Imbalanced classes in Python

When one class is rare, accuracy lies and the weights need saying.

import numpy as np
from sklearn.linear_model import LogisticRegression
from sklearn.metrics import recall_score
from sklearn.utils.class_weight import compute_class_weight

rng = np.random.default_rng(0)
X = rng.normal(size=(200, 2))
y = (rng.random(200) < 0.1).astype(int)

weights = compute_class_weight("balanced", classes=np.array([0, 1]), y=y)
print(weights.round(2))

plain = LogisticRegression().fit(X, y)
weighted = LogisticRegression(class_weight="balanced").fit(X, y)
print(f"plain recall    {recall_score(y, plain.predict(X)):.3f}")
print(f"weighted recall {recall_score(y, weighted.predict(X)):.3f}")

How it works

Setting class_weight to balanced scales the loss by frequency.
compute_class_weight shows the numbers it uses.
Score such a model on recall or AUC, not accuracy.

Keywords and builtins used here

as
int
print

The run, in numbers

Lines: 16
Characters to type: 612
Tokens: 172
Three-star pace: 115 tpm

At the three-star pace of 115 tokens a minute, this run takes about 90 seconds.

Type this snippet

Step 2 of 3 in Beyond the basics, step 23 of 25 in Machine learning with scikit-learn.

← Previous Next →