Python crashes in LogisticRegression on v0.22

Description

With certain input data, the LogisticRegression class is reliably crashing Python (without an error message) when fit() is called.

Steps/Code to Reproduce

Please see: https://github.com/fairlearn/sklearn-repro

This repository contains pickled data, the outputs of pip freeze and a sample script. As noted, the script successfully completes with v0.21.3, but crashes silently with v0.22

Expected Results

Either a successful completion (i.e. the final log message appears) or some sort of usable error message.

Actual Results

Python crashes.

Versions

Included in the linked repo, but

System:
    python: 3.6.9 |Anaconda, Inc.| (default, Jul 30 2019, 14:00:49) [MSC v.1915 64 bit (AMD64)]
executable: C:\Users\MyUser\AppData\Local\Continuum\Miniconda3\envs\sklearn-latest\python.exe
   machine: Windows-10-10.0.18362-SP0

Python dependencies:
       pip: 19.3.1
setuptools: 42.0.2.post20191203
   sklearn: 0.22
     numpy: 1.17.4
     scipy: 1.3.3
    Cython: None
    pandas: 0.25.3
matplotlib: None
    joblib: 0.14.0

Built with OpenMP: True

[TomDLT]

Thanks for the report.

This is a subtle bug in check_array, when dtype is a tuple, and when called on pandas dataframes with mixed dtypes:

import pandas as pd
import numpy as np
from sklearn.utils.validation import check_array

df = pd.DataFrame([[33., False]])
print(check_array(df, dtype=[np.float64, np.float32]))
# [[33.0 False]]

df = [[33., False]]
print(check_array(df, dtype=[np.float64, np.float32]))
# [[33.  0.]]

df = pd.DataFrame([[33., False]])
print(check_array(df, dtype=np.float64))
# [[33.  0.]]

[jnothman]

Right, so it’s an effect of #15795?