Update CI with onnxruntime==1.19.0

.github/workflows/linux-ci.yml

@@ -14,14 +14,14 @@ jobs:
             numpy_version: '>=1.21.1,<2.0'
             scipy_version: '>=1.7.0'
             onnx_version: 'onnx==1.16.0'
-            onnxrt_version: 'onnxruntime==1.18.0'
+            onnxrt_version: 'onnxruntime==1.19.0'
             python_version: '3.12'
           - python_version: '3.12'
             documentation: 0
             numpy_version: '>=1.21.1,<2.0'
             scipy_version: '>=1.7.0'
             onnx_version: 'onnx==1.16.0'
-            onnxrt_version: 'onnxruntime==1.18.0'
+            onnxrt_version: 'onnxruntime==1.18.1'
             sklearn_version: '==1.4.2'
           - python_version: '3.11'
             documentation: 1

.github/workflows/windows-macos-ci.yml

@@ -9,12 +9,18 @@ jobs:
         os: [windows-latest, macos-latest]
         sklearn_version: ['==1.5.0', '==1.4.2', '==1.3.2', '==1.2.2', '==1.1.3']
         include:
+          - sklearn_version: '==1.5.0'
+            python_version: '3.12'
+            numpy_version: '>=1.21.1,<2.0'
+            scipy_version: '>=1.7.0'
+            onnx_version: 'onnx==1.16.0'
+            onnxrt_version: 'onnxruntime<1.20.0'
           - sklearn_version: '==1.5.0'
             python_version: '3.11'
             numpy_version: '>=1.21.1,<2.0'
             scipy_version: '>=1.7.0'
             onnx_version: 'onnx==1.16.0'
-            onnxrt_version: 'onnxruntime==1.18.0'
+            onnxrt_version: 'onnxruntime<1.19.0'
           - python_version: '3.11'
             numpy_version: '>=1.21.1,<2.0'
             scipy_version: '>=1.7.0'

benchmarks/bench_plot_onnxruntime_decision_tree.py

@@ -4,6 +4,7 @@
 """
 Benchmark of onnxruntime on DecisionTree.
 """
+
 # Authors: Xavier Dupré (benchmark)
 from io import BytesIO
 from time import perf_counter as time
@@ -101,7 +102,7 @@ def bench(n_obs, n_features, max_depths, methods, repeat=10, verbose=False):
 
                     # creates different inputs to avoid caching in any ways
                     Xs = []
-                    for r in range(repeat):
+                    for _r in range(repeat):
                         x = np.empty((n, nfeat))
                         x[:, :] = rand(n, nfeat)[:, :]
                         Xs.append(x.astype(np.float32))
@@ -123,7 +124,7 @@ def bench(n_obs, n_features, max_depths, methods, repeat=10, verbose=False):
                     r2 = 0
                     for X in Xs:
                         p2 = fct2(X)
-                        r2 += 1
+                        r2 += 1  # noqa: SIM113
                         if r2 >= repeated:
                             break
                     end = time()

benchmarks/bench_plot_onnxruntime_hgb.py

@@ -4,6 +4,7 @@
 """
 Benchmark of onnxruntime on RandomForestRegressor.
 """
+
 import warnings
 from io import BytesIO
 from time import perf_counter as time
@@ -100,7 +101,7 @@ def bench(
 
                         # creates different inputs to avoid caching in any ways
                         Xs = []
-                        for r in range(repeat):
+                        for _r in range(repeat):
                             x = np.empty((n, nfeat))
                             x[:, :] = rand(n, nfeat)[:, :]
                             Xs.append(x.astype(np.float32))
@@ -122,7 +123,7 @@ def bench(
                         r2 = 0
                         for X in Xs:
                             p2 = fct2(X)
-                            r2 += 1
+                            r2 += 1  # noqa: SIM113
                             if r2 >= repeated:
                                 break
                         end = time()
@@ -140,7 +141,7 @@ def bench(
                             try:
                                 assert_almost_equal(p1.ravel(), p2.ravel(), decimal=5)
                             except AssertionError as e:
-                                warnings.warn(str(e))
+                                warnings.warn(str(e), stacklevel=0)
     return res
 
 
@@ -158,7 +159,7 @@ def plot_results(df, verbose=False):
     for n_obs in sorted(set(df.n_obs)):
         for max_depth in sorted(set(df.max_depth)):
             pos = 0
-            for n_jobs in [1]:
+            for _n_jobs in [1]:
                 a = ax[row, pos]
                 if row == ax.shape[0] - 1:
                     a.set_xlabel("N features", fontsize="x-small")

benchmarks/bench_plot_onnxruntime_linreg.py

@@ -4,6 +4,7 @@
 """
 Benchmark of onnxruntime on LinearRegression.
 """
+
 import warnings
 from io import BytesIO
 from time import perf_counter as time
@@ -100,7 +101,7 @@ def bench(n_obs, n_features, fit_intercepts, methods, repeat=10, verbose=False):
 
                     # creates different inputs to avoid caching in any ways
                     Xs = []
-                    for r in range(loop_repeat):
+                    for _r in range(loop_repeat):
                         x = np.empty((n, nfeat))
                         x[:, :] = rand(n, nfeat)[:, :]
                         Xs.append(x.astype(np.float32))
@@ -120,21 +121,20 @@ def bench(n_obs, n_features, fit_intercepts, methods, repeat=10, verbose=False):
                     r2 = 0
                     for X in Xs:
                         p2 = fct2(X)
-                        r2 += 1
+                        r2 += 1  # noqa: SIM113
                     end = time()
                     obs["time_ort"] = (end - st) / repeated
                     res.append(obs)
                     if verbose and (len(res) % 1 == 0 or n >= 10000):
                         print("bench", len(res), ":", obs)
 
                     # checks that both produce the same outputs
-                    if n <= 10000:
-                        if len(p1.shape) == 1 and len(p2.shape) == 2:
-                            p2 = p2.ravel()
-                            try:
-                                assert_almost_equal(p1.ravel(), p2.ravel(), decimal=5)
-                            except AssertionError as e:
-                                warnings.warn(str(e))
+                    if n <= 10000 and len(p1.shape) == 1 and len(p2.shape) == 2:
+                        p2 = p2.ravel()
+                        try:
+                            assert_almost_equal(p1.ravel(), p2.ravel(), decimal=5)
+                        except AssertionError as e:
+                            warnings.warn(str(e), stacklevel=0)
     return res
 
 

benchmarks/bench_plot_onnxruntime_logreg.py

@@ -4,6 +4,7 @@
 """
 Benchmark of onnxruntime on LogisticRegression.
 """
+
 # Authors: Xavier Dupré (benchmark)
 from io import BytesIO
 from time import perf_counter as time
@@ -113,7 +114,7 @@ def bench(n_obs, n_features, fit_intercepts, methods, repeat=10, verbose=False):
 
                     # creates different inputs to avoid caching in any ways
                     Xs = []
-                    for r in range(repeat):
+                    for _r in range(repeat):
                         x = np.empty((n, nfeat))
                         x[:, :] = rand(n, nfeat)[:, :]
                         Xs.append(x.astype(np.float32))
@@ -133,7 +134,7 @@ def bench(n_obs, n_features, fit_intercepts, methods, repeat=10, verbose=False):
                     r2 = 0
                     for X in Xs:
                         p2 = fct2(X)
-                        r2 += 1
+                        r2 += 1  # noqa: SIM113
                     end = time()
                     obs["time_ort"] = (end - st) / repeated
                     res.append(obs)

benchmarks/bench_plot_onnxruntime_random_forest.py

@@ -4,6 +4,7 @@
 """
 Benchmark of onnxruntime on RandomForest.
 """
+
 # Authors: Xavier Dupré (benchmark)
 from io import BytesIO
 from time import perf_counter as time
@@ -118,7 +119,7 @@ def bench(
 
                         # creates different inputs to avoid caching in any ways
                         Xs = []
-                        for r in range(repeat):
+                        for _r in range(repeat):
                             x = np.empty((n, nfeat))
                             x[:, :] = rand(n, nfeat)[:, :]
                             Xs.append(x)
@@ -140,7 +141,7 @@ def bench(
                         r2 = 0
                         for X in Xs:
                             p2 = fct2(X)
-                            r2 += 1
+                            r2 += 1  # noqa: SIM113
                             if r2 >= repeated:
                                 break
                         end = time()

benchmarks/bench_plot_onnxruntime_random_forest_reg.py

@@ -4,6 +4,7 @@
 """
 Benchmark of onnxruntime on RandomForestRegressor.
 """
+
 import sys
 import warnings
 from io import BytesIO
@@ -137,7 +138,7 @@ def bench(
                             # creates different inputs to avoid caching
                             # in any ways
                             Xs = []
-                            for r in range(repeat):
+                            for _r in range(repeat):
                                 x = np.empty((n, nfeat))
                                 x[:, :] = rand(n, nfeat)[:, :]
                                 Xs.append(x.astype(np.float32))
@@ -159,7 +160,7 @@ def bench(
                             r2 = 0
                             for X in Xs:
                                 p2 = fct2(X)
-                                r2 += 1
+                                r2 += 1  # noqa: SIM113
                                 if r2 >= repeated:
                                     break
                             end = time()
@@ -191,7 +192,7 @@ def bench(
                                         p1.ravel(), p2.ravel(), decimal=5
                                     )
                                 except AssertionError as e:
-                                    warnings.warn(str(e))
+                                    warnings.warn(str(e), stacklevel=0)
     return res
 
 

benchmarks/bench_plot_onnxruntime_svm_reg.py

@@ -4,6 +4,7 @@
 """
 Benchmark of onnxruntime on SVM.
 """
+
 # Authors: Xavier Dupré (benchmark)
 import warnings
 from io import BytesIO
@@ -86,7 +87,7 @@ def bench(n_obs, n_features, kernels, methods, repeat=10, verbose=False):
 
                     # creates different inputs to avoid caching in any ways
                     Xs = []
-                    for r in range(repeat):
+                    for _r in range(repeat):
                         x = np.empty((n, nfeat))
                         x[:, :] = rand(n, nfeat)[:, :]
                         Xs.append(x.astype(np.float32))
@@ -108,7 +109,7 @@ def bench(n_obs, n_features, kernels, methods, repeat=10, verbose=False):
                     r2 = 0
                     for X in Xs:
                         p2 = fct2(X)
-                        r2 += 1
+                        r2 += 1  # noqa: SIM113
                         if r2 >= repeated:
                             break
                     end = time()
@@ -126,7 +127,7 @@ def bench(n_obs, n_features, kernels, methods, repeat=10, verbose=False):
                         try:
                             assert_almost_equal(p1.ravel(), p2.ravel(), decimal=3)
                         except AssertionError as e:
-                            warnings.warn(str(e))
+                            warnings.warn(str(e), stacklevel=0)
     return res
 
 

benchmarks/post_graph.py

@@ -103,7 +103,7 @@ def svm_models():
             tick.label.set_fontsize(8)
         for tick in ax.yaxis.get_major_ticks():
             tick.label.set_fontsize(8)
-        pos += 1
+        pos += 1  # noqa: SIM113
 
     fig.tight_layout()
     fig.savefig("svm_model.png", dpi=1000)

docs/conf.py

@@ -1,6 +1,4 @@
 # SPDX-License-Identifier: Apache-2.0
-# -*- coding: utf-8 -*-
-#
 # Configuration file for the Sphinx documentation builder.
 
 import os
@@ -10,7 +8,7 @@
 import skl2onnx
 
 sys.path.append(os.path.abspath("exts"))
-from github_link import make_linkcode_resolve  # noqa
+from github_link import make_linkcode_resolve
 
 
 # -- Project information -----------------------------------------------------

docs/examples/plot_backend.py

@@ -17,6 +17,7 @@
 Let's use the API to compute the prediction
 of a simple logistic regression model.
 """
+
 import skl2onnx
 import onnxruntime
 import onnx

docs/examples/plot_benchmark_cdist.py

@@ -18,6 +18,7 @@
 generated/scipy.spatial.distance.cdist.html>`_
 function computes pairwise distances.
 """
+
 from pprint import pprint
 from timeit import Timer
 import numpy as np

docs/examples/plot_benchmark_pipeline.py

@@ -20,6 +20,7 @@
 does not handle string type. This cannot be part of the final ONNX pipeline
 and must be removed. Look for comment starting with ``---`` below.
 """
+
 import skl2onnx
 import onnx
 import sklearn
@@ -105,7 +106,7 @@
 
 pipe.predict_proba(X_digits[:2])
 
-for i, step in enumerate(steps):
+for _i, step in enumerate(steps):
     onnx_step = step["onnx_step"]
     sess = rt.InferenceSession(
         onnx_step.SerializeToString(), providers=["CPUExecutionProvider"]

docs/examples/plot_black_op.py

@@ -19,6 +19,7 @@
 The first converter to change its behaviour depending on a black list
 of operators is for model *GaussianMixture*.
 """
+
 import onnxruntime
 import onnx
 import numpy
@@ -167,11 +168,11 @@
 #################################
 # **Versions used for this example**
 
-import sklearn  # noqa
+import sklearn
 
 print("numpy:", numpy.__version__)
 print("scikit-learn:", sklearn.__version__)
-import skl2onnx  # noqa
+import skl2onnx
 
 print("onnx: ", onnx.__version__)
 print("onnxruntime: ", onnxruntime.__version__)

docs/examples/plot_cast_transformer.py

@@ -28,6 +28,7 @@
 based on the assumption ``(x / y)`` is usually different from
 ``x * ( 1 / y)`` on a computer.
 """
+
 import onnxruntime
 import onnx
 import os
@@ -169,11 +170,11 @@
 #################################
 # **Versions used for this example**
 
-import sklearn  # noqa
+import sklearn
 
 print("numpy:", np.__version__)
 print("scikit-learn:", sklearn.__version__)
-import skl2onnx  # noqa
+import skl2onnx
 
 print("onnx: ", onnx.__version__)
 print("onnxruntime: ", onnxruntime.__version__)

docs/examples/plot_complex_pipeline.py

@@ -29,6 +29,7 @@
 does not handle string type. This cannot be part of the final ONNX pipeline
 and must be removed. Look for comment starting with ``---`` below.
 """
+
 import os
 import pprint
 import pandas as pd

docs/examples/plot_convert_decision_function.py

@@ -19,6 +19,7 @@
 ++++++++++++++++++++++++++++
 
 """
+
 import numpy
 import sklearn
 from sklearn.datasets import load_iris