[BUG] `ComplementNB` fails with an ambiguous truth-value error for per-feature `alpha`

[apiqwe]

Describe the bug

cuml.naive_bayes.ComplementNB accepts a NumPy array as its alpha constructor argument but fails during fit() with an internal scalar comparison:

ValueError: The truth value of an array with more than one element is ambiguous. Use a.any() or a.all()

The equivalent scikit-learn estimator supports an alpha array of shape (n_features,), fits successfully, and predicts [1].

This is a scikit-learn compatibility gap, and the current failure does not clearly report that cuML expects a scalar smoothing value.

Steps/Code to reproduce bug

cuML reproducer:

import numpy as np
from cuml.naive_bayes import ComplementNB

X = np.array([[1, 2], [3, 4], [5, 6]])
y = np.array([0, 1, 0])

alpha = np.array([1.0, 2.0])

print(ComplementNB(alpha=alpha).fit(X, y).predict([[1, 2]]))

Output:

Traceback (most recent call last):
  File "/workspace/apibughub/cuml/ComplementNB/error_bug1/error_bug_cuml.py", line 9, in <module>
    print(ComplementNB(alpha=alpha).fit(X, y).predict([[1, 2]]))
          ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^
  File "/opt/conda/envs/rapids-26.08/lib/python3.14/site-packages/cuml/internals/outputs.py", line 874, in inner
    res = func(*args, **kwargs)
  File "/opt/conda/envs/rapids-26.08/lib/python3.14/site-packages/cuml/naive_bayes/naive_bayes.py", line 702, in fit
    return self._partial_fit(X, y, reset=True)
           ~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^
  File "/opt/conda/envs/rapids-26.08/lib/python3.14/site-packages/cuml/naive_bayes/naive_bayes.py", line 661, in _partial_fit
    if self.alpha < 0:
       ^^^^^^^^^^^^^^
ValueError: The truth value of an array with more than one element is ambiguous. Use a.any() or a.all()

For comparison, the equivalent scikit-learn code:

import numpy as np
from sklearn.naive_bayes import ComplementNB

X = np.array([[1, 2], [3, 4], [5, 6]])
y = np.array([0, 1, 0])

alpha = np.array([1.0, 2.0])

print(ComplementNB(alpha=alpha).fit(X, y).predict([[1, 2]]))

Output:

[1]

Expected behavior

Preferably, cuml.naive_bayes.ComplementNB should support a per-feature smoothing array with shape (n_features,), matching scikit-learn. In this reproducer, alpha=np.array([1.0, 2.0]) matches the two input features, so fitting should complete and prediction should return:

[1]

If array-valued alpha is intentionally unsupported, cuML should validate the argument and raise a clear TypeError or ValueError stating that alpha must be a scalar. It should not evaluate an array in a scalar Boolean context and expose NumPy's ambiguous truth-value error.

Environment details (please complete the following information):

• Environment location: Docker
• Linux Distro/Architecture: Ubuntu 24.04 / x86_64
• GPU Model/Driver: NVIDIA GeForce RTX 4090 / 595.71.05
• CUDA: 13.2
• Method of cuDF & cuML install: conda

conda list:

conda list
# packages in environment at /opt/conda/envs/rapids-26.08:
#
# Name                              Version          Build                                         Channel
# Name              Version       Build                                      Channel
python              3.14.6        h242f9ac_102_cp314                         conda-forge
numpy               2.4.6         py314h2b28147_0                            conda-forge
scipy               1.16.3        py314hf07bd8e_2                            conda-forge
scikit-learn        1.9.0         np2py314hf09ca88_0                         conda-forge
rapids              26.08.00      cuda13_260806_c2656556                     rapidsai
cuml                26.08.00      cuda13_cp311_abi3_260805_265b9da6          rapidsai
libcuml             26.08.00      cuda13_260805_265b9da6                     rapidsai
cudf                26.08.00      cuda13_cp311_abi3_260805_ff5b362d          rapidsai
libraft             26.08.00      cuda13_260805_ebf92684                     rapidsai
libraft-headers     26.08.00      cuda13_260805_ebf92684                     rapidsai
pylibraft           26.08.00      cuda13_cp311_abi3_260805_ebf92684          rapidsai
cuvs                26.08.01      cuda13_cp311_abi3_260806_25b1be43          rapidsai
libcuvs             26.08.01      cuda13_260806_25b1be43                     rapidsai
cupy                14.1.1        py314hdea9c46_0                            conda-forge
cupy-core           14.1.1        py314hcd3b49b_0                            conda-forge
numba               0.64.0        py314h8169c2f_0                            conda-forge
numba-cuda          0.30.4        py314h42812f9_0                            conda-forge
rmm                 26.08.00      cuda13_cp311_abi3_260805_42d059f1          rapidsai
librmm              26.08.00      cuda13_260805_42d059f1                     rapidsai
cuda-version        13.3           hcbadf70_3                                 conda-forge
cuda-bindings       13.3.1        py314h42812f9_1                            conda-forge
cuda-cudart         13.3.29       hecca717_0                                 conda-forge
cuda-nvrtc          13.3.33       hecca717_0                                 conda-forge
libcublas           13.6.0.2      h676940d_0                                 conda-forge
libcusolver         12.2.6.9      h676940d_0                                 conda-forge
libcusparse         12.8.2.51     hecca717_0                                 conda-forge
libcurand           10.4.3.29     h676940d_0                                 conda-forge

Additional context

The failure comes from this scalar-only validation in ComplementNB._partial_fit():

if self.alpha < 0:

For an array-valued alpha, the comparison produces a Boolean array, which cannot be used directly as an if condition.

The input has two features and the smoothing array has exactly two elements:

X.shape = (3, 2)
alpha.shape = (2,)

All smoothing values are finite and strictly positive, so a per-element non-negativity check would pass.

The scikit-learn ComplementNB documentation defines alpha as a float or an array-like of shape (n_features,). The cuML ComplementNB documentation currently documents only a scalar float. The report therefore requests either compatible per-feature smoothing support or explicit validation of the narrower cuML contract.