Support feature input types other than float in function transformer (#…

…208)

* Support feature input types other than float in function transformer converter
* Remove unused import

skl2onnx/shape_calculators/function_transformer.py

@@ -4,15 +4,15 @@
 # license information.
 # --------------------------------------------------------------------------
 
+import copy
 from ..common._registration import register_shape_calculator
-from ..common.data_types import FloatTensorType
 
 
 def calculate_sklearn_function_transformer_output_shapes(operator):
-    '''
+    """
     This operator is used only to merge columns in a pipeline.
-    Only id function is supported.
-    '''
+    Only identity function is supported.
+    """
     if operator.raw_operator.func is not None:
         raise RuntimeError("FunctionTransformer is not supported unless the "
                            "transform function is None (= identity). "
@@ -27,7 +27,8 @@ def calculate_sklearn_function_transformer_output_shapes(operator):
             C = 'None'
             break
 
-    operator.outputs[0].type = FloatTensorType([N, C])
+    operator.outputs[0].type = copy.deepcopy(operator.inputs[0].type)
+    operator.outputs[0].type.shape = [N, C]
 
 
 register_shape_calculator('SklearnFunctionTransformer',

tests/test_sklearn_function_transformer_converter.py

@@ -2,8 +2,9 @@
 Tests scikit-imputer converter.
 """
 import unittest
+import numpy as np
 import pandas
-from sklearn.datasets import load_iris
+from sklearn.datasets import load_digits, load_iris
 from sklearn.pipeline import Pipeline
 
 try:
@@ -44,10 +45,10 @@ def convert_dataframe_schema(df, drop=None):
                 inputs.append((k, t))
             return inputs
 
-        data = load_iris()
+        data = load_digits()
         X = data.data[:, :2]
         y = data.target
-        data = pandas.DataFrame(X, columns=["X1", "X2"])
+        data = pandas.DataFrame(X, columns=["X1", "X2"], dtype=np.int64)
 
         pipe = Pipeline(steps=[
             (