Fix conversion of Booster for xgboost>=1.6.1 (onnx#567)

* fix base_score for binary classification Signed-off-by: xadupre <[email protected]> * Update for xgboost 1.6.1 Signed-off-by: xadupre <[email protected]> * update ci Signed-off-by: xadupre <[email protected]> * fix for catboost Signed-off-by: xadupre <[email protected]> * fix for svm Signed-off-by: xadupre <[email protected]> * lint Signed-off-by: xadupre <[email protected]>
xadupre · Jun 30, 2022 · d7db0ff · d7db0ff
1 parent a699341
commit d7db0ff
Show file tree

Hide file tree

Showing 6 changed files with 95 additions and 44 deletions.
diff --git a/.azure-pipelines/linux-conda-CI.yml b/.azure-pipelines/linux-conda-CI.yml
@@ -15,11 +15,19 @@ jobs:
   strategy:
     matrix:
 
-      Python39-1120-RT1110:
+      Python39-1120-RT1110-xgb161:
         python.version: '3.9'
         ONNX_PATH: 'onnx==1.12.0' #'-i https://test.pypi.org/simple/ onnx==1.12.0rc4'
         ONNXRT_PATH: onnxruntime==1.11.0 #'-i https://test.pypi.org/simple/ ort-nightly==1.11.0.dev20220311003'
         COREML_PATH: git+https://github.com/apple/[email protected]
+        xgboost.version: '>=1.6.1'
+
+      Python39-1120-RT1110-xgb142:
+        python.version: '3.9'
+        ONNX_PATH: 'onnx==1.12.0' #'-i https://test.pypi.org/simple/ onnx==1.12.0rc4'
+        ONNXRT_PATH: onnxruntime==1.11.0 #'-i https://test.pypi.org/simple/ ort-nightly==1.11.0.dev20220311003'
+        COREML_PATH: git+https://github.com/apple/[email protected]
+        xgboost.version: '==1.4.2'
 
       Python39-1110-RT1110:
         python.version: '3.9'
@@ -126,7 +134,10 @@ jobs:
       export PYTHONPATH=.
       python -c "import onnxruntime;print('onnx:',onnx.__version__)"
       python -c "import onnxconverter_common;print('cc:',onnxconverter_common.__version__)"
+      python -c "import onnx;print('onnx:',onnx.__version__)"
       python -c "import onnxruntime;print('ort:',onnxruntime.__version__)"
+      python -c "import xgboost;print('xgboost:',xgboost.__version__)"
+      python -c "import lightgbm;print('lightgbm:',lightgbm.__version__)"
     displayName: 'version'
 
   - script: |

diff --git a/.azure-pipelines/win32-conda-CI.yml b/.azure-pipelines/win32-conda-CI.yml
@@ -122,7 +122,10 @@ jobs:
       python -m pip install -e .
       export PYTHONPATH=.
       python -c "import onnxconverter_common;print(onnxconverter_common.__version__)"
+      python -c "import onnx;print(onnx.__version__)"
       python -c "import onnxruntime;print(onnxruntime.__version__)"
+      python -c "import xgboost;print(xgboost.__version__)"
+      python -c "import lightgbm;print(lightgbm.__version__)"
     displayName: 'version'
 
   - script: |

diff --git a/onnxmltools/convert/xgboost/_parse.py b/onnxmltools/convert/xgboost/_parse.py
@@ -2,8 +2,10 @@
 
 import json
 import re
+import pprint
+from packaging.version import Version
 import numpy as np
-from xgboost import XGBRegressor, XGBClassifier
+from xgboost import XGBRegressor, XGBClassifier, __version__
 from onnxconverter_common.data_types import FloatTensorType
 from ..common._container import XGBoostModelContainer
 from ..common._topology import Topology
@@ -27,23 +29,36 @@ def _append_covers(node):
 
 
 def _get_attributes(booster):
-    # num_class
-    state = booster.__getstate__()
-    bstate = bytes(state['handle'])
-    reg = re.compile(b'("tree_info":\\[[0-9,]*\\])')
-    objs = list(set(reg.findall(bstate)))
-    assert len(objs) == 1, 'Missing required property "tree_info".'
-    tree_info = json.loads("{{{}}}".format(objs[0].decode('ascii')))['tree_info']
-    num_class = len(set(tree_info))
-
     atts = booster.attributes()
     dp = booster.get_dump(dump_format='json', with_stats=True)
     res = [json.loads(d) for d in dp]
-    trees = len(res)
-    try:
+
+    # num_class
+    if Version(__version__) < Version('1.5'):
+        state = booster.__getstate__()
+        bstate = bytes(state['handle'])
+        reg = re.compile(b'("tree_info":\\[[0-9,]*\\])')
+        objs = list(set(reg.findall(bstate)))
+        if len(objs) != 1:
+            raise RuntimeError(
+                "Unable to retrieve the tree coefficients from\n%s"
+                "" % bstate.decode("ascii", errors="ignore"))
+        tree_info = json.loads("{{{}}}".format(objs[0].decode('ascii')))['tree_info']
+        num_class = len(set(tree_info))
+        trees = len(res)
+        try:
+            ntrees = booster.best_ntree_limit
+        except AttributeError:
+            ntrees = trees // num_class if num_class > 0 else trees
+    else:
+        trees = len(res)
         ntrees = booster.best_ntree_limit
-    except AttributeError:
-        ntrees = trees // num_class if num_class > 0 else trees
+        num_class = trees // ntrees
+        if num_class == 0:
+            raise RuntimeError(
+                "Unable to retrieve the number of classes, trees=%d, ntrees=%d." % (
+                    trees, ntrees))
+
     kwargs = atts.copy()
     kwargs['feature_names'] = booster.feature_names
     kwargs['n_estimators'] = ntrees
@@ -62,14 +77,23 @@ def _get_attributes(booster):
         # classification
         kwargs['num_class'] = num_class
         if num_class != 1:
-            reg = re.compile(b'(multi:[a-z]{1,15})')
-            objs = list(set(reg.findall(bstate)))
-            if len(objs) == 1:
-                kwargs["objective"] = objs[0].decode('ascii')
+            if Version(__version__) < Version('1.5'):
+                reg = re.compile(b'(multi:[a-z]{1,15})')
+                objs = list(set(reg.findall(bstate)))
+                if len(objs) == 1:
+                    kwargs["objective"] = objs[0].decode('ascii')
+                else:
+                    raise RuntimeError(
+                        "Unable to guess objective in %r (trees=%r, ntrees=%r, num_class=%r)"
+                        "." % (objs, trees, ntrees, kwargs['num_class']))
             else:
-                raise RuntimeError(
-                    "Unable to guess objective in %r (trees=%r, ntrees=%r, num_class=%r)"
-                    "." % (objs, trees, ntrees, kwargs['num_class']))
+                att = json.loads(booster.save_config())
+                kwargs["objective"] = att['learner']['objective']['name']
+                nc = int(att['learner']['learner_model_param']['num_class'])
+                if nc != num_class:
+                    raise RuntimeError(
+                        "Mismatched value %r != %r from\n%s" % (
+                            nc, num_class, pprint.pformat(att)))
         else:
             kwargs["objective"] = "binary:logistic"
 

diff --git a/onnxmltools/utils/tests_helper.py b/onnxmltools/utils/tests_helper.py
@@ -113,7 +113,18 @@ def dump_data_and_model(data, model, onnx=None, basename="model", folder=None,
                 prediction = [model.predict(datax)]
         elif hasattr(model, "predict_proba"):
             # Classifier
-            prediction = [model.predict(data), model.predict_proba(data)]
+            if hasattr(model, 'get_params'):
+                params = model.get_params()
+                if 'objective' in params:
+                    objective = params['objective']
+                    if objective == "multi:softmax":
+                        prediction = [model.predict(data)]
+                    else:
+                        prediction = [model.predict(data), model.predict_proba(data)]
+                else:
+                    prediction = [model.predict(data), model.predict_proba(data)]
+            else:
+                prediction = [model.predict(data), model.predict_proba(data)]
         elif hasattr(model, "predict_with_probabilities"):
             # Classifier that returns all in one go
             prediction = model.predict_with_probabilities(data)

diff --git a/onnxmltools/utils/utils_backend.py b/onnxmltools/utils/utils_backend.py
@@ -4,7 +4,6 @@
 Helpers to test runtimes.
 """
 import os
-import sys
 import glob
 import pickle
 import packaging.version as pv
@@ -75,13 +74,9 @@ def compare_backend(backend, test, decimal=5, options=None, verbose=False, conte
     if the comparison failed.
     """
     if backend == "onnxruntime":
-        if sys.version_info[0] == 2:
-            # onnxruntime is not available on Python 2.
-            return
         from .utils_backend_onnxruntime import compare_runtime
         return compare_runtime(test, decimal, options, verbose)
-    else:
-        raise ValueError("Does not support backend '{0}'.".format(backend))
+    raise ValueError("Does not support backend '{0}'.".format(backend))
 
 
 def search_converted_models(root=None):

diff --git a/tests/xgboost/test_xgboost_converters.py b/tests/xgboost/test_xgboost_converters.py
@@ -109,24 +109,10 @@ def test_xgb_classifier_reglog(self):
             x_test, xgb, conv_model,
             basename="SklearnXGBClassifierRegLog")
 
-    def test_xgb_classifier_multi_str_labels(self):
-        xgb, x_test = _fit_classification_model(
-            XGBClassifier(n_estimators=4), 5, is_str=True)
-        conv_model = convert_xgboost(
-            xgb, initial_types=[('input', FloatTensorType(shape=[None, None]))],
-            target_opset=TARGET_OPSET)
-        self.assertTrue(conv_model is not None)
-        dump_data_and_model(
-            x_test, xgb, conv_model,
-            basename="SklearnXGBClassifierMultiStrLabels")
-
     def test_xgb_classifier_multi_discrete_int_labels(self):
         iris = load_iris()
         x = iris.data[:, :2]
         y = iris.target
-        y[y == 0] = 10
-        y[y == 1] = 20
-        y[y == 2] = -30
         x_train, x_test, y_train, _ = train_test_split(x,
                                                        y,
                                                        test_size=0.5,
@@ -241,7 +227,7 @@ def test_xgboost_10(self):
             X_test.astype(np.float32), regressor, model_onnx,
             basename="XGBBoosterRegBug")
 
-    def test_xgboost_classifier_i5450(self):
+    def test_xgboost_classifier_i5450_softmax(self):
         iris = load_iris()
         X, y = iris.data, iris.target
         X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=10)
@@ -255,6 +241,26 @@ def test_xgboost_classifier_i5450(self):
         predict_list = [1.,  20., 466.,   0.]
         predict_array = np.array(predict_list).reshape((1,-1)).astype(np.float32)
         pred_onx = sess.run([label_name], {input_name: predict_array})[0]
+        bst = clr.get_booster()
+        bst.dump_model('dump.raw.txt')
+        dump_data_and_model(
+            X_test.astype(np.float32) + 1e-5, clr, onx,
+            basename="XGBClassifierIris-Out0")
+
+    def test_xgboost_classifier_i5450(self):
+        iris = load_iris()
+        X, y = iris.data, iris.target
+        X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=10)
+        clr = XGBClassifier(objective="multi:softprob", max_depth=1, n_estimators=2)
+        clr.fit(X_train, y_train, eval_set=[(X_test, y_test)], early_stopping_rounds=40)
+        initial_type = [('float_input', FloatTensorType([None, 4]))]
+        onx = convert_xgboost(clr, initial_types=initial_type, target_opset=TARGET_OPSET)
+        sess = InferenceSession(onx.SerializeToString())
+        input_name = sess.get_inputs()[0].name
+        label_name = sess.get_outputs()[1].name
+        predict_list = [1.,  20., 466.,   0.]
+        predict_array = np.array(predict_list).reshape((1,-1)).astype(np.float32)
+        pred_onx = sess.run([label_name], {input_name: predict_array})[0]
         pred_xgboost = sessresults=clr.predict_proba(predict_array)
         bst = clr.get_booster()
         bst.dump_model('dump.raw.txt')
@@ -364,4 +370,5 @@ def test_onnxrt_python_xgbclassifier(self):
 
 
 if __name__ == "__main__":
+    # TestXGBoostModels().test_xgboost_booster_classifier_multiclass_softprob()
     unittest.main()