Closed
Description
After #16224 it looks like this code no longer produces the correct result:
import numpy as np
from sklearn.decomposition import PCA
n_samples, n_dim = 1000, 10
X = np.random.RandomState(0).randn(n_samples, n_dim)
X[:, -1] = np.mean(X[:, :-1], axis=-1) # true X dim is ndim - 1
pca_skl = PCA('mle', svd_solver='full')
pca_skl.fit(X)
assert pca_skl.n_components_ == n_dim - 1
Before #16224 this passed (n_components_ == 9) but after #16224 it gives 8. Not sure why this would happen given the singular value spectrum looks good:
import matplotlib.pyplot as plt
s = np.linalg.svdvals(X)
plt.stem(s)
Maybe an off-by-one error somewhere?
cc'ing @lschwetlick since it was your PR