mescla.emma.rank.parallel_analysis

mescla.emma.rank.parallel_analysis(Z, n_iter=200, percentile=95.0, seed=0)

Horn’s parallel analysis: retain components beating uncorrelated noise.

Generates n_iter random data sets of the same shape, decomposes each, and keeps the components whose eigenvalue exceeds the given percentile of the corresponding random eigenvalue. The most defensible of the automatic criteria because it accounts for sample size – with few samples, eigenvalues above 1 arise by chance routinely.

Parameters:

• Z (ndarray, shape (n_samples, n_species)) – Standardised data.

• n_iter (int, default 200)

• percentile (float, default 95.0)

• seed (int, optional)

Return type:

int