# Copyright (C) Microsoft Corporation. All rights reserved.
# Licensed under the MIT License. See LICENSE in project root for information.
import sys
if sys.version >= '3':
basestring = str
from pyspark import SparkContext, SQLContext
from pyspark.sql import DataFrame
from pyspark.ml.param.shared import *
from pyspark import keyword_only
from pyspark.ml.util import JavaMLReadable, JavaMLWritable
from synapse.ml.core.serialize.java_params_patch import *
from pyspark.ml.wrapper import JavaTransformer, JavaEstimator, JavaModel
from pyspark.ml.evaluation import JavaEvaluator
from pyspark.ml.common import inherit_doc
from synapse.ml.core.schema.Utils import *
from pyspark.ml.param import TypeConverters
from synapse.ml.core.schema.TypeConversionUtils import generateTypeConverter, complexTypeConverter
[docs]@inherit_doc
class TabularLIME(ComplexParamsMixin, JavaMLReadable, JavaMLWritable, JavaTransformer):
"""
Args:
backgroundData (object): A dataframe containing background data
categoricalFeatures (list): Name of features that should be treated as categorical variables.
inputCols (list): input column names
kernelWidth (float): Kernel width. Default value: sqrt (number of features) * 0.75
metricsCol (str): Column name for fitting metrics
model (object): The model to be interpreted.
numSamples (int): Number of samples to generate.
outputCol (str): output column name
regularization (float): Regularization param for the lasso. Default value: 0.
targetClasses (list): The indices of the classes for multinomial classification models. Default: 0.For regression models this parameter is ignored.
targetClassesCol (str): The name of the column that specifies the indices of the classes for multinomial classification models.
targetCol (str): The column name of the prediction target to explain (i.e. the response variable). This is usually set to "prediction" for regression models and "probability" for probabilistic classification models. Default value: probability
"""
backgroundData = Param(Params._dummy(), "backgroundData", "A dataframe containing background data")
categoricalFeatures = Param(Params._dummy(), "categoricalFeatures", "Name of features that should be treated as categorical variables.", typeConverter=TypeConverters.toListString)
inputCols = Param(Params._dummy(), "inputCols", "input column names", typeConverter=TypeConverters.toListString)
kernelWidth = Param(Params._dummy(), "kernelWidth", "Kernel width. Default value: sqrt (number of features) * 0.75", typeConverter=TypeConverters.toFloat)
metricsCol = Param(Params._dummy(), "metricsCol", "Column name for fitting metrics", typeConverter=TypeConverters.toString)
model = Param(Params._dummy(), "model", "The model to be interpreted.")
numSamples = Param(Params._dummy(), "numSamples", "Number of samples to generate.", typeConverter=TypeConverters.toInt)
outputCol = Param(Params._dummy(), "outputCol", "output column name", typeConverter=TypeConverters.toString)
regularization = Param(Params._dummy(), "regularization", "Regularization param for the lasso. Default value: 0.", typeConverter=TypeConverters.toFloat)
targetClasses = Param(Params._dummy(), "targetClasses", "The indices of the classes for multinomial classification models. Default: 0.For regression models this parameter is ignored.", typeConverter=TypeConverters.toListInt)
targetClassesCol = Param(Params._dummy(), "targetClassesCol", "The name of the column that specifies the indices of the classes for multinomial classification models.", typeConverter=TypeConverters.toString)
targetCol = Param(Params._dummy(), "targetCol", "The column name of the prediction target to explain (i.e. the response variable). This is usually set to \"prediction\" for regression models and \"probability\" for probabilistic classification models. Default value: probability", typeConverter=TypeConverters.toString)
@keyword_only
def __init__(
self,
java_obj=None,
backgroundData=None,
categoricalFeatures=[],
inputCols=None,
kernelWidth=0.75,
metricsCol="r2",
model=None,
numSamples=1000,
outputCol="TabularLIME_4e15a4bc7791__output",
regularization=0.0,
targetClasses=[],
targetClassesCol=None,
targetCol="probability"
):
super(TabularLIME, self).__init__()
if java_obj is None:
self._java_obj = self._new_java_obj("com.microsoft.azure.synapse.ml.explainers.TabularLIME", self.uid)
else:
self._java_obj = java_obj
self._setDefault(categoricalFeatures=[])
self._setDefault(kernelWidth=0.75)
self._setDefault(metricsCol="r2")
self._setDefault(numSamples=1000)
self._setDefault(outputCol="TabularLIME_4e15a4bc7791__output")
self._setDefault(regularization=0.0)
self._setDefault(targetClasses=[])
self._setDefault(targetCol="probability")
if hasattr(self, "_input_kwargs"):
kwargs = self._input_kwargs
else:
kwargs = self.__init__._input_kwargs
if java_obj is None:
for k,v in kwargs.items():
if v is not None:
getattr(self, "set" + k[0].upper() + k[1:])(v)
[docs] @keyword_only
def setParams(
self,
backgroundData=None,
categoricalFeatures=[],
inputCols=None,
kernelWidth=0.75,
metricsCol="r2",
model=None,
numSamples=1000,
outputCol="TabularLIME_4e15a4bc7791__output",
regularization=0.0,
targetClasses=[],
targetClassesCol=None,
targetCol="probability"
):
"""
Set the (keyword only) parameters
"""
if hasattr(self, "_input_kwargs"):
kwargs = self._input_kwargs
else:
kwargs = self.__init__._input_kwargs
return self._set(**kwargs)
[docs] @classmethod
def read(cls):
""" Returns an MLReader instance for this class. """
return JavaMMLReader(cls)
[docs] @staticmethod
def getJavaPackage():
""" Returns package name String. """
return "com.microsoft.azure.synapse.ml.explainers.TabularLIME"
@staticmethod
def _from_java(java_stage):
module_name=TabularLIME.__module__
module_name=module_name.rsplit(".", 1)[0] + ".TabularLIME"
return from_java(java_stage, module_name)
[docs] def setBackgroundData(self, value):
"""
Args:
backgroundData: A dataframe containing background data
"""
self._set(backgroundData=value)
return self
[docs] def setCategoricalFeatures(self, value):
"""
Args:
categoricalFeatures: Name of features that should be treated as categorical variables.
"""
self._set(categoricalFeatures=value)
return self
[docs] def setKernelWidth(self, value):
"""
Args:
kernelWidth: Kernel width. Default value: sqrt (number of features) * 0.75
"""
self._set(kernelWidth=value)
return self
[docs] def setMetricsCol(self, value):
"""
Args:
metricsCol: Column name for fitting metrics
"""
self._set(metricsCol=value)
return self
[docs] def setModel(self, value):
"""
Args:
model: The model to be interpreted.
"""
self._set(model=value)
return self
[docs] def setNumSamples(self, value):
"""
Args:
numSamples: Number of samples to generate.
"""
self._set(numSamples=value)
return self
[docs] def setOutputCol(self, value):
"""
Args:
outputCol: output column name
"""
self._set(outputCol=value)
return self
[docs] def setRegularization(self, value):
"""
Args:
regularization: Regularization param for the lasso. Default value: 0.
"""
self._set(regularization=value)
return self
[docs] def setTargetClasses(self, value):
"""
Args:
targetClasses: The indices of the classes for multinomial classification models. Default: 0.For regression models this parameter is ignored.
"""
self._set(targetClasses=value)
return self
[docs] def setTargetClassesCol(self, value):
"""
Args:
targetClassesCol: The name of the column that specifies the indices of the classes for multinomial classification models.
"""
self._set(targetClassesCol=value)
return self
[docs] def setTargetCol(self, value):
"""
Args:
targetCol: The column name of the prediction target to explain (i.e. the response variable). This is usually set to "prediction" for regression models and "probability" for probabilistic classification models. Default value: probability
"""
self._set(targetCol=value)
return self
[docs] def getBackgroundData(self):
"""
Returns:
backgroundData: A dataframe containing background data
"""
ctx = SparkContext._active_spark_context
sql_ctx = SQLContext.getOrCreate(ctx)
return DataFrame(self._java_obj.getBackgroundData(), sql_ctx)
[docs] def getCategoricalFeatures(self):
"""
Returns:
categoricalFeatures: Name of features that should be treated as categorical variables.
"""
return self.getOrDefault(self.categoricalFeatures)
[docs] def getKernelWidth(self):
"""
Returns:
kernelWidth: Kernel width. Default value: sqrt (number of features) * 0.75
"""
return self.getOrDefault(self.kernelWidth)
[docs] def getMetricsCol(self):
"""
Returns:
metricsCol: Column name for fitting metrics
"""
return self.getOrDefault(self.metricsCol)
[docs] def getModel(self):
"""
Returns:
model: The model to be interpreted.
"""
return JavaParams._from_java(self._java_obj.getModel())
[docs] def getNumSamples(self):
"""
Returns:
numSamples: Number of samples to generate.
"""
return self.getOrDefault(self.numSamples)
[docs] def getOutputCol(self):
"""
Returns:
outputCol: output column name
"""
return self.getOrDefault(self.outputCol)
[docs] def getRegularization(self):
"""
Returns:
regularization: Regularization param for the lasso. Default value: 0.
"""
return self.getOrDefault(self.regularization)
[docs] def getTargetClasses(self):
"""
Returns:
targetClasses: The indices of the classes for multinomial classification models. Default: 0.For regression models this parameter is ignored.
"""
return self.getOrDefault(self.targetClasses)
[docs] def getTargetClassesCol(self):
"""
Returns:
targetClassesCol: The name of the column that specifies the indices of the classes for multinomial classification models.
"""
return self.getOrDefault(self.targetClassesCol)
[docs] def getTargetCol(self):
"""
Returns:
targetCol: The column name of the prediction target to explain (i.e. the response variable). This is usually set to "prediction" for regression models and "probability" for probabilistic classification models. Default value: probability
"""
return self.getOrDefault(self.targetCol)