Fixes #234, fix lightgbm converter when a root is a leave (#235)

* Fixes #234, fixes lightgbm when the root of a tree is a leave

Author: xadupre

.azure-pipelines/win32-conda-CI.yml

@@ -19,13 +19,11 @@ jobs:
       Python35:
         python.version: '3.5'
         ONNX_PATH: onnx==1.2.3
-        KERAS: keras==2.1.6
         COREML_PATH: https://github.com/apple/coremltools/archive/v2.0.zip
 
       Python36:
         python.version: '3.6'
         ONNX_PATH: onnx==1.3.0
-        KERAS: keras
         COREML_PATH: git+https://github.com/apple/coremltools
 
     maxParallel: 3
@@ -49,7 +47,6 @@ jobs:
       pip install %COREML_PATH% %ONNX_PATH%
       pip install -r requirements-dev.txt
       echo Test onnxruntime installation... && python -c "import onnxruntime"
-      pip install %KERAS%
       REM install libsvm from github
       git clone --recursive https://github.com/cjlin1/libsvm libsvm
       copy libsvm\windows\*.dll libsvm\python

onnxmltools/convert/lightgbm/operator_converters/LightGbm.py

@@ -42,6 +42,12 @@ def _parse_tree_structure(tree_id, class_id, learning_rate, tree_structure, attr
     node_id_pool = set()
 
     node_id = _create_node_id(node_id_pool)
+
+    # The root node is a leaf node.
+    if not 'left_child' in tree_structure or not 'right_child' in tree_structure:
+        _parse_node(tree_id, class_id, node_id, node_id_pool, learning_rate, tree_structure, attrs)
+        return
+
     left_id = _create_node_id(node_id_pool)
     right_id = _create_node_id(node_id_pool)
 

onnxmltools/utils/tests_helper.py

@@ -34,8 +34,7 @@ def dump_data_and_model(data, model, onnx=None, basename="model", folder=None,
     :param backend: backend used to compare expected output and runtime output.
         Two options are currently supported: None for no test,
         `'onnxruntime'` to use module *onnxruntime*.
-    :param context: used if the model contains a custom operator such
-        as a custom Keras function...
+    :param context: used if the model contains a custom operator
     :param allow_failure: None to raise an exception if comparison fails
         for the backends, otherwise a string which is then evaluated to check
         whether or not the test can fail, example:
@@ -108,15 +107,10 @@ def dump_data_and_model(data, model, onnx=None, basename="model", folder=None,
     with open(dest, "wb") as f:
         pickle.dump(data, f)
 
-    if hasattr(model, 'save'):
-        dest = os.path.join(folder, basename + ".model.keras")
-        names.append(dest)
-        model.save(dest)
-    else:
-        dest = os.path.join(folder, basename + ".model.pkl")
-        names.append(dest)
-        with open(dest, "wb") as f:
-            pickle.dump(model, f)
+    dest = os.path.join(folder, basename + ".model.pkl")
+    names.append(dest)
+    with open(dest, "wb") as f:
+        pickle.dump(model, f)
 
     if onnx is None:
         array = numpy.array(data)
@@ -168,7 +162,7 @@ def convert_model(model, name, input_types):
     """
     Runs the appropriate conversion method.
     
-    :param model: model, *scikit-learn*, *keras*, or *coremltools* object
+    :param model: model
     :return: *onnx* model
     """
     from sklearn.base import BaseEstimator
@@ -179,13 +173,8 @@ def convert_model(model, name, input_types):
         from onnxmltools.convert import convert_sklearn
         model, prefix = convert_sklearn(model, name, input_types), "Sklearn"
     else:
-        from keras.models import Model
-        if isinstance(model, Model):
-            from onnxmltools.convert import convert_keras
-            model, prefix = convert_keras(model, name, input_types), "Keras"
-        else:
-            from onnxmltools.convert import convert_coreml
-            model, prefix = convert_coreml(model, name, input_types), "Cml"
+        from onnxmltools.convert import convert_coreml
+        model, prefix = convert_coreml(model, name, input_types), "Cml"
     if model is None:
         raise RuntimeError("Unable to convert model of type '{0}'.".format(type(model)))
     return model, prefix

tests/lightgbm/example.pkl

@@ -0,0 +1,32 @@
+# -------------------------------------------------------------------------
+# Copyright (c) Microsoft Corporation. All rights reserved.
+# Licensed under the MIT License. See License.txt in the project root for
+# license information.
+# --------------------------------------------------------------------------
+import sys
+import unittest
+import numpy
+import pickle
+import os
+from onnxmltools import convert_lightgbm
+from onnxmltools.convert.common.data_types import FloatTensorType
+from onnxmltools.utils import dump_data_and_model
+
+
+class TestLightGbmTreeEnsembleModelsPkl(unittest.TestCase):
+
+    @unittest.skipIf(sys.version_info[0] == 2, reason="pickled with Python 3, cannot unpickle with 2")
+    @unittest.skipIf(sys.platform.startswith('win'), reason="pickled on linux, may not work on windows")
+    def test_root_leave(self):
+        this = os.path.abspath(os.path.dirname(__file__))
+        for name in ["example.pkl"]:
+            with open(os.path.join(this, name), "rb") as f:
+                model = pickle.load(f)
+            X = [[0., 1.], [1., 1.], [2., 0.]]
+            X = numpy.array(X, dtype=numpy.float32)
+            model_onnx = convert_lightgbm(model.steps[1][1], 'pkl1', [('input', FloatTensorType([1, X.shape[1]]))])
+            dump_data_and_model(X, model.steps[1][1], model_onnx, basename="LightGbmPkl1")
+
+
+if __name__ == "__main__":
+    unittest.main()