refactoring of operators convert folder (#216)

hummingbird/ml/operator_converters/__init__.py

@@ -17,24 +17,24 @@ from .._utils import _Constants
 constants = _Constants(converter_constants, hummingbird_constants)
 
 # To register a converter for scikit-learn API operators, import associated modules here.
-from . import lightgbm  # noqa: E402
-from . import onnx_operator  # noqa: E402
-from . import onnxml_array_feature_extractor  # noqa: E402
-from . import onnxml_linear  # noqa: E402
-from . import onnxml_normalizer  # noqa: E402
-from . import onnxml_one_hot_encoder  # noqa: E402
-from . import onnxml_scaler  # noqa: E402
-from . import onnxml_tree_ensemble  # noqa: E402
-from . import skl_array_feature_extractor  # noqa: E402
-from . import skl_decision_tree  # noqa: E402
-from . import skl_gbdt  # noqa: E402
-from . import skl_iforest  # noqa: E402
-from . import skl_linear  # noqa: E402
-from . import skl_normalizer  # noqa: E402
-from . import skl_one_hot_encoder  # noqa: E402
-from . import skl_scaler  # noqa: E402
-from . import skl_sv  # noqa: E402
-from . import xgb  # noqa: E402
+from .onnx import onnx_operator  # noqa: E402
+from .onnx import onnxml_array_feature_extractor  # noqa: E402
+from .onnx import onnxml_linear  # noqa: E402
+from .onnx import onnxml_normalizer  # noqa: E402
+from .onnx import onnxml_one_hot_encoder  # noqa: E402
+from .onnx import onnxml_scaler  # noqa: E402
+from .onnx import onnxml_tree_ensemble  # noqa: E402
+from .sklearn import lightgbm  # noqa: E402
+from .sklearn import skl_array_feature_extractor  # noqa: E402
+from .sklearn import skl_decision_tree  # noqa: E402
+from .sklearn import skl_gbdt  # noqa: E402
+from .sklearn import skl_iforest  # noqa: E402
+from .sklearn import skl_linear  # noqa: E402
+from .sklearn import skl_normalizer  # noqa: E402
+from .sklearn import skl_one_hot_encoder  # noqa: E402
+from .sklearn import skl_scaler  # noqa: E402
+from .sklearn import skl_sv  # noqa: E402
+from .sklearn import xgb  # noqa: E402
 
 
 __pdoc__ = {}

hummingbird/ml/operator_converters/onnx/__init__.py

@@ -0,0 +1,9 @@
+# -------------------------------------------------------------------------
+# Copyright (c) Microsoft Corporation. All rights reserved.
+# Licensed under the MIT License. See License.txt in the project root for
+# license information.
+# --------------------------------------------------------------------------
+
+"""
+All onnx operators converters are stored under this package.
+"""

hummingbird/ml/operator_converters/onnx/onnx_operator.py

@@ -9,12 +9,11 @@ Converters for ONNX operators.
 """
 
 import numpy as np
+from onnxconverter_common.registration import register_converter
 import torch
 
-from onnxconverter_common.registration import register_converter
-
-from . import constants
-from ._base_operator import BaseOperator
+from .. import constants
+from .._base_operator import BaseOperator
 
 
 class Cast(BaseOperator, torch.nn.Module):

hummingbird/ml/operator_converters/onnx/onnxml_array_feature_extractor.py

@@ -3,12 +3,15 @@
 # Licensed under the MIT License. See License.txt in the project root for
 # license information.
 # --------------------------------------------------------------------------
-import torch
+
+"""
+Converter for ONNX-ML Array Feature Extractor.
+"""
+
 from onnxconverter_common.registration import register_converter
 
-from ._base_operator import BaseOperator
-from . import constants
-from ._array_feature_extractor_implementations import ArrayFeatureExtractor
+from .. import constants
+from .._array_feature_extractor_implementations import ArrayFeatureExtractor
 
 
 def convert_onnx_array_feature_extractor(operator, device, extra_config):

hummingbird/ml/operator_converters/onnx/onnxml_linear.py

@@ -11,7 +11,7 @@ Converters for ONNX-ML linear models.
 import numpy as np
 from onnxconverter_common.registration import register_converter
 
-from ._linear_implementations import LinearModel
+from .._linear_implementations import LinearModel
 
 
 def convert_onnx_linear_model(operator, device=None, extra_config={}):

hummingbird/ml/operator_converters/onnx/onnxml_normalizer.py

@@ -3,11 +3,14 @@
 # Licensed under the MIT License. See License.txt in the project root for
 # license information.
 # --------------------------------------------------------------------------
-import torch
+
 from onnxconverter_common.registration import register_converter
 
-from ._base_operator import BaseOperator
-from ._normalizer_implementations import Normalizer
+from .._normalizer_implementations import Normalizer
+
+"""
+Converter for ONNX-ML Normalizer.
+"""
 
 
 def convert_onnx_normalizer(operator, device=None, extra_config={}):

hummingbird/ml/operator_converters/onnx/onnxml_one_hot_encoder.py

@@ -3,12 +3,16 @@
 # Licensed under the MIT License. See License.txt in the project root for
 # license information.
 # --------------------------------------------------------------------------
+
+"""
+Converter for ONNX-ML One Hot Encoder.
+"""
+
 import numpy as np
 from onnxconverter_common.registration import register_converter
-import torch
 
-from ._base_operator import BaseOperator
-from ._one_hot_encoder_implementations import OneHotEncoderString, OneHotEncoder
+
+from .._one_hot_encoder_implementations import OneHotEncoderString, OneHotEncoder
 
 
 def convert_onnx_one_hot_encoder(operator, device=None, extra_config={}):

hummingbird/ml/operator_converters/onnx/onnxml_scaler.py

@@ -3,12 +3,16 @@
 # Licensed under the MIT License. See License.txt in the project root for
 # license information.
 # --------------------------------------------------------------------------
+
+"""
+Converter for ONNX-ML Scaler.
+"""
+
 import numpy as np
 from onnxconverter_common.registration import register_converter
-import torch
 
-from ._base_operator import BaseOperator
-from ._scaler_implementations import Scaler
+from .._base_operator import BaseOperator
+from .._scaler_implementations import Scaler
 
 
 def convert_onnx_scaler(operator, device=None, extra_config={}):

hummingbird/ml/operator_converters/onnx/onnxml_tree_ensemble.py

@@ -11,9 +11,9 @@ Converters for ONNX-ML tree-ensemble models.
 import numpy as np
 from onnxconverter_common.registration import register_converter
 
-from . import constants
-from ._gbdt_commons import convert_gbdt_classifier_common, convert_gbdt_common
-from ._tree_commons import TreeParameters
+from .. import constants
+from .._gbdt_commons import convert_gbdt_classifier_common, convert_gbdt_common
+from .._tree_commons import TreeParameters
 
 
 def _get_tree_infos_from_onnx_ml_operator(model):

hummingbird/ml/operator_converters/sklearn/__init__.py

@@ -0,0 +1,9 @@
+# -------------------------------------------------------------------------
+# Copyright (c) Microsoft Corporation. All rights reserved.
+# Licensed under the MIT License. See License.txt in the project root for
+# license information.
+# --------------------------------------------------------------------------
+
+"""
+All scikit-learn operators converters are stored under this package.
+"""

hummingbird/ml/operator_converters/sklearn/lightgbm.py

@@ -11,8 +11,8 @@ Converters for LightGBM models.
 import numpy as np
 from onnxconverter_common.registration import register_converter
 
-from ._gbdt_commons import convert_gbdt_classifier_common, convert_gbdt_common
-from ._tree_commons import TreeParameters
+from .._gbdt_commons import convert_gbdt_classifier_common, convert_gbdt_common
+from .._tree_commons import TreeParameters
 
 
 def _tree_traversal(node, lefts, rights, features, thresholds, values, count):

hummingbird/ml/operator_converters/sklearn/skl_array_feature_extractor.py

@@ -5,15 +5,13 @@
 # --------------------------------------------------------------------------
 
 """
-Converters for scikit-learn feature selectors: SelectKBest, SelectPercentile, VarianceThreshold
+Converters for scikit-learn feature selectors: SelectKBest, SelectPercentile, VarianceThreshold.
 """
 
-
 import numpy as np
 from onnxconverter_common.registration import register_converter
-import torch
 
-from ._array_feature_extractor_implementations import ArrayFeatureExtractor
+from .._array_feature_extractor_implementations import ArrayFeatureExtractor
 
 
 def convert_sklearn_select_k_best(operator, device, extra_config):

hummingbird/ml/operator_converters/sklearn/skl_decision_tree.py

@@ -9,12 +9,10 @@ Converters for scikit-learn decision-tree-based models: DecisionTree, RandomFore
 """
 
 import copy
-
-import torch
 from onnxconverter_common.registration import register_converter
 
-from ._tree_commons import get_parameters_for_sklearn_common, get_parameters_for_tree_trav_sklearn
-from ._tree_commons import convert_decision_ensemble_tree_common
+from .._tree_commons import get_parameters_for_sklearn_common, get_parameters_for_tree_trav_sklearn
+from .._tree_commons import convert_decision_ensemble_tree_common
 
 
 def convert_sklearn_random_forest_classifier(operator, device, extra_config):

hummingbird/ml/operator_converters/sklearn/skl_gbdt.py

@@ -11,9 +11,9 @@ Converters for Sklearn's GradientBoosting models.
 import numpy as np
 from onnxconverter_common.registration import register_converter
 
-from . import constants
-from ._gbdt_commons import convert_gbdt_common, convert_gbdt_classifier_common
-from ._tree_commons import get_parameters_for_sklearn_common, get_parameters_for_tree_trav_sklearn, TreeParameters
+from .. import constants
+from .._gbdt_commons import convert_gbdt_common, convert_gbdt_classifier_common
+from .._tree_commons import get_parameters_for_sklearn_common, get_parameters_for_tree_trav_sklearn, TreeParameters
 
 
 def _get_parameters_hist_gbdt(trees):

hummingbird/ml/operator_converters/sklearn/skl_iforest.py

@@ -7,13 +7,18 @@
 """
 Converters for scikit-learn isolation forest.
 """
+
 import numpy as np
 from onnxconverter_common.registration import register_converter
 
-from . import constants
-from ._tree_commons import get_parameters_for_sklearn_common, \
-    get_parameters_for_tree_trav_sklearn, get_tree_params_and_type, get_parameters_for_gemm_common
-from ._tree_implementations import TreeImpl, GEMMTreeImpl, TreeTraversalTreeImpl, PerfectTreeTraversalTreeImpl
+from .. import constants
+from .._tree_commons import (
+    get_parameters_for_sklearn_common,
+    get_parameters_for_tree_trav_sklearn,
+    get_tree_params_and_type,
+    get_parameters_for_gemm_common,
+)
+from .._tree_implementations import TreeImpl, GEMMTreeImpl, TreeTraversalTreeImpl, PerfectTreeTraversalTreeImpl
 
 
 def _average_path_length(n_samples_leaf):
@@ -37,8 +42,8 @@ def _average_path_length(n_samples_leaf):
     mask_2 = n_samples_leaf == 2
     not_mask = ~np.logical_or(mask_1, mask_2)
 
-    average_path_length[mask_1] = 0.
-    average_path_length[mask_2] = 1.
+    average_path_length[mask_1] = 0.0
+    average_path_length[mask_2] = 1.0
     average_path_length[not_mask] = (
         2.0 * (np.log(n_samples_leaf[not_mask] - 1.0) + np.euler_gamma)
         - 2.0 * (n_samples_leaf[not_mask] - 1.0) / n_samples_leaf[not_mask]
@@ -81,12 +86,9 @@ def _get_parameters_for_sklearn_iforest(tree_infos):
         The tree parameters wrapped into an instance of `operator_converters._tree_commons_TreeParameters`
     """
     tree_parameters = get_parameters_for_sklearn_common(tree_infos)
-    tree_parameters.values = (
-        _get_iforest_anomaly_score_per_node(tree_parameters.lefts,
-                                            tree_parameters.rights,
-                                            tree_infos.tree_.n_node_samples)
-        .reshape(tree_parameters.values.shape)
-    )
+    tree_parameters.values = _get_iforest_anomaly_score_per_node(
+        tree_parameters.lefts, tree_parameters.rights, tree_infos.tree_.n_node_samples
+    ).reshape(tree_parameters.values.shape)
     return tree_parameters
 
 
@@ -119,7 +121,7 @@ class GEMMIsolationForestImpl(GEMMTreeImpl):
         if self.final_probability_divider > 1:
             output = output / self.final_probability_divider
         # Further normalize to match "decision_function" in sklearn implementation.
-        output = -1.0 * 2**(-output / self.average_path_length) - self.offset
+        output = -1.0 * 2 ** (-output / self.average_path_length) - self.offset
         return output
 
 
@@ -154,7 +156,7 @@ class TreeTraversalIsolationForestImpl(TreeTraversalTreeImpl):
         if self.final_probability_divider > 1:
             output = output / self.final_probability_divider
         # Further normalize to match "decision_function" in sklearn implementation.
-        output = -1.0 * 2**(-output / self.average_path_length) - self.offset
+        output = -1.0 * 2 ** (-output / self.average_path_length) - self.offset
         return output
 
 
@@ -189,7 +191,7 @@ class PerfectTreeTraversalIsolationForestImpl(PerfectTreeTraversalTreeImpl):
         if self.final_probability_divider > 1:
             output = output / self.final_probability_divider
         # Further normalize to match "decision_function" in sklearn implementation.
-        output = -1.0 * 2**(-output / self.average_path_length) - self.offset
+        output = -1.0 * 2 ** (-output / self.average_path_length) - self.offset
         return output
 
 
@@ -217,15 +219,15 @@ def convert_sklearn_isolation_forest(operator, device, extra_config):
     # Predict in isolation forest sklearn implementation produce 2 classes: -1 (normal) & 1 (anomaly).
     classes = [-1, 1]
 
-    tree_parameters, max_depth, tree_type = get_tree_params_and_type(tree_infos, _get_parameters_for_sklearn_iforest,
-                                                                     extra_config)
+    tree_parameters, max_depth, tree_type = get_tree_params_and_type(
+        tree_infos, _get_parameters_for_sklearn_iforest, extra_config
+    )
 
     # Generate the tree implementation based on the selected strategy.
     if tree_type == TreeImpl.gemm:
         net_parameters = [
             get_parameters_for_gemm_common(
-                tree_param.lefts, tree_param.rights, tree_param.features, tree_param.thresholds, tree_param.values,
-                n_features
+                tree_param.lefts, tree_param.rights, tree_param.features, tree_param.thresholds, tree_param.values, n_features
             )
             for tree_param in tree_parameters
         ]
@@ -238,11 +240,11 @@ def convert_sklearn_isolation_forest(operator, device, extra_config):
         for tree_param in tree_parameters
     ]
     if tree_type == TreeImpl.tree_trav:
-        return TreeTraversalIsolationForestImpl(net_parameters, max_depth, n_features, classes,
-                                                extra_config=extra_config)
+        return TreeTraversalIsolationForestImpl(net_parameters, max_depth, n_features, classes, extra_config=extra_config)
     else:  # Remaining possible case: tree_type == TreeImpl.perf_tree_trav
-        return PerfectTreeTraversalIsolationForestImpl(net_parameters, max_depth, n_features, classes,
-                                                       extra_config=extra_config)
+        return PerfectTreeTraversalIsolationForestImpl(
+            net_parameters, max_depth, n_features, classes, extra_config=extra_config
+        )
 
 
 # Register the converters.

hummingbird/ml/operator_converters/sklearn/skl_linear.py

@@ -4,15 +4,14 @@
 # license information.
 # --------------------------------------------------------------------------
 
-import torch
+"""
+Converters for scikit-learn linear models: LinearRegression, LogisticRegression, LinearSVC, SGDClassifier, LogisticRegressionCV.
+"""
+
 import numpy as np
 from onnxconverter_common.registration import register_converter
-from ._linear_implementations import LinearModel
 
-
-"""
-Converters for scikit-learn linear models: LinearRegression, LogisticRegression, LinearSVC, SGDClassifier, LogisticRegressionCV
-"""
+from .._linear_implementations import LinearModel
 
 
 def convert_sklearn_linear_model(operator, device, extra_config):

hummingbird/ml/operator_converters/sklearn/skl_normalizer.py

@@ -3,11 +3,14 @@
 # Licensed under the MIT License. See License.txt in the project root for
 # license information.
 # --------------------------------------------------------------------------
-import torch
+
+"""
+Converter for scikit-learn Normalizer.
+"""
+
 from onnxconverter_common.registration import register_converter
 
-from ._base_operator import BaseOperator
-from ._normalizer_implementations import Normalizer
+from .._normalizer_implementations import Normalizer
 
 
 def convert_sklearn_normalizer(operator, device, extra_config):

hummingbird/ml/operator_converters/sklearn/skl_one_hot_encoder.py

@@ -4,12 +4,14 @@
 # license information.
 # --------------------------------------------------------------------------
 
+"""
+Converter for scikit-learn one hot encoder.
+"""
+
 import numpy as np
 from onnxconverter_common.registration import register_converter
-import torch
 
-from ._base_operator import BaseOperator
-from ._one_hot_encoder_implementations import OneHotEncoderString, OneHotEncoder
+from .._one_hot_encoder_implementations import OneHotEncoderString, OneHotEncoder
 
 
 def convert_sklearn_one_hot_encoder(operator, device, extra_config):

hummingbird/ml/operator_converters/sklearn/skl_scaler.py

@@ -4,12 +4,14 @@
 # license information.
 # --------------------------------------------------------------------------
 
-import torch
+"""
+Converters for scikit-learn scalers: RobustScaler, MaxAbsScaler, MinMaxScaler, StandardScaler.
+"""
+
 import numpy as np
 from onnxconverter_common.registration import register_converter
 
-from ._base_operator import BaseOperator
-from ._scaler_implementations import Scaler
+from .._scaler_implementations import Scaler
 
 
 def convert_sklearn_robust_scaler(operator, device, extra_config):

hummingbird/ml/operator_converters/sklearn/skl_sv.py

@@ -4,16 +4,16 @@
 # license information.
 # --------------------------------------------------------------------------
 
-import torch
-import scipy
+"""
+Converters for scikit-learn SV models: SVC, NuSVC.  (LinearSVC is covered by linear_classifier.py).
+"""
+
 import numpy as np
+import scipy
+import torch
 from onnxconverter_common.registration import register_converter
 
-from ._base_operator import BaseOperator
-
-"""
-Converters for scikit-learn linear models: SVC, NuSVC.  (LinearSVC is covered by linear_classifier.py)
-"""
+from .._base_operator import BaseOperator
 
 
 class SVC(BaseOperator, torch.nn.Module):

hummingbird/ml/operator_converters/sklearn/xgb.py

@@ -11,9 +11,9 @@ Converters for XGBoost models.
 import numpy as np
 from onnxconverter_common.registration import register_converter
 
-from . import constants
-from ._gbdt_commons import convert_gbdt_classifier_common, convert_gbdt_common
-from ._tree_commons import TreeParameters
+from .. import constants
+from .._gbdt_commons import convert_gbdt_classifier_common, convert_gbdt_common
+from .._tree_commons import TreeParameters
 
 
 def _tree_traversal(tree_info, lefts, rights, features, thresholds, values):

tests/tree_utils.py

@@ -14,7 +14,7 @@ dt_implementation_map = {
 }
 
 iforest_implementation_map = {
-    "tree_trav": "<class 'hummingbird.ml.operator_converters.skl_iforest.TreeTraversalIsolationForestImpl'>",
-    "perf_tree_trav": "<class 'hummingbird.ml.operator_converters.skl_iforest.PerfectTreeTraversalIsolationForestImpl'>",
-    "gemm": "<class 'hummingbird.ml.operator_converters.skl_iforest.GEMMIsolationForestImpl'>",
+    "tree_trav": "<class 'hummingbird.ml.operator_converters.sklearn.skl_iforest.TreeTraversalIsolationForestImpl'>",
+    "perf_tree_trav": "<class 'hummingbird.ml.operator_converters.sklearn.skl_iforest.PerfectTreeTraversalIsolationForestImpl'>",
+    "gemm": "<class 'hummingbird.ml.operator_converters.sklearn.skl_iforest.GEMMIsolationForestImpl'>",
 }