

process_metadata

Management of meta-information