Prediction Interpreter

This package contains a wrapper to interpret any predictor. Use this package:

• to interpret a machine learning model that runs over a service
• to interpret a trained ML- model in python, that has insufficient features for interpretation
• to interpret a trained ML- model in python, using the methods already set up for you
• to get ideas about how to use the interpretation methods

By using established interpretation- techniques, the package only needs a function that takes a dataset and returns a prediction, as well as some information about the data and columns. The package then automatically sets up a wrapper containing the important information and configurations to run interpretations.

To set up the predictor, state the necessary information as in file StartPredictionInterpreter given: Here, we use a DummyML-Model that predicts the amount of visitors on a day, depending on whether it is holiday, how much the ticket price is, and what'S the weather like. (For your additional information: There are many visitors if its holiday, or when the price is low. The weather has a random effect on the amount of visitors. But all of this you will see when running the dummy data.)

Additionally to the DummyMLModel, it is necessary to give

• the testdata as a panda-Dataframe

• as well as the result column's name

• all data columns' names

• the numerical column names (if there are any...)

• the categories in result column (classes_)

• rather the result is a continuous value

  #get data and object that has singlepredict in correct format dm = DummyMLModel() data = dm.testdata

  #define necessary variables for techniques standardColumns = data.columns.to_list() resultcolumn = "visitorsOnThisDay" listOfNumericalColumns = ["ticketPrice"] classes = data[resultcolumn].unique().tolist() resultIsContinuous = False

After that, create the the interpreter with the before defined parameters, and run the interpretation techniques that are of interest for you.

#create interpreter 
predictionInterpreter = PredictionInterpreterClass(dm.predict, listOfNumericalColumns, standardColumns, resultcolumn, _classes_, data, resultIsContinuous)

#call interpretation technique  s you want to use:
predictionInterpreter.plotpdpOfDistanceToTrueResultSklearn() # only works if called without any prior methods
predictionInterpreter.plotpdpOfDistanceToTrueResultSklearn2D()
predictionInterpreter.writeDistribution("visitorsOnThisDay")
predictionInterpreter.plotConfusionTable()
predictionInterpreter.printImportanceEli5(exceptedColumns = resultcolumn)
predictionInterpreter.printImportanceEli5(distanceAnalysis=True)
predictionInterpreter.featureAnnulation(annulationValue = "0")
predictionInterpreter.plotIce()
predictionInterpreter.plotpdpOfDistanceToTrueResultPdpbox(featureToExamine="ticketPrice")
predictionInterpreter.plotpdpOfDistanceToTrueResultPdpbox(featuresToExamine=["holidayYN", "ticketPrice"])
predictionInterpreter.plotpdpOfDistanceToTrueResultPdpbox(featureToExamine="ticketPrice", featuresToExamine=["holidayYN", "ticketPrice"])
predictionInterpreter.globalSurrogateModel()