maxfuse.utils.svd_denoise

maxfuse.utils.svd_denoise(arr, n_components=20, randomized=False, n_runs=1)

Compute best rank-n_components approximation of arr by SVD.

Parameters:

• arr (np.array of shape (n_samples, n_features)) – Data matrix

• n_components (int, default=20) – Number of components to keep

• randomized (bool, default=False) – Whether to use randomized SVD

• n_runs (int, default=1) – Run multiple times and take the realization with the lowest Frobenious reconstruction error

Returns:

arr (array_like of shape (n_samples, n_features)) – Rank-n_comopnents approximation of the input arr.