CCA_align_by_class

CCA_align_by_class(X_a, X_b, y_a, y_b, return_space='b_to_a')[source][source]

CCA Alignment between 2 datasets with correspondence by averaging within class conditions.

The number of features must be the same for datasets A and B. For example, if the datasets have different feature sizes, you can use PCA to reduce both datasets to the same number of PCs first.

Parameters:

  • X_a (ndarray) – Data matrix for dataset A of shape (n_trials_a, n_timepoints, n_features)

  • X_b (ndarray) – Data matrix for dataset B of shape (n_trials_b, n_timepoints, n_features)

  • y_a (ndarray) – Label matrix for dataset A of shape (n_trials_a, …). The first dimension must be the trial dimension. This can be a 1D array, or a 2D array if each trial has multiple labels (e.g. a sequence of phonemes). Label sequences are converted to a single string so that only the same label sequences have correspondence between the datasets.

  • y_b (ndarray) – Label matrix for dataset B of shape (n_trials_a, …). See y_a for more details.

  • return_space (str, optional) – How to perform alignment. Dataset B can be aligned to A, and vice versa (‘b_to_a’ and ‘a_to_b’, respectively), or both datasets can be aligned to a shared space (‘shared’). Defaults to ‘b_to_a’.

Returns:

Tuple containing aligned data matrix for dataset A and dataset B.

Return type:

tuple