Merge branch 'master' of https://github.com/onnx/onnxmltools into speed

README.md

@@ -35,17 +35,16 @@ If you choose to install `onnxmltools` from its source code, you must set the en
 ## Dependencies
 This package relies on ONNX, NumPy, and ProtoBuf. If you are converting a model from scikit-learn, Core ML, Keras, LightGBM, SparkML, XGBoost, H2O, CatBoost or LibSVM, you will need an environment with the respective package installed from the list below:
 1. scikit-learn
-2. CoreMLTools
+2. CoreMLTools (version 3.1 or lower)
 3. Keras (version 2.0.8 or higher) with the corresponding Tensorflow version
-4. LightGBM (scikit-learn interface)
+4. LightGBM
 5. SparkML
-6. XGBoost (scikit-learn interface)
+6. XGBoost
 7. libsvm
 8. H2O
 9. CatBoost
 
-ONNXMLTools has been tested with Python **3.5**, **3.6**, and **3.7**.
-Version 1.6.1 is the latest version supporting Python 2.7.
+ONNXMLTools is tested with Python **3.7+**.
 
 # Examples
 If you want the converted ONNX model to be compatible with a certain ONNX version, please specify the target_opset parameter upon invoking the convert function. The following Keras model conversion example demonstrates this below. You can identify the mapping from ONNX Operator Sets (referred to as opsets) to ONNX releases in the [versioning documentation](https://github.com/onnx/onnx/blob/master/docs/Versioning.md#released-versions).

docs/api_summary.rst

@@ -14,14 +14,20 @@ in *onnxmltools*.
 Converters
 ==========
 
+.. autofunction:: onnxmltools.convert.h2o.catboost
+
 .. autofunction:: onnxmltools.convert.coreml.convert
 
+.. autofunction:: onnxmltools.convert.h2o.convert
+
 .. autofunction:: onnxmltools.convert.keras.convert
 
 .. autofunction:: onnxmltools.convert.lightgbm.convert
 
 .. autofunction:: onnxmltools.convert.sklearn.convert
 
+.. autofunction:: onnxmltools.convert.xgboost.convert
+
 Utils
 =====
 

docs/examples/plot_convert_keras.py

@@ -48,6 +48,8 @@
               optimizer='sgd',
               metrics=['accuracy'])
 model.fit(X_train, y_train, epochs=5, batch_size=16)
+print("keras prediction")
+print(model.predict(X_test.astype(numpy.float32)))
 
 ###########################
 # Convert a model into ONNX
@@ -62,9 +64,10 @@
 
 sess = rt.InferenceSession(onx.SerializeToString())
 input_name = sess.get_inputs()[0].name
-label_name = sess.get_outputs()[0].name
+output_name = sess.get_outputs()[0].name
 pred_onx = sess.run(
-    [label_name], {input_name: X_test.astype(numpy.float32)})[0]
+    [output_name], {input_name: X_test.astype(numpy.float32)})[0]
+print("ONNX prediction")
 print(pred_onx)
 
 ##################################

docs/index.rst

@@ -4,23 +4,14 @@
 onnxmltools: Convert your model into ONNX
 =========================================
 
-.. list-table:
-    :header-rows: 1
-    :widths: 5 5
-    * - Linux
-      - Windows
-    * - .. image:: https://travis-ci.org/onnx/onnxmltools.svg?branch=master
-            :target: https://travis-ci.org/onnx/onnxmltools
-      - .. image:: https://ci.appveyor.com/api/projects/status/d1xav3amubypje4n?svg=true
-            :target: https://ci.appveyor.com/project/xadupre/onnxmltools
-
 ONNXMLTools enables you to convert models from different machine learning
 toolkits into `ONNX <https://onnx.ai>`_.
 Currently the following toolkits are supported:
 
 * `Apple Core ML <https://developer.apple.com/documentation/coreml>`_,
   (`onnx-coreml <https://github.com/onnx/onnx-coreml>`_ does the reverse
-  conversion from *onnx* to *Apple Core ML*),
+  conversion from *onnx* to *Apple Core ML*) (up to version 3.1)
+* `catboost <https://catboost.ai/>`_
 * `h2o <http://docs.h2o.ai/h2o/latest-stable/h2o-py/docs/intro.html>`_
   (a subset only)
 * `Keras <https://keras.io/>`_

onnxmltools/convert/common/utils.py

@@ -14,4 +14,16 @@ def hummingbird_installed():
         except ImportError:
             return False
 
+
+def tf2onnx_installed():
+    """
+    Checks that *tf2onnx* is available.
+    """
+    try:
+        import tf2onnx  # noqa F401
+        return True
+    except ImportError:
+        return False
+
+
 from onnxconverter_common.utils import *  # noqa

onnxmltools/convert/main.py

@@ -1,12 +1,15 @@
 # SPDX-License-Identifier: Apache-2.0
 
+import warnings
+from distutils.version import StrictVersion
 import onnx
 from .common import utils
-import warnings
 
 
 def convert_coreml(model, name=None, initial_types=None, doc_string='', target_opset=None,
-                   targeted_onnx=onnx.__version__, custom_conversion_functions=None, custom_shape_calculators=None):
+                   targeted_onnx=None, custom_conversion_functions=None, custom_shape_calculators=None):
+    if targeted_onnx is not None:
+        warnings.warn("targeted_onnx is deprecated. Use target_opset.", DeprecationWarning)
     if not utils.coreml_installed():
         raise RuntimeError('coremltools is not installed. Please install coremltools to use this feature.')
 
@@ -15,22 +18,93 @@ def convert_coreml(model, name=None, initial_types=None, doc_string='', target_o
                    custom_conversion_functions, custom_shape_calculators)
 
 
-def convert_keras(model, name=None, initial_types=None, doc_string='',
-                  target_opset=None, targeted_onnx=onnx.__version__,
-                  channel_first_inputs=None, custom_conversion_functions=None, custom_shape_calculators=None,
+def convert_keras(model, name=None,
+                  initial_types=None,
+                  doc_string='',
+                  target_opset=None,
+                  targeted_onnx=None,
+                  channel_first_inputs=None,
+                  custom_conversion_functions=None,
+                  custom_shape_calculators=None,
                   default_batch_size=1):
-    if not utils.keras2onnx_installed():
-        raise RuntimeError('keras2onnx is not installed. Please install it to use this feature.')
-
-    if custom_conversion_functions:
-        warnings.warn('custom_conversion_functions is not supported any more. Please set it to None.')
-
-    from keras2onnx import convert_keras as convert
-    return convert(model, name, doc_string, target_opset, channel_first_inputs)
+    """
+    .. versionchanged:: 1.9.0
+        The conversion is now using *tf2onnx*.
+    """
+    if targeted_onnx is not None:
+        warnings.warn("targeted_onnx is deprecated and unused. Use target_opset.", DeprecationWarning)
+    import tensorflow as tf
+    if StrictVersion(tf.__version__) < StrictVersion('2.0'):
+        # Former converter for tensorflow<2.0.
+        from keras2onnx import convert_keras as convert
+        return convert(model, name, doc_string, target_opset, channel_first_inputs)
+    else:
+        # For tensorflow>=2.0, new converter based on tf2onnx.
+        import tf2onnx
+
+        if not utils.tf2onnx_installed():
+            raise RuntimeError('tf2onnx is not installed. Please install it to use this feature.')
+
+        if custom_conversion_functions is not None:
+            warnings.warn('custom_conversion_functions is not supported any more. Please set it to None.')
+        if custom_shape_calculators is not None:
+            warnings.warn('custom_shape_calculators is not supported any more. Please set it to None.')
+        if default_batch_size != 1:
+            warnings.warn('default_batch_size is not supported any more. Please set it to 1.')
+        if default_batch_size != 1:
+            warnings.warn('default_batch_size is not supported any more. Please set it to 1.')
+
+        if initial_types is not None:
+            from onnxconverter_common import (
+                FloatTensorType, DoubleTensorType,
+                Int64TensorType, Int32TensorType,
+                StringTensorType, BooleanTensorType)
+            spec = []
+            for name, kind in initial_types:
+                if isinstance(kind, FloatTensorType):
+                    dtype = tf.float32
+                elif isinstance(kind, Int64TensorType):
+                    dtype = tf.int64
+                elif isinstance(kind, Int32TensorType):
+                    dtype = tf.int32
+                elif isinstance(kind, DoubleTensorType):
+                    dtype = tf.float64
+                elif isinstance(kind, StringTensorType):
+                    dtype = tf.string
+                elif isinstance(kind, BooleanTensorType):
+                    dtype = tf.bool
+                else:
+                    raise TypeError(
+                        "Unexpected type %r, cannot infer tensorflow type." % type(kind))
+                spec.append(tf.TensorSpec(tuple(kind.shape), dtype, name=name))
+            input_signature = tuple(spec)
+        else:
+            input_signature = None
+
+        model_proto, external_tensor_storage = tf2onnx.convert.from_keras(
+            model,
+            input_signature=input_signature,
+            opset=target_opset,
+            custom_ops=None,
+            custom_op_handlers=None,
+            custom_rewriter=None,
+            inputs_as_nchw=channel_first_inputs,
+            extra_opset=None,
+            shape_override=None,
+            target=None,
+            large_model=False,
+            output_path=None)
+        if external_tensor_storage is not None:
+            warnings.warn("The current API does not expose the second result 'external_tensor_storage'. "
+                          "Use tf2onnx directly to get it.")
+        model_proto.doc_string = doc_string
+        return model_proto
 
 
 def convert_libsvm(model, name=None, initial_types=None, doc_string='', target_opset=None,
-                   targeted_onnx=onnx.__version__, custom_conversion_functions=None, custom_shape_calculators=None):
+                   targeted_onnx=None, custom_conversion_functions=None, custom_shape_calculators=None):
+    if targeted_onnx is not None:
+        warnings.warn("targeted_onnx is deprecated. Use target_opset.", DeprecationWarning)
     if not utils.libsvm_installed():
         raise RuntimeError('libsvm is not installed. Please install libsvm to use this feature.')
 
@@ -51,8 +125,10 @@ def convert_catboost(model, name=None, initial_types=None, doc_string='', target
 
 
 def convert_lightgbm(model, name=None, initial_types=None, doc_string='', target_opset=None,
-                     targeted_onnx=onnx.__version__, custom_conversion_functions=None,
+                     targeted_onnx=None, custom_conversion_functions=None,
                      custom_shape_calculators=None, without_onnx_ml=False, zipmap=True):
+    if targeted_onnx is not None:
+        warnings.warn("targeted_onnx is deprecated. Use target_opset.", DeprecationWarning)
     if not utils.lightgbm_installed():
         raise RuntimeError('lightgbm is not installed. Please install lightgbm to use this feature.')
 
@@ -63,7 +139,9 @@ def convert_lightgbm(model, name=None, initial_types=None, doc_string='', target
 
 
 def convert_sklearn(model, name=None, initial_types=None, doc_string='', target_opset=None,
-                    targeted_onnx=onnx.__version__, custom_conversion_functions=None, custom_shape_calculators=None):
+                    targeted_onnx=None, custom_conversion_functions=None, custom_shape_calculators=None):
+    if targeted_onnx is not None:
+        warnings.warn("targeted_onnx is deprecated. Use target_opset.", DeprecationWarning)
     if not utils.sklearn_installed():
         raise RuntimeError('scikit-learn is not installed. Please install scikit-learn to use this feature.')
 
@@ -76,8 +154,10 @@ def convert_sklearn(model, name=None, initial_types=None, doc_string='', target_
 
 
 def convert_sparkml(model, name=None, initial_types=None, doc_string='', target_opset=None,
-                    targeted_onnx=onnx.__version__, custom_conversion_functions=None,
+                    targeted_onnx=None, custom_conversion_functions=None,
                     custom_shape_calculators=None, spark_session=None):
+    if targeted_onnx is not None:
+        warnings.warn("targeted_onnx is deprecated. Use target_opset.", DeprecationWarning)
     if not utils.sparkml_installed():
         raise RuntimeError('Spark is not installed. Please install Spark to use this feature.')
 
@@ -87,6 +167,8 @@ def convert_sparkml(model, name=None, initial_types=None, doc_string='', target_
 
 
 def convert_xgboost(*args, **kwargs):
+    if kwargs.get('targeted_onnx', None) is not None:
+        warnings.warn("targeted_onnx is deprecated. Use target_opset.", DeprecationWarning)
     if not utils.xgboost_installed():
         raise RuntimeError('xgboost is not installed. Please install xgboost to use this feature.')
 
@@ -95,6 +177,8 @@ def convert_xgboost(*args, **kwargs):
 
 
 def convert_h2o(*args, **kwargs):
+    if kwargs.get('targeted_onnx', None) is not None:
+        warnings.warn("targeted_onnx is deprecated. Use target_opset.", DeprecationWarning)
     if not utils.h2o_installed():
         raise RuntimeError('h2o is not installed. Please install h2o to use this feature.')
 

onnxmltools/convert/xgboost/_parse.py

@@ -59,10 +59,16 @@ def _get_attributes(booster):
             reg = re.compile(b'(multi:[a-z]{1,15})')
             objs = list(set(reg.findall(bstate)))
             if len(objs) != 1:
-                raise RuntimeError(
-                    "Unable to guess objective in {}.".format(objs))
-            kwargs['num_class'] = trees // ntrees
-            kwargs["objective"] = objs[0].decode('ascii')
+                if '"name":"binary:logistic"' in str(bstate):
+                    kwargs['num_class'] = 1
+                    kwargs["objective"] = "binary:logistic"
+                else:
+                    raise RuntimeError(
+                        "Unable to guess objective in %r (trees=%r, ntrees=%r)"
+                        "." % (objs, trees, ntrees))
+            else:
+                kwargs['num_class'] = trees // ntrees
+                kwargs["objective"] = objs[0].decode('ascii')
         else:
             kwargs['num_class'] = 1
             kwargs["objective"] = "binary:logistic"