rsatoolbox.inference.crossvalsets module

generation of crossvalidation splits

rsatoolbox.inference.crossvalsets.sets_k_fold(rdms, k_rdm=None, k_pattern=None, random=True, pattern_descriptor='index', rdm_descriptor='index')[source]

generates training and test set combinations by splitting into k similar sized groups. This version splits both over rdms and over patterns resulting in k_rdm * k_pattern (training, test) pairs.

If a k is set to 1 the corresponding dimension is not crossvalidated.

Parameters
  • rdms (rsatoolbox.rdm.RDMs) – rdms to use

  • pattern_descriptor (String) – descriptor to select pattern groups

  • rdm_descriptor (String) – descriptor to select rdm groups

  • k_rdm (int) – number of rdm groups

  • k_pattern (int) – number of pattern groups

  • random (bool) – whether the assignment shall be randomized

Returns

list of tuples (rdms, pattern_idx) test_set(list): list of tuples (rdms, pattern_idx) ceil_set(list): list of tuples (rdms, pattern_idx)

Return type

train_set(list)

rsatoolbox.inference.crossvalsets.sets_k_fold_pattern(rdms, pattern_descriptor='index', k=None, random=False)[source]

generates training and test set combinations by splitting into k similar sized groups. This version splits in the given order or randomizes the order. For k=1 training and test_set are whole dataset, i.e. no crossvalidation is performed.

For only crossvalidating over patterns there is no independent training set for calculating a noise ceiling for the patterns. To express this we set ceil_set to None, which makes the crossvalidation function calculate a leave one rdm out noise ceiling for the right patterns instead.

Parameters
  • rdms (rsatoolbox.rdm.RDMs) – rdms to use

  • pattern_descriptor (String) – descriptor to select groups

  • k (int) – number of groups

  • random (bool) – whether the assignment shall be randomized

Returns

list of tuples (rdms, pattern_idx) test_set(list): list of tuples (rdms, pattern_idx) ceil_set = None

Return type

train_set(list)

rsatoolbox.inference.crossvalsets.sets_k_fold_rdm(rdms, k_rdm=None, random=True, rdm_descriptor='index')[source]

generates training and test set combinations by splitting into k similar sized groups. This version splits both over rdms and over patterns resulting in k_rdm * k_pattern (training, test) pairs.

Parameters
  • rdms (rsatoolbox.rdm.RDMs) – rdms to use

  • rdm_descriptor (String) – descriptor to select rdm groups

  • k_rdm (int) – number of rdm groups

  • random (bool) – whether the assignment shall be randomized

Returns

list of tuples (rdms, pattern_idx) test_set(list): list of tuples (rdms, pattern_idx)

Return type

train_set(list)

rsatoolbox.inference.crossvalsets.sets_leave_one_out_pattern(rdms, pattern_descriptor)[source]

generates training and test set combinations by leaving one level of pattern_descriptor out as a test set. This is only sensible if pattern_descriptor already defines larger groups!

the ceil_train_set contains the rdms for the test-patterns from the training-rdms. This is required for computing the noise-ceiling

Parameters
  • rdms (rsatoolbox.rdm.RDMs) – rdms to use

  • pattern_descriptor (String) – descriptor to select groups

Returns

list of tuples (rdms, pattern_idx) test_set(list): list of tuples (rdms, pattern_idx) ceil_set(list): list of tuples (rdms, pattern_idx)

Return type

train_set(list)

rsatoolbox.inference.crossvalsets.sets_leave_one_out_rdm(rdms, rdm_descriptor='index')[source]

generates training and test set combinations by leaving one level of rdm_descriptor out as a test set. :param rdms: rdms to use :type rdms: rsatoolbox.rdm.RDMs :param rdm_descriptor: descriptor to select groups :type rdm_descriptor: String

Returns

list of tuples (rdms, pattern_idx) test_set(list): list of tuples (rdms, pattern_idx) ceil_set(list): list of tuples (rdms, pattern_idx)

Return type

train_set(list)

rsatoolbox.inference.crossvalsets.sets_of_k_pattern(rdms, pattern_descriptor=None, k=5, random=False)[source]

generates training and test set combinations by splitting into groups of k. This version splits in the given order or randomizes the order. If the number of patterns is not divisible by k patterns are added to the first groups such that those have k+1 patterns

Parameters
  • rdms (rsatoolbox.rdm.RDMs) – rdms to use

  • pattern_descriptor (String) – descriptor to select groups

  • k (int) – number of groups

  • random (bool) – whether the assignment shall be randomized

Returns

list of tuples (rdms, pattern_idx) test_set(list): list of tuples (rdms, pattern_idx)

Return type

train_set(list)

rsatoolbox.inference.crossvalsets.sets_of_k_rdm(rdms, rdm_descriptor='index', k=5, random=False)[source]

generates training and test set combinations by splitting into groups of k. This version splits in the given order or randomizes the order. If the number of patterns is not divisible by k patterns are added to the first groups such that those have k+1 patterns

Parameters
  • rdms (rsatoolbox.rdm.RDMs) – rdms to use

  • pattern_descriptor (String) – descriptor to select groups

  • k (int) – number of groups

  • random (bool) – whether the assignment shall be randomized

Returns

list of tuples (rdms, pattern_idx) test_set(list): list of tuples (rdms, pattern_idx) ceil_set(list): list of tuples (rdms, pattern_idx)

Return type

train_set(list)

rsatoolbox.inference.crossvalsets.sets_random(rdms, n_rdm=None, n_pattern=None, n_cv=2, pattern_descriptor='index', rdm_descriptor='index')[source]

generates training and test set combinations by selecting random test sets of n_rdm RDMs and n_pattern patterns and using the rest of the data as the training set.

If a n is set to 0 the corresponding dimension is not crossvalidated.

Parameters
  • rdms (rsatoolbox.rdm.RDMs) – rdms to split

  • pattern_descriptor (String) – descriptor to select pattern groups

  • rdm_descriptor (String) – descriptor to select rdm groups

  • n_rdm (int) – number of rdms per test set

  • n_pattern (int) – number of patterns per test set

Returns

list of tuples (rdms, pattern_idx) test_set(list): list of tuples (rdms, pattern_idx) ceil_set(list): list of tuples (rdms, pattern_idx)

Return type

train_set(list)