pca_svd.py

#!/usr/bin/env python
"""
   pca_svd.py
"""

from sklearn.decomposition import TruncatedSVD, PCA
import numpy as np

X = np.random.normal(0, 1, (100, 10))
X -= X.mean(axis=0)

pca = PCA(svd_solver='full').fit(X)
u, s, v = np.linalg.svd(X, full_matrices=False)
np.allclose(np.abs(pca.components_), np.abs(v)) # True

p = PCA(svd_solver='full').fit_transform(X)
np.allclose(X.dot(pca.components_.T), p) # True

np.allclose(np.abs(u.dot(np.diag(s))), np.abs(p)) # True

p2 = PCA(svd_solver='full', whiten=True).fit_transform(X)
np.allclose(np.abs(p2), np.abs(u * np.sqrt(100 - 1))) # True