There is a conflict between matplotlib and PCA

[hewenkui]

Using sklearn.decomposition.PCA, if the matplotlib library is used to draw a picture, there will be an exception ( numpy.linalg.LinAlgError)

The test code(windows x64, matplotlib==3.2.2, scipy==1.5.0, numpy==1.19.0):

import numpy as np
from sklearn.decomposition import PCA
import matplotlib.pylab as plt

def save_pic():
    fig = plt.figure(figsize=(12, 6))
    ax = fig.add_axes([0.1, 0.13, 0.85, 0.80])
    ax.plot([0, 1], [0, 1], 'k--')
    fig.savefig('./a.png')
    plt.close(fig)

def test(is_call_save_pic):
    a = np.arange(1200).reshape((30, -1))
    if is_call_save_pic:
        save_pic()
    pca = PCA(n_components=2)
    pca.fit(a)
    print('OK')

if __name__ == '__main__':
    test(False)
    try:
        test(True)
    except np.linalg.LinAlgError as e:
        print(e)

[glemaitre]

I cannot reproduce on linux.

[LilianBoulard]

Could not reproduce either on Windows 10 x64 (with scikit-learn 0.23.2 and specified numpy, scipy and matplotlib versions).

[NicolasHug]

Thanks @glemaitre and @Phaide for trying to reproduce the bug. @hewenkui can you try again with the latest version and confirm whether you still observe the exception? thanks

[thomasjpfan]

As suggested in the related issue: #19285 (comment) , the solution is to update NumPy.