def segmentation(self):
"""
Build a segmentation (sequence of estimators)
:return: Segmentation element
"""
# there is no notion of weighted sum, so we should take weighted average and multiply result by total weight
# in output transformation
segmentation = pmml.Segmentation(multipleModelMethod="weightedAverage")
# build the context for the nested regression models by replacing output categorical feature
# with the continuous numeric feature
regression_context = copy(self.context)
regression_context.schemas[Schema.OUTPUT] = [
RealNumericFeature(
name=self.context.schemas[Schema.OUTPUT][0].name,
namespace=Schema.NUMERIC.value)
]
# first, transform initial estimator
init_segment = pmml.Segment(weight=1)
init_segment.append(pmml.True_())
init_segment.append(
find_converter(self.estimator.init_)(self.estimator.init_,
regression_context).model())
segmentation.append(init_segment)
for est in self.estimator.estimators_[:, 0]:
s = pmml.Segment(weight=self.estimator.learning_rate)
s.append(pmml.True_())
s.append(
DecisionTreeConverter(est, regression_context,
ModelMode.REGRESSION)._model())
segmentation.append(s)
return segmentation
def segmentation(self):
"""
Build a segmentation (sequence of estimators)
:return: Segmentation element
"""
# there is no notion of weighted sum, so we should take weighted average and multiply result by total weight
# in output transformation
segmentation = pmml.Segmentation(multipleModelMethod="weightedAverage")
# build the context for the nested regression models by replacing output categorical feature
# with the continuous numeric feature
regression_context = TransformationContext(schemas=dict(self.context.schemas))
regression_context.schemas[Schema.OUTPUT] = [RealNumericFeature(
name=self.context.schemas[Schema.OUTPUT][0].name,
namespace=Schema.NUMERIC.namespace
)]
# first, transform initial estimator
init_segment = pmml.Segment(weight=1)
init_segment.append(pmml.True_())
init_segment.append(find_converter(self.estimator.init_)(self.estimator.init_, regression_context).model())
segmentation.append(init_segment)
for est in self.estimator.estimators_[:, 0]:
s = pmml.Segment(weight=self.estimator.learning_rate)
s.append(pmml.True_())
s.append(DecisionTreeConverter(est, regression_context, ModelMode.REGRESSION)._model())
segmentation.append(s)
return segmentation
def segmentation(self):
"""
Build a segmentation (sequence of estimators)
:return: Segmentation element
"""
# there is no notion of weighted sum, so we should take weighted average and multiply result by total weight
# in output transformation
segmentation = pmml.Segmentation(multipleModelMethod="weightedAverage")
# first, transform initial estimator
init_segment = pmml.Segment(weight=1)
init_segment.append(pmml.True_())
init_segment.append(
find_converter(self.estimator.init_)(self.estimator.init_,
self.context).model())
segmentation.append(init_segment)
for est in self.estimator.estimators_[:, 0]:
s = pmml.Segment(weight=self.estimator.learning_rate)
s.append(pmml.True_())
s.append(
DecisionTreeConverter(est, self.context,
self.MODE_REGRESSION)._model())
segmentation.append(s)
return segmentation
def __init__(self, estimator, context):
super(GradientBoostingConverter, self).__init__(estimator, context, self.MODE_CLASSIFICATION)
assert isinstance(estimator, GradientBoostingClassifier), \
'This converter can only process GradientBoostingClassifier instances'
assert len(context.schemas[self.SCHEMA_OUTPUT]) == 1, 'Only one-label classification is supported'
assert not estimator.loss_.is_multi_class, 'Only one-label classification is supported'
assert context.schemas[self.SCHEMA_OUTPUT][0].data_type == 'double', 'PMML version only returns probabilities'
assert context.schemas[self.SCHEMA_OUTPUT][0].optype == 'continuous', 'PMML version only returns probabilities'
assert find_converter(estimator.init_) is not None, 'Can not find a converter for {}'.format(estimator.init_)
def __init__(self, estimator, context):
super(GradientBoostingConverter, self).__init__(estimator, context)
assert isinstance(estimator, GradientBoostingClassifier), \
'This converter can only process GradientBoostingClassifier instances'
assert len(context.schemas[Schema.OUTPUT]) == 1, 'Only one-label classification is supported'
assert not estimator.loss_.is_multi_class, 'Only one-label classification is supported'
assert context.schemas[Schema.OUTPUT][0].optype == FeatureOpType.CATEGORICAL, \
'Classification output must be categorical'
assert len(context.schemas[Schema.OUTPUT][0].value_list) == 2, 'Only binary classifier is supported'
assert find_converter(estimator.init_) is not None, 'Can not find a converter for {}'.format(estimator.init_)
def __init__(self, estimator, context):
super(GradientBoostingConverter, self).__init__(estimator, context)
assert isinstance(estimator, GradientBoostingClassifier), \
'This converter can only process GradientBoostingClassifier instances'
assert len(context.schemas[
Schema.OUTPUT]) == 1, 'Only one-label classification is supported'
assert not estimator.loss_.is_multi_class, 'Only one-label classification is supported'
assert context.schemas[Schema.OUTPUT][0].optype == FeatureOpType.CATEGORICAL, \
'Classification output must be categorical'
assert len(context.schemas[Schema.OUTPUT]
[0].value_list) == 2, 'Only binary classifier is supported'
assert find_converter(
estimator.init_
) is not None, 'Can not find a converter for {}'.format(
estimator.init_)
def __init__(self, estimator, context):
super(GradientBoostingConverter,
self).__init__(estimator, context, self.MODE_CLASSIFICATION)
assert isinstance(estimator, GradientBoostingClassifier), \
'This converter can only process GradientBoostingClassifier instances'
assert len(context.schemas[self.SCHEMA_OUTPUT]
) == 1, 'Only one-label classification is supported'
assert not estimator.loss_.is_multi_class, 'Only one-label classification is supported'
assert context.schemas[self.SCHEMA_OUTPUT][
0].data_type == 'double', 'PMML version only returns probabilities'
assert context.schemas[self.SCHEMA_OUTPUT][
0].optype == 'continuous', 'PMML version only returns probabilities'
assert find_converter(
estimator.init_
) is not None, 'Can not find a converter for {}'.format(
estimator.init_)
def segmentation(self):
"""
Build a segmentation (sequence of estimators)
:return: Segmentation element
"""
# there is no notion of weighted sum, so we should take weighted average and multiply result by total weight
# in output transformation
segmentation = pmml.Segmentation(multipleModelMethod="weightedAverage")
# first, transform initial estimator
init_segment = pmml.Segment(weight=1)
init_segment.append(pmml.True_())
init_segment.append(find_converter(self.estimator.init_)(self.estimator.init_, self.context).model())
segmentation.append(init_segment)
for est in self.estimator.estimators_[:, 0]:
s = pmml.Segment(weight=self.estimator.learning_rate)
s.append(pmml.True_())
s.append(DecisionTreeConverter(est, self.context, self.MODE_REGRESSION)._model())
segmentation.append(s)
return segmentation
