Enable Double input type for RandomForest and ZipMap (#826)

* Enable Double input type for RandomForest and ZipMap

Signed-off-by: BowenBao <bowbao@microsoft.com>

* Adjust attr type based on opset; revert regressor changes

Signed-off-by: BowenBao <bowbao@microsoft.com>

* flake8

Signed-off-by: BowenBao <bowbao@microsoft.com>

skl2onnx/_parse.py

@@ -13,7 +13,8 @@
     class OutlierMixin:
         pass
 
-from sklearn.ensemble import IsolationForest, RandomTreesEmbedding
+from sklearn.ensemble import (
+    IsolationForest, RandomTreesEmbedding, RandomForestClassifier)
 from sklearn.gaussian_process import GaussianProcessRegressor
 from sklearn.linear_model import BayesianRidge
 from sklearn.model_selection import GridSearchCV
@@ -47,7 +48,8 @@ class OutlierMixin:
 from .common._topology import Topology, Variable
 from .common.data_types import (
     DictionaryType, Int64TensorType, SequenceType,
-    StringTensorType, TensorType, guess_tensor_type)
+    StringTensorType, TensorType, FloatTensorType,
+    guess_tensor_type)
 from .common.utils import get_column_indices
 from .common.utils_checking import check_signature
 from .common.utils_classifier import get_label_classes
@@ -153,8 +155,12 @@ def _parse_sklearn_simple_model(scope, model, inputs, custom_parsers=None,
         # be fixed in shape inference phase.
         label_variable = scope.declare_local_variable(
             'label', Int64TensorType())
+        if type(model) in [RandomForestClassifier]:
+            prob_dtype = FloatTensorType()
+        else:
+            prob_dtype = guess_tensor_type(inputs[0].type)
         probability_tensor_variable = scope.declare_local_variable(
-            'probabilities', guess_tensor_type(inputs[0].type))
+            'probabilities', prob_dtype)
         this_operator.outputs.append(label_variable)
         this_operator.outputs.append(probability_tensor_variable)
 

skl2onnx/operator_converters/random_forest.py

@@ -92,6 +92,7 @@ def convert_sklearn_random_forest_classifier(
     dtype = guess_numpy_type(operator.inputs[0].type)
     if dtype != np.float64:
         dtype = np.float32
+    attr_dtype = dtype if op_version >= 3 else np.float32
     op = operator.raw_operator
 
     if hasattr(op, 'n_outputs_'):
@@ -209,7 +210,7 @@ def convert_sklearn_random_forest_classifier(
                          'target_weights', 'nodes_hitrates',
                          'base_values'):
                     attr_pairs[k] = np.array(
-                        attr_pairs[k], dtype=dtype).ravel()
+                        attr_pairs[k], dtype=attr_dtype).ravel()
 
         container.add_node(
             op_type, input_name,
@@ -249,7 +250,7 @@ def convert_sklearn_random_forest_classifier(
                     if k in ('nodes_values', 'class_weights',
                              'target_weights', 'nodes_hitrates',
                              'base_values'):
-                        attrs[k] = np.array(attrs[k], dtype=dtype).ravel()
+                        attrs[k] = np.array(attrs[k], dtype=attr_dtype).ravel()
 
             if options['decision_path']:
                 # decision_path

skl2onnx/shape_calculators/zip_map.py

@@ -12,6 +12,9 @@ def calculate_sklearn_zipmap(operator):
     if len(operator.inputs) == 2:
         operator.outputs[0].type = operator.inputs[0].type.__class__(
             operator.inputs[0].type.shape)
+        if operator.outputs[1].type is not None:
+            operator.outputs[1].type.element_type.value_type = \
+                operator.inputs[1].type.__class__([])
 
 
 def calculate_sklearn_zipmap_columns(operator):

tests/test_sklearn_random_forest_converters.py

@@ -27,7 +27,9 @@
 from skl2onnx.common.data_types import (
     BooleanTensorType,
     FloatTensorType,
-    Int64TensorType)
+    DoubleTensorType,
+    Int64TensorType,
+)
 from skl2onnx import convert_sklearn, to_onnx
 from test_utils import (
     binary_array_to_string,
@@ -252,6 +254,22 @@ def test_extra_trees_classifier_bool(self):
             X, model, model_onnx,
             basename="SklearnExtraTreesClassifierBool")
 
+    @ignore_warnings(category=FutureWarning)
+    def test_random_forest_classifier_double(self):
+        model, X = fit_classification_model(
+            RandomForestClassifier(n_estimators=5, random_state=42),
+            3, is_double=True)
+        for opv in [1, 2, 3]:
+            model_onnx = convert_sklearn(
+                model, "random forest classifier",
+                [("input", DoubleTensorType([None, X.shape[1]]))],
+                target_opset={'ai.onnx.ml': opv,
+                              '': TARGET_OPSET})
+            self.assertIsNotNone(model_onnx)
+            dump_data_and_model(
+                X, model, model_onnx,
+                basename="SklearnRandomForestClassifierDouble")
+
     @ignore_warnings(category=FutureWarning)
     def common_test_model_hgb_regressor(self, add_nan=False):
         model = HistGradientBoostingRegressor(max_iter=5, max_depth=2)

tests/test_utils/tests_helper.py

@@ -51,7 +51,8 @@ def fit_classification_model(model, n_classes, is_int=False,
                              pos_features=False, label_string=False,
                              random_state=42, is_bool=False,
                              n_features=20, n_redundant=None,
-                             n_repeated=None, cls_dtype=None):
+                             n_repeated=None, cls_dtype=None,
+                             is_double=False,):
     X, y = make_classification(
         n_classes=n_classes, n_features=n_features, n_samples=250,
         random_state=random_state, n_informative=min(7, n_features),
@@ -62,6 +63,7 @@ def fit_classification_model(model, n_classes, is_int=False,
     if label_string:
         y = numpy.array(['cl%d' % cl for cl in y])
     X = X.astype(numpy.int64) if is_int or is_bool else X.astype(numpy.float32)
+    X = X.astype(numpy.double) if is_double else X
     if pos_features:
         X = numpy.abs(X)
     if is_bool: