ENH: dpnp interop for sklearnex

.ci/pipeline/build-and-test-lnx.yml

@@ -47,6 +47,7 @@ steps:
       bash .ci/scripts/setup_sklearn.sh $(SKLEARN_VERSION)
       pip install --upgrade -r requirements-test.txt -r requirements-test-optional.txt
       pip install $(python .ci/scripts/get_compatible_scipy_version.py)
+      if [ $(echo $(PYTHON_VERSION) | grep '3.8\|3.9\|3.10') ]; then conda install -q -y -c intel dpnp; fi
       pip list
     displayName: 'Install testing requirements'
   - script: |

examples/sklearnex/knn_bf_classification_dpnp_batch.py

@@ -0,0 +1,58 @@
+# ===============================================================================
+# Copyright 2023 Intel Corporation
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ===============================================================================
+
+# sklearnex kNN example for GPU offloading with DPNP ndarray:
+#    python ./knn_bf_classification_dpnp_batch.py.py
+
+import dpctl
+import dpnp
+import numpy as np
+from sklearn.datasets import make_classification
+from sklearn.metrics import accuracy_score
+from sklearn.model_selection import train_test_split
+
+from sklearnex.neighbors import KNeighborsClassifier
+
+X, y = make_classification(
+    n_samples=1000,
+    n_features=4,
+    n_informative=2,
+    n_redundant=0,
+    random_state=0,
+    shuffle=False,
+)
+
+X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.33, random_state=42)
+
+# Make sure that all DPNP ndarrays using the same device.
+q = dpctl.SyclQueue("gpu")  # GPU
+
+dpnp_X_train = dpnp.asarray(X_train, usm_type="device", sycl_queue=q)
+dpnp_y_train = dpnp.asarray(y_train, usm_type="device", sycl_queue=q)
+dpnp_X_test = dpnp.asarray(X_test, usm_type="device", sycl_queue=q)
+
+knn_mdl = KNeighborsClassifier(
+    algorithm="brute", n_neighbors=20, weights="uniform", p=2, metric="minkowski"
+)
+knn_mdl.fit(dpnp_X_train, dpnp_y_train)
+
+y_predict = knn_mdl.predict(dpnp_X_test)
+
+print("Brute Force Distributed kNN classification results:")
+print("Ground truth (first 5 observations):\n{}".format(y_test[:5]))
+print("Classification results (first 5 observations):\n{}".format(y_predict[:5]))
+print("Accuracy (2 classes): {}\n".format(accuracy_score(y_test, y_predict.asnumpy())))
+print("Are predicted results on GPU: {}".format(y_predict.sycl_device.is_gpu))

examples/sklearnex/random_forest_classifier_dpctl_batch.py

@@ -0,0 +1,53 @@
+# ===============================================================================
+# Copyright 2023 Intel Corporation
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ===============================================================================
+
+# sklearnex RF example for GPU offloading with DPCtl tensor:
+#    python ./random_forest_classifier_dpctl_batch.py
+
+import dpctl
+import dpctl.tensor as dpt
+import numpy as np
+from sklearn.datasets import make_classification
+from sklearn.model_selection import train_test_split
+
+from sklearnex.preview.ensemble import RandomForestClassifier
+
+# Make sure that all DPCtl tensors using the same device.
+q = dpctl.SyclQueue("gpu")  # GPU
+
+X, y = make_classification(
+    n_samples=1000,
+    n_features=4,
+    n_informative=2,
+    n_redundant=0,
+    random_state=0,
+    shuffle=False,
+)
+
+X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.33, random_state=42)
+
+dpt_X_train = dpt.asarray(X_train, usm_type="device", sycl_queue=q)
+dpt_y_train = dpt.asarray(y_train, usm_type="device", sycl_queue=q)
+dpt_X_test = dpt.asarray(X_test, usm_type="device", sycl_queue=q)
+
+rf = RandomForestClassifier(max_depth=2, random_state=0).fit(dpt_X_train, dpt_y_train)
+
+pred = rf.predict(dpt_X_test)
+
+print("Random Forest classification results:")
+print("Ground truth (first 5 observations):\n{}".format(y_test[:5]))
+print("Classification results (first 5 observations):\n{}".format(pred[:5]))
+print("Are predicted results on GPU: {}".format(pred.sycl_device.is_gpu))

examples/sklearnex/random_forest_regressor_dpnp_batch.py

@@ -0,0 +1,46 @@
+# ===============================================================================
+# Copyright 2023 Intel Corporation
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ===============================================================================
+
+# sklearnex RF example for GPU offloading with DPNP ndarray:
+#    python ./random_forest_regressor_dpnp_batch.py.py
+
+import dpnp
+import numpy as np
+from sklearn.datasets import make_regression
+from sklearn.model_selection import train_test_split
+
+from sklearnex.preview.ensemble import RandomForestRegressor
+
+sycl_device = "gpu:0"
+
+X, y = make_regression(
+    n_samples=1000, n_features=4, n_informative=2, random_state=0, shuffle=False
+)
+
+X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.33, random_state=42)
+
+dpnp_X_train = dpnp.asarray(X_train, device=sycl_device)
+dpnp_y_train = dpnp.asarray(y_train, device=sycl_device)
+dpnp_X_test = dpnp.asarray(X_test, device=sycl_device)
+
+rf = RandomForestRegressor(max_depth=2, random_state=0).fit(dpnp_X_train, dpnp_y_train)
+
+pred = rf.predict(dpnp_X_test)
+
+print("Random Forest regression results:")
+print("Ground truth (first 5 observations):\n{}".format(y_test[:5]))
+print("Regression results (first 5 observations):\n{}".format(pred[:5]))
+print("Are predicted results on GPU: {}".format(pred.sycl_device.is_gpu))

sklearnex/_device_offload.py

@@ -20,9 +20,6 @@
 
 import numpy as np
 
-from ._config import get_config
-from ._utils import get_patch_message
-
 try:
     from dpctl import SyclQueue
     from dpctl.memory import MemoryUSMDevice, as_usm_memory
@@ -32,6 +29,16 @@
 except ImportError:
     dpctl_available = False
 
+try:
+    import dpnp
+
+    dpnp_available = True
+except ImportError:
+    dpnp_available = False
+
+from ._config import get_config
+from ._utils import get_patch_message
+
 oneapi_is_available = "daal4py.oneapi" in sys.modules
 if oneapi_is_available:
     from daal4py.oneapi import _get_device_name_sycl_ctxt, _get_sycl_ctxt_params
@@ -197,7 +204,9 @@ def wrapper(self, *args, **kwargs):
             usm_iface = getattr(data[0], "__sycl_usm_array_interface__", None)
         result = func(self, *args, **kwargs)
         if usm_iface is not None:
-            return _copy_to_usm(usm_iface["syclobj"], result)
+            result = _copy_to_usm(usm_iface["syclobj"], result)
+            if dpnp_available and isinstance(data[0], dpnp.ndarray):
+                result = dpnp.array(result, copy=False)
         return result
 
     return wrapper