piBreakDown

python version of iBreakDown

Installation

Clone the repository, then open console in piBreakDown/piBreakDown directory and run

pip install . 

Example

Import necessary packages

from piBreakDown.Attributions import Attributions
from piBreakDown.PlotUtils import PlotUtils
from sklearn.ensemble import RandomForestClassifier
import pandas as pd
from sklearn.datasets import load_iris

import some dataset

iris = load_iris()
data = pd.DataFrame(iris['data'])
target = iris['target']

You need to set the column names to strings

data.columns = iris.feature_names

Fit your data

rf = RandomForestClassifier()
rf.fit(data, target)

Use local_attributions

attr = Attributions(rf, data, 'Class')
results = attr.local_attributions(data.loc[0,:], # observation to explain
                                classes_names = [0,1,2]) # available classes_names

Draw the plot

PlotUtils.plot_contribution(results, 
                                plot_class = 0) # class to plot

The results are displayed below