Memory leak in fit of Logistic Regression using KFold

Describe the bug

It has been noticed that some memory is leaked during the training phase of Logistic Regression. I generate the dataset with make_classification and then use KFold.split

Steps/Code to Reproduce

Example:

import numpy as np
import tracemalloc

from sklearn.linear_model import LogisticRegression
from sklearn.datasets import make_classification
from sklearn.model_selection import KFold

def gen_clsf_data():
    data, label = make_classification(
        n_samples=5, n_features=5, random_state=777)
    return data, label, \
        data.size * data.dtype.itemsize + label.size * label.dtype.itemsize


x, y, data_memory_size = gen_clsf_data()
x = np.ascontiguousarray(x)

kf = KFold(n_splits=2)

tracemalloc.start()
for train_index, test_index in kf.split(x):
    x_train, x_test = x[train_index], x[test_index]
    y_train, y_test = y[train_index], y[test_index]
    
    mem_before, _ = tracemalloc.get_traced_memory()    
    alg = LogisticRegression()
    alg.fit(x_train, y_train)
    mem_after, _ = tracemalloc.get_traced_memory()
    
    mem_diff = mem_after - mem_before
    assert mem_diff < 0.25 * data_memory_size, \
        'Size of extra allocated memory is greater than 25% of input data:' \
        f'\n\tInput data size: {data_memory_size} bytes' \
        f'\n\tExtra allocated memory size: {mem_diff} bytes' \
        f' / {round((mem_diff) / data_memory_size * 100, 2)} %'
tracemalloc.stop()

Expected Results

I guess that memory usage should be less than 25% of the input dataset.

Actual Results

Input data size: 240 bytes Extra allocated memory size: 103117 bytes / 42965.42 %

Versions

scikit-learn == 0.24.2

pip install scikit-learn

[OnlyDeniko]

Thanks for the answer. It became clear to me. It seems to me that the issue can be closed

[thomasjpfan]

Thank you for the help with debugging this memory issue @i-aki-y !