mojolearn.sparse_encode#

mojolearn.sparse_encode(X, dictionary, *, algorithm='lasso_lars', n_nonzero_coefs=None, alpha=None, init=None, max_iter=1000, positive=False, numeric_mode=None, **_ignored)#

Sparse-code each row of X against a fixed dictionary.

Follows scikit-learn’s sparse_encode. "lasso_lars" and "lasso_cd" solve the same Lasso problem (both by coordinate descent; the optimum does not depend on the path). Other scikit-learn keywords (such as n_jobs or check_input) are accepted and ignored.

Parameters:

  • X (array-like of shape (n_samples, n_features))

  • dictionary (array-like of shape (n_components, n_features)) – Dictionary atoms, one per row.

  • algorithm ({“lasso_lars”, “lasso_cd”, “lars”, “omp”, “threshold”}, default “lasso_lars”) – Coding method.

  • n_nonzero_coefs (int or None, default None) – Nonzero coefficients per row for "lars" and "omp". None uses min(max(n_features / 10, 1), n_components).

  • alpha (float or None, default None) – Penalty for the Lasso methods, or threshold for "threshold". None uses 1.0.

  • init (array-like of shape (n_samples, n_components) or None, default None) – Starting code for "lasso_cd".

  • max_iter (int, default 1000) – Maximum iterations for the Lasso methods.

  • positive (bool, default False) – Force the code to be non-negative (Lasso methods and "threshold").

  • numeric_mode ({“identical”, “fast”} or None, default None) – Numeric mode for this call. None uses the package default ("identical" unless changed). In identical mode, mojolearn returns the same bits on Apple, NVIDIA and AMD GPUs. "fast" is available on Apple GPUs only.

Returns:

ndarray of shape (n_samples, n_components), float32

Raises:

ValueError – For an unknown algorithm.

Examples

>>> import numpy as np, mojolearn
>>> rng = np.random.default_rng(0)
>>> D = rng.random((5, 8), dtype=np.float32)
>>> code = mojolearn.sparse_encode(rng.random((100, 8), dtype=np.float32), D, algorithm="omp", n_nonzero_coefs=2)
On this page