SMOTENC crashes on documented categorical_features array inputs (bool mask / str-name array) due to == "auto" check

Author: siddhant-shahhhCreated Sep 20, 2026Updated Sep 20, 2026

Describe the bug

SMOTENC documents that categorical_features accepts a boolean mask array and a string-name array:

mask array of shape (n_features,) and bool dtype for which True indicates the categorical features

array of str corresponding to the feature names

Both crash before being used. In _validate_column_types (imblearn/over_sampling/_smote/base.py:561):

python

if self.categorical_features == "auto":

When categorical_features is a NumPy array, array == "auto" returns an element-wise boolean array, so if <array>: raises ValueError: The truth value of an array with more than one element is ambiguous. A Python list compares to "auto" as a scalar (False) and passes — which is why list inputs work while the documented array inputs do not.

Steps/Code to Reproduce

python

import numpy as np
import pandas as pd
from imblearn.over_sampling import SMOTENC

X = np.hstack([
    np.random.default_rng(3).standard_normal((240, 3)),
    np.random.default_rng(3).integers(0, 4, size=(240, 2)).astype(float),
])
y = np.array([0] * 200 + [1] * 40)

# 1) documented boolean mask -> crashes
mask = np.array([False, False, False, True, True])
SMOTENC(categorical_features=mask, random_state=7).fit_resample(X, y)

# 2) documented str-name array -> crashes the same way
Xdf = pd.DataFrame(X, columns=["c0", "c1", "c2", "cat0", "cat1"])
SMOTENC(categorical_features=np.array(["cat0", "cat1"]), random_state=7).fit_resample(Xdf, y)

# 3) equivalent Python lists both work fine
SMOTENC(categorical_features=[3, 4], random_state=7).fit_resample(X, y)
SMOTENC(categorical_features=["cat0", "cat1"], random_state=7).fit_resample(Xdf, y)

Expected Results

The boolean mask and the str-name array are accepted, identically to their list equivalents, since both are documented input forms.

Actual Results

Traceback (most recent call last):
  ...
  File ".../imblearn/over_sampling/_smote/base.py", line 597, in _fit_resample
    self._validate_column_types(X)
  File ".../imblearn/over_sampling/_smote/base.py", line 561, in _validate_column_types
    if self.categorical_features == "auto":
ValueError: The truth value of an array with more than one element is ambiguous. Use a.any() or a.all()

Versions

System:
    python: 3.12.10 [MSC v.1943 64 bit (AMD64)]
   machine: Windows-11
Python dependencies:
      sklearn: 1.9.1
        numpy: 2.5.3
        scipy: 1.18.1
       pandas: 3.0.5
       joblib: 1.6.0
     imblearn: 0.14.2

I'd be happy to open a PR. The fix is to make the sentinel check string-safe, e.g. if isinstance(self.categorical_features, str) and self.categorical_features == "auto":, plus regression tests passing a boolean mask and a str-name array. One question before I do: would you prefer that minimal guard, or normalizing categorical_features to indices up front in _validate_column_types?

Source: scikit-learn-contrib/imbalanced-learn

View original on GitHub View discussion on GitHub