1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
|
# Maintainer: Smoolak <smoolak@gmail.com>
_pkgname=skl2onnx
pkgname=python-skl2onnx
pkgver=1.20.0
pkgrel=14
pkgdesc='Convert scikit-learn models to ONNX'
arch=('any')
url='https://github.com/onnx/sklearn-onnx'
license=('Apache-2.0')
depends=(
'python'
'python-numpy'
'python-onnx'
'python-packaging'
'python-scikit-learn'
'python-scipy'
# Arch's python-onnx 1.21 imports this at runtime but does not currently
# declare it, so keep the compatibility dependency here until it does.
'python-typing_extensions'
)
makedepends=(
'python-build'
'python-installer'
'python-wheel'
'python-setuptools'
'git'
)
checkdepends=(
'python-pytest'
'python-pytest-xdist'
'python-numpy'
'python-onnxruntime-cpu'
'python-jinja'
'python-pandas'
)
optdepends=(
'python-pandas: for DataFrame type inference'
'python-jinja: for ONNX operator code generation'
)
source=(
"$pkgname-$pkgver.tar.gz::https://github.com/onnx/sklearn-onnx/archive/refs/tags/$pkgver.tar.gz"
'nearest-neighbor-radius.patch::https://github.com/onnx/sklearn-onnx/commit/1c9ef2845b313919b0d756d2c85e55b476a8ed68.patch'
'hist-gradient-boosting-float32.patch::https://github.com/onnx/sklearn-onnx/commit/3ca82079d98e74c0dabb91510e2cf1699fce5dc2.patch'
'tuned-threshold-classifier.patch::https://github.com/onnx/sklearn-onnx/commit/dd3bec163c3d77ac0b9f7a015e69f7aa99b2b966.patch'
'pipeline-raw-scores.patch::https://github.com/onnx/sklearn-onnx/commit/7bdfd51ac7653ae35e36781fbc1ffb5b0dd86d87.patch'
'radius-safe-index.patch'
'onnx-1.21-integer-tree-attributes.patch'
'current-dependencies.patch'
'onnxruntime-1.27-labels.patch'
)
sha256sums=(
'80cfee91f8e545f2746c229855c5bf7c211aee374ad12b9584b6b64f38abfe1d'
'302f423c49014b268991b8ed83a843a8830fd3efde3092e7bb7d32cdcfce8301'
'4b4c062a491671db84eab6631ea973d4283244f8fb9f8325dfcc8021851beb10'
'6dfb419f513b625152df6a2c2cde9d8de3e26007d76b6fb9d8e5629e49fb41e8'
'16300e170569a1e577ad3b4c0d483fd5cda7966ac195aed49c4fc9a255c18198'
'69655b56eefb5940487a855bfeb38929cc936bbdc0ea29971a8d4da484307487'
'97baba298c7dc4c4029bf3a51710d166424fabb104b120d5ea7fd0b736435955'
'1581c51ff4c2955201d23e2616d0d6dc2df1b841e542ca1aa1e3e3343a6b8254'
'6093d5e5468122c73da3264e64851e8f3d36a834cb81625b3376291ca3912adb'
)
prepare() {
cd "sklearn-onnx-$pkgver"
# Backport the converter fixes already merged upstream after 1.20.0. Git
# can omit the rolling changelog hunk while still applying every code and
# regression-test hunk exactly.
local patch
for patch in \
nearest-neighbor-radius.patch \
hist-gradient-boosting-float32.patch \
tuned-threshold-classifier.patch \
pipeline-raw-scores.patch; do
git apply --check --exclude=CHANGELOGS.md "$srcdir/$patch"
git apply --exclude=CHANGELOGS.md "$srcdir/$patch"
done
patch -Np1 --fuzz=0 -i "$srcdir/radius-safe-index.patch"
patch -Np1 --fuzz=0 -i "$srcdir/onnx-1.21-integer-tree-attributes.patch"
patch -Np1 --fuzz=0 -i "$srcdir/current-dependencies.patch"
patch -Np1 --fuzz=0 -i "$srcdir/onnxruntime-1.27-labels.patch"
}
build() {
cd "sklearn-onnx-$pkgver"
python -m build --wheel --no-isolation
}
check() {
cd "sklearn-onnx-$pkgver"
local _checkroot="$srcdir/_check" _site _locale="$srcdir/_locale"
local _bootstrap="$srcdir/_bootstrap"
rm -rf "$_checkroot" "$_locale" "$_bootstrap"
python -m installer --destdir="$_checkroot" dist/*.whl
_site=$(python -c 'import site; print(site.getsitepackages()[0])')
cp -a tests "$_checkroot/tests"
mkdir -p "$_locale" "$_bootstrap"
# ONNX Runtime's StringNormalizer nodes explicitly construct the named
# en_US.UTF-8 C++ locale. Minimal clean chroots only ship C.UTF-8, so
# generate the required locale privately without modifying the host.
localedef --no-archive -i en_US -f UTF-8 "$_locale/en_US.UTF-8"
# Python 3.14 changed multiprocessing's POSIX default to forkserver. A
# handful of upstream conversion tests intentionally use local callables;
# retain their historical fork semantics in the test process only.
cat > "$_bootstrap/sitecustomize.py" <<'PY'
import multiprocessing
try:
multiprocessing.set_start_method("fork")
except RuntimeError:
pass
PY
cd "$_checkroot"
# Exercise every offline upstream test against the installed wheel. Arch
# now ships the CPU ONNX Runtime backend required by its inference checks.
# The three exact TF-IDF nodes below unconditionally fetch the 20 Newsgroups
# corpus from Figshare; no fixture or offline fallback is shipped upstream.
LOCPATH="$_locale" LC_ALL=en_US.UTF-8 \
PYTHONPATH="$_bootstrap:$_checkroot$_site:$_checkroot/tests" \
PYTEST_DISABLE_PLUGIN_AUTOLOAD=1 \
pytest -ra --import-mode=importlib \
-k 'not (test_x_issue_789 or test_x_issue_789_cdist)' \
--deselect=tests/test_sklearn_tfidf_transformer_converter_sparse.py::TestSklearnTfidfVectorizerSparse::test_model_tfidf_transform_bug \
--deselect=tests/test_sklearn_tfidf_vectorizer_converter_dataset.py::TestSklearnTfidfVectorizerDataSet::test_tfidf_20newsgroups \
--deselect=tests/test_sklearn_tfidf_vectorizer_converter_dataset.py::TestSklearnTfidfVectorizerDataSet::test_tfidf_20newsgroups_nolowercase \
tests
# These two upstream tests each fit a 7,500-row Gaussian process (cubic
# Cholesky cost). Run the two independent cases concurrently; they remain
# part of the complete suite without needlessly serializing their cost.
LOCPATH="$_locale" LC_ALL=en_US.UTF-8 \
PYTHONPATH="$_bootstrap:$_checkroot$_site:$_checkroot/tests" \
PYTEST_DISABLE_PLUGIN_AUTOLOAD=1 \
OPENBLAS_NUM_THREADS="${SLURM_CPUS_PER_TASK:-1}" \
OMP_NUM_THREADS="${SLURM_CPUS_PER_TASK:-1}" \
pytest -ra -p xdist.plugin -n 2 --dist=load --import-mode=importlib \
tests/test_sklearn_gaussian_process_regressor.py::TestSklearnGaussianProcessRegressor::test_x_issue_789 \
tests/test_sklearn_gaussian_process_regressor.py::TestSklearnGaussianProcessRegressor::test_x_issue_789_cdist
PYTHONPATH="$_checkroot$_site" python - <<'PY'
import numpy as np
import onnx
from onnx.reference import ReferenceEvaluator
from sklearn.datasets import load_iris
from sklearn.linear_model import LogisticRegression
from sklearn.pipeline import make_pipeline
from sklearn.preprocessing import StandardScaler
from skl2onnx import to_onnx
x, y = load_iris(return_X_y=True)
x = x.astype(np.float32)
model = make_pipeline(StandardScaler(), LogisticRegression(max_iter=300)).fit(x, y)
converted = to_onnx(model, x[:1], options={id(model[-1]): {"zipmap": False}})
onnx.checker.check_model(converted)
labels, probabilities = ReferenceEvaluator(converted).run(None, {"X": x[:12]})
np.testing.assert_array_equal(labels, model.predict(x[:12]))
np.testing.assert_allclose(probabilities, model.predict_proba(x[:12]), rtol=1e-4, atol=1e-5)
print("scikit-learn to ONNX reference-evaluator workflow passed")
PY
}
package() {
cd "sklearn-onnx-$pkgver"
python -m installer --destdir="$pkgdir" dist/*.whl
install -Dm644 LICENSE "$pkgdir/usr/share/licenses/$pkgname/LICENSE"
}
|