rsatoolbox.util.searchlight module

This code was initially inspired by the following : https://github.com/machow/pysearchlight

@author: Daniel Lindh

rsatoolbox.util.searchlight.evaluate_models_searchlight(sl_RDM, models, eval_function, method='corr', theta=None, n_jobs=1)[source]

evaluates each searchlighth with the given model/models

Parameters:

  • sl_RDM ([rsatoolbox.rdm.RDMs]) – RDMs object

  • rsatoolbox.util.searchlight.get_searchlight_RDMs (as computed by)

  • ([rsatoolbox.model] (models) – models to evaluate - can also be list of models

  • eval_function (rsatoolbox.inference evaluation-function) – [description]

  • method (str, optional) – see rsatoolbox.rdm.compare for specifics. Defaults to ‘corr’.

  • n_jobs (int, optional) – how many jobs to run. Defaults to 1.

Returns:

list of with the model evaluation for each searchlight center

Return type:

list

rsatoolbox.util.searchlight.get_searchlight_RDMs(data_2d, centers, neighbors, events, method='correlation', verbose=True)[source]

Iterates over all the searchlight centers and calculates the RDM

Parameters:

  • data_2d (2D numpy array) – brain data,

  • n_channels (shape n_observations x)

  • centers (1D numpy array) – center indices for all searchlights as provided

  • rsatoolbox.util.searchlight.get_volume_searchlight (as provided by)

  • neighbors (list) – list of lists with neighbor voxel indices for all searchlights

  • rsatoolbox.util.searchlight.get_volume_searchlight

  • events (1D numpy array) – 1D array of length n_observations

  • method (str, optional) – distance metric,

  • 'correlation'. (see rsatoolbox.rdm.calc for options. Defaults to)

  • verbose (bool, optional) – Defaults to True.

Returns:

RDMs object with the RDM for each searchlight

the RDM.rdm_descriptors[‘voxel_index’] describes the center voxel index each RDM is associated with

Return type:

RDM [rsatoolbox.rdm.RDMs]

rsatoolbox.util.searchlight.get_volume_searchlight(mask, radius=2, threshold=1.0, truncate_at_boundary=False)[source]

Searches through the non-zero voxels of the mask, selects centers where proportion of sphere voxels >= self.threshold.

Parameters:

  • mask ([numpy array]) – binary brain mask

  • radius (int, optional) – the radius of each searchlight, defined in voxels.

  • 2. (Defaults to)

  • threshold (float, optional) – Threshold of the proportion of voxels that need to

  • be (be inside the brain mask in order for it to)

  • center. (considered a good searchlight)

  • that (Values go between 0.0 - 1.0 where 1.0 means)

  • inside (100% of the voxels need to be)

  • mask. (the brain)

  • 1.0. (Defaults to)

  • truncate_at_boundary (bool, optional) – if True, searchlight spheres will only

  • True (include voxels where the mask is)

  • at (effectively truncating spheres)

  • False (the mask boundary. if)

  • value (the radius regardless of mask)

  • 1.0 (When False and threshold <)

  • in (this can lead to artifacts as reported)

  • for (issue #466. Setting to True fixes this but may require accounting)

  • analysis. (different variance characteristics in second-level)

  • False. (Defaults to)

Returns:

array of centers of size n_centers x 3

list: list of lists with neighbors - the length of the list will correspond to: n_centers x 3 x n_neighbors

Return type:

numpy array