Skip to content

Commit 0809a3e

Browse files
icfausticfaust
authored
[enhancement] enforce use of sklearn's check_is_fitted before checking oneDAL support (#2110)
* Update test_common.py * formatting * change to fstrings * fix all but SVM algos * add SVM fixes * Update test_common.py * forgotten index switch after merge
1 parent 2ec180c commit 0809a3e

File tree

9 files changed

+61
-12
lines changed

9 files changed

+61
-12
lines changed

‎sklearnex/decomposition/pca.py‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -170,6 +170,7 @@ def _onedal_fit(self, X, queue=None):
170170

171171
@wrap_output_data
172172
def transform(self, X):
173+
check_is_fitted(self)
173174
return dispatch(
174175
self,
175176
"transform",
@@ -181,7 +182,6 @@ def transform(self, X):
181182
)
182183

183184
def _onedal_transform(self, X, queue=None):
184-
check_is_fitted(self)
185185
if sklearn_check_version("1.0"):
186186
X = validate_data(
187187
self,

‎sklearnex/ensemble/_forest.py‎

Lines changed: 5 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -598,6 +598,7 @@ def _onedal_fit_ready(self, patching_status, X, y, sample_weight):
598598

599599
@wrap_output_data
600600
def predict(self, X):
601+
check_is_fitted(self)
601602
return dispatch(
602603
self,
603604
"predict",
@@ -613,7 +614,7 @@ def predict_proba(self, X):
613614
# TODO:
614615
# _check_proba()
615616
# self._check_proba()
616-
617+
check_is_fitted(self)
617618
return dispatch(
618619
self,
619620
"predict_proba",
@@ -639,6 +640,7 @@ def predict_log_proba(self, X):
639640

640641
@wrap_output_data
641642
def score(self, X, y, sample_weight=None):
643+
check_is_fitted(self)
642644
return dispatch(
643645
self,
644646
"score",
@@ -786,7 +788,6 @@ def _onedal_gpu_supported(self, method_name, *data):
786788
return patching_status
787789

788790
def _onedal_predict(self, X, queue=None):
789-
check_is_fitted(self, "_onedal_estimator")
790791

791792
if sklearn_check_version("1.0"):
792793
X = validate_data(
@@ -822,7 +823,6 @@ def _onedal_predict(self, X, queue=None):
822823
return np.take(self.classes_, res.ravel().astype(np.int64, casting="unsafe"))
823824

824825
def _onedal_predict_proba(self, X, queue=None):
825-
check_is_fitted(self, "_onedal_estimator")
826826

827827
if sklearn_check_version("1.0"):
828828
X = validate_data(
@@ -1166,6 +1166,7 @@ def fit(self, X, y, sample_weight=None):
11661166

11671167
@wrap_output_data
11681168
def predict(self, X):
1169+
check_is_fitted(self)
11691170
return dispatch(
11701171
self,
11711172
"predict",
@@ -1178,6 +1179,7 @@ def predict(self, X):
11781179

11791180
@wrap_output_data
11801181
def score(self, X, y, sample_weight=None):
1182+
check_is_fitted(self)
11811183
return dispatch(
11821184
self,
11831185
"score",

‎sklearnex/linear_model/logistic_regression.py‎

Lines changed: 4 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -129,6 +129,7 @@ def fit(self, X, y, sample_weight=None):
129129

130130
@wrap_output_data
131131
def predict(self, X):
132+
check_is_fitted(self)
132133
return dispatch(
133134
self,
134135
"predict",
@@ -141,6 +142,7 @@ def predict(self, X):
141142

142143
@wrap_output_data
143144
def predict_proba(self, X):
145+
check_is_fitted(self)
144146
return dispatch(
145147
self,
146148
"predict_proba",
@@ -153,6 +155,7 @@ def predict_proba(self, X):
153155

154156
@wrap_output_data
155157
def predict_log_proba(self, X):
158+
check_is_fitted(self)
156159
return dispatch(
157160
self,
158161
"predict_log_proba",
@@ -165,6 +168,7 @@ def predict_log_proba(self, X):
165168

166169
@wrap_output_data
167170
def score(self, X, y, sample_weight=None):
171+
check_is_fitted(self)
168172
return dispatch(
169173
self,
170174
"score",
@@ -321,7 +325,6 @@ def _onedal_predict(self, X, queue=None):
321325
if queue is None or queue.sycl_device.is_cpu:
322326
return daal4py_predict(self, X, "computeClassLabels")
323327

324-
check_is_fitted(self)
325328
if sklearn_check_version("1.0"):
326329
X = validate_data(
327330
self,
@@ -346,7 +349,6 @@ def _onedal_predict_proba(self, X, queue=None):
346349
if queue is None or queue.sycl_device.is_cpu:
347350
return daal4py_predict(self, X, "computeClassProbabilities")
348351

349-
check_is_fitted(self)
350352
if sklearn_check_version("1.0"):
351353
X = validate_data(
352354
self,
@@ -371,7 +373,6 @@ def _onedal_predict_log_proba(self, X, queue=None):
371373
if queue is None or queue.sycl_device.is_cpu:
372374
return daal4py_predict(self, X, "computeClassLogProbabilities")
373375

374-
check_is_fitted(self)
375376
if sklearn_check_version("1.0"):
376377
X = validate_data(
377378
self,

‎sklearnex/neighbors/knn_unsupervised.py‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -101,6 +101,7 @@ def radius_neighbors(
101101
and self._fit_method == "kd_tree"
102102
):
103103
_sklearn_NearestNeighbors.fit(self, self._fit_X, getattr(self, "_y", None))
104+
check_is_fitted(self)
104105
return dispatch(
105106
self,
106107
"radius_neighbors",

‎sklearnex/svm/nusvc.py‎

Lines changed: 10 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -18,7 +18,11 @@
1818
from sklearn.exceptions import NotFittedError
1919
from sklearn.metrics import accuracy_score
2020
from sklearn.svm import NuSVC as _sklearn_NuSVC
21-
from sklearn.utils.validation import _deprecate_positional_args, check_array
21+
from sklearn.utils.validation import (
22+
_deprecate_positional_args,
23+
check_array,
24+
check_is_fitted,
25+
)
2226

2327
from daal4py.sklearn._n_jobs_support import control_n_jobs
2428
from daal4py.sklearn._utils import sklearn_check_version
@@ -115,6 +119,7 @@ def fit(self, X, y, sample_weight=None):
115119

116120
@wrap_output_data
117121
def predict(self, X):
122+
check_is_fitted(self)
118123
return dispatch(
119124
self,
120125
"predict",
@@ -127,6 +132,7 @@ def predict(self, X):
127132

128133
@wrap_output_data
129134
def score(self, X, y, sample_weight=None):
135+
check_is_fitted(self)
130136
return dispatch(
131137
self,
132138
"score",
@@ -169,6 +175,7 @@ def predict_proba(self, X):
169175
predict. Also, it will produce meaningless results on very small
170176
datasets.
171177
"""
178+
check_is_fitted(self)
172179
return self._predict_proba(X)
173180

174181
@available_if(_sklearn_NuSVC._check_proba)
@@ -208,6 +215,7 @@ def predict_log_proba(self, X):
208215
@property
209216
def predict_proba(self):
210217
self._check_proba()
218+
check_is_fitted(self)
211219
return self._predict_proba
212220

213221
def _predict_log_proba(self, X):
@@ -236,6 +244,7 @@ def _predict_proba(self, X):
236244

237245
@wrap_output_data
238246
def decision_function(self, X):
247+
check_is_fitted(self)
239248
return dispatch(
240249
self,
241250
"decision_function",

‎sklearnex/svm/nusvr.py‎

Lines changed: 7 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -16,7 +16,11 @@
1616

1717
import numpy as np
1818
from sklearn.svm import NuSVR as _sklearn_NuSVR
19-
from sklearn.utils.validation import _deprecate_positional_args, check_array
19+
from sklearn.utils.validation import (
20+
_deprecate_positional_args,
21+
check_array,
22+
check_is_fitted,
23+
)
2024

2125
from daal4py.sklearn._n_jobs_support import control_n_jobs
2226
from daal4py.sklearn._utils import sklearn_check_version
@@ -97,6 +101,7 @@ def fit(self, X, y, sample_weight=None):
97101

98102
@wrap_output_data
99103
def predict(self, X):
104+
check_is_fitted(self)
100105
return dispatch(
101106
self,
102107
"predict",
@@ -109,6 +114,7 @@ def predict(self, X):
109114

110115
@wrap_output_data
111116
def score(self, X, y, sample_weight=None):
117+
check_is_fitted(self)
112118
return dispatch(
113119
self,
114120
"score",

‎sklearnex/svm/svc.py‎

Lines changed: 10 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -19,7 +19,11 @@
1919
from sklearn.exceptions import NotFittedError
2020
from sklearn.metrics import accuracy_score
2121
from sklearn.svm import SVC as _sklearn_SVC
22-
from sklearn.utils.validation import _deprecate_positional_args, check_array
22+
from sklearn.utils.validation import (
23+
_deprecate_positional_args,
24+
check_array,
25+
check_is_fitted,
26+
)
2327

2428
from daal4py.sklearn._n_jobs_support import control_n_jobs
2529
from daal4py.sklearn._utils import sklearn_check_version
@@ -117,6 +121,7 @@ def fit(self, X, y, sample_weight=None):
117121

118122
@wrap_output_data
119123
def predict(self, X):
124+
check_is_fitted(self)
120125
return dispatch(
121126
self,
122127
"predict",
@@ -129,6 +134,7 @@ def predict(self, X):
129134

130135
@wrap_output_data
131136
def score(self, X, y, sample_weight=None):
137+
check_is_fitted(self)
132138
return dispatch(
133139
self,
134140
"score",
@@ -171,6 +177,7 @@ def predict_proba(self, X):
171177
predict. Also, it will produce meaningless results on very small
172178
datasets.
173179
"""
180+
check_is_fitted(self)
174181
return self._predict_proba(X)
175182

176183
@available_if(_sklearn_SVC._check_proba)
@@ -210,6 +217,7 @@ def predict_log_proba(self, X):
210217
@property
211218
def predict_proba(self):
212219
self._check_proba()
220+
check_is_fitted(self)
213221
return self._predict_proba
214222

215223
def _predict_log_proba(self, X):
@@ -238,6 +246,7 @@ def _predict_proba(self, X):
238246

239247
@wrap_output_data
240248
def decision_function(self, X):
249+
check_is_fitted(self)
241250
return dispatch(
242251
self,
243252
"decision_function",

‎sklearnex/svm/svr.py‎

Lines changed: 3 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -16,7 +16,7 @@
1616

1717
import numpy as np
1818
from sklearn.svm import SVR as _sklearn_SVR
19-
from sklearn.utils.validation import _deprecate_positional_args
19+
from sklearn.utils.validation import _deprecate_positional_args, check_is_fitted
2020

2121
from daal4py.sklearn._n_jobs_support import control_n_jobs
2222
from daal4py.sklearn._utils import sklearn_check_version
@@ -98,6 +98,7 @@ def fit(self, X, y, sample_weight=None):
9898

9999
@wrap_output_data
100100
def predict(self, X):
101+
check_is_fitted(self)
101102
return dispatch(
102103
self,
103104
"predict",
@@ -110,6 +111,7 @@ def predict(self, X):
110111

111112
@wrap_output_data
112113
def score(self, X, y, sample_weight=None):
114+
check_is_fitted(self)
113115
return dispatch(
114116
self,
115117
"score",

‎sklearnex/tests/test_common.py‎

Lines changed: 20 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -365,7 +365,26 @@ def runtime_property_check(text, estimator, method):
365365
), f"{estimator}.{method} should only use 'property' at instantiation"
366366

367367

368-
DESIGN_RULES = [n_jobs_check, runtime_property_check]
368+
def fit_check_before_support_check(text, estimator, method):
369+
if "fit" not in method:
370+
if "_onedal_cpu_supported" in text["funcs"]:
371+
onedal_support = "_onedal_cpu_supported"
372+
elif "_onedal_gpu_supported" in text["funcs"]:
373+
onedal_support = "_onedal_gpu_supported"
374+
else:
375+
pytest.skip(f"onedal dispatching not used in {estimator}.{method}")
376+
# get location of _onedal_*_supported
377+
idx = len(text["funcs"]) - 1 - text["funcs"][::-1].index(onedal_support)
378+
validfuncs = text["funcs"][:idx]
379+
assert (
380+
"check_is_fitted" in validfuncs
381+
), f"sklearn's check_is_fitted must be called before checking oneDAL support"
382+
383+
else:
384+
pytest.skip(f"fitting occurs in {estimator}.{method}")
385+
386+
387+
DESIGN_RULES = [n_jobs_check, runtime_property_check, fit_check_before_support_check]
369388

370389

371390
if sklearn_check_version("1.0"):

0 commit comments

Comments
 (0)