[BUG] `TargetEncoder` treats a fitted `None` category as unseen during `transform`

[poiuyt009]

Describe the bug

cuml.preprocessing.TargetEncoder does not preserve or match a None category learned during fit when transforming new data.

In the reproducer below, None occurs in the training feature with target value 1.5. With smooth=0, transforming the same category should therefore produce its learned category mean, 1.5.

Instead, cuML returns 2.25, which is the overall mean of the complete target array:

(1.0 + 2.0 + 1.5 + 3.0 + 2.5 + 3.5) / 6 = 2.25

This indicates that cuML treats the previously fitted None value as an unseen category and incorrectly applies the global fallback. The equivalent scikit-learn estimator correctly returns 1.5.

Steps/Code to reproduce bug

cuML reproducer:

import numpy as np
from cuml.preprocessing import TargetEncoder 

X = np.array([
    ["a"],
    ["b"],
    [None],
    ["c"],
    ["b"],
    ["c"],
], dtype=object)

y = np.array([1.0, 2.0, 1.5, 3.0, 2.5, 3.5])

X_test = np.array([
    ["a"],
    [None],
], dtype=object)

k = TargetEncoder(
    smooth=0,
).fit(X, y)

print(k.transform(X_test))

Output:

[[1.  ]
 [2.25]]

For comparison, the equivalent scikit-learn code:

import numpy as np
from sklearn.preprocessing import TargetEncoder 

X = np.array([
    ["a"],
    ["b"],
    [None],
    ["c"],
    ["b"],
    ["c"],
], dtype=object)

y = np.array([1.0, 2.0, 1.5, 3.0, 2.5, 3.5])

X_test = np.array([
    ["a"],
    [None],
], dtype=object)

k = TargetEncoder(
    smooth=0,
).fit(X, y)

print(k.transform(X_test))

Output:

[[1. ]
 [1.5]]

Expected behavior

A missing-value category that occurs during fit should remain a known category during transform.

With smooth=0, each known category should be encoded using its observed target mean. For this input:

• category "a" should be encoded as 1.0;
• category None should be encoded as 1.5.

The expected result, consistent with scikit-learn, is:

[[1. ]
 [1.5]]

The global target statistic 2.25 should be used only for genuinely unseen categories, not for a None category present in the fitted data.

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 training input is a NumPy object array with shape (6, 1).
• None appears once during fitting and is also present in X_test.
• smooth=0 removes smoothing as a possible explanation for the discrepancy.
• The value returned by cuML for None, 2.25, is exactly the global mean of y, strongly indicating that the category lookup did not match the fitted missing-value key.
• This may result from null values being excluded during the groupby aggregation, or from inconsistent null-key handling between the fitted encoding table and the transform-time join.
• The ordinary string category "a" is encoded correctly as 1.0.

[betatim]

Thanks for reporting this. I've created a PR that should fix this