Preprocessing with L-inf Normalizer of all-negative elements vector returns all-positive elements vector

[maurapintor]

Describe the bug

L inf normalizer should take the maximum of the absolute value before dividing the array.

Steps/Code to Reproduce

from sklearn.preprocessing import Normalizer
import numpy as np

a = np.array([[1]])

normalizer = Normalizer(norm='max')

b = normalizer.fit_transform(a)
c = normalizer.fit_transform(-a)

# they should be different
print(b, c)

Expected Results

The normalized version of a vector with only negative elements should remain negative. It should take maximum absolute value, which is positive, hence remains negative.

Actual Results

The normalization applied to a positive-only vector and to a negative-only vector returns the same results.

Versions

Python dependencies:
pip: 19.0.3
setuptools: 40.8.0
sklearn: 0.22.2.post1
numpy: 1.18.1
scipy: 1.4.1
Cython: None
pandas: None
matplotlib: 3.1.3
joblib: 0.14.1

Possible fix

Take the absolute value before taking the max in this line.

---

Thanks to @freezebat for identifying the problem

[rth]

Thanks for reporting this!