simple_example_classification.py

import numpy
import sklearn.datasets
import sklearn.model_selection
from matplotlib import pyplot as plt
from sklearn.tree import DecisionTreeClassifier

from pytolemaic import PyTrust, FeatureTypes


def run(fast=False):
    # Dataset: xtrain, ytrain, xtest, ytest
    # noinspection PyUnresolvedReferences
    data = sklearn.datasets.load_wine(return_X_y=False)

    x = data['data']
    y = data['target']
    feature_names = data['feature_names']
    labels = data['target_names']

    train_inds, test_inds = sklearn.model_selection.train_test_split(
        numpy.arange(len(data['data'])), test_size=0.3)

    xtrain, ytrain = x[train_inds], y[train_inds]
    xtest, ytest = x[test_inds], y[test_inds]

    if fast:
        inds = numpy.random.permutation(len(xtrain))[:500]
        xtrain = xtrain[inds,:]
        ytrain = ytrain[inds]
        inds = numpy.random.permutation(len(xtest))[:500]
        xtest = xtest[inds, :]
        ytest = ytest[inds]


    # Train estimator
    estimator = DecisionTreeClassifier()
    estimator.fit(xtrain, ytrain)

    # Initiating PyTrust
    pytrust = PyTrust(
        model=estimator,
        xtrain=xtrain, ytrain=ytrain,
        xtest=xtest, ytest=ytest,
        metric='recall')

    # Initiating PyTrust with more information
    pytrust = PyTrust(
        model=estimator,
        xtrain=xtrain, ytrain=ytrain,
        xtest=xtest, ytest=ytest,
        feature_names=feature_names,
        feature_types=[FeatureTypes.numerical] * xtrain.shape[1],
        target_labels={i: label for i, label in enumerate(labels)},
        splitter='stratified',
        metric='recall')

    pytrust.scoring_report.plot()

    pytrust.sensitivity_report.plot()

    pytrust.dataset_analysis_report.plot()

    pytrust.quality_report.plot()

    pytrust.anomalies_in_data_report.plot()

    sample = xtest[0, :].reshape(1, -1)
    explainer = pytrust.create_lime_explainer(max_samples=4000 if fast else 16000)
    explainer.explain(sample=sample)

    uncertainty_model = pytrust.create_uncertainty_model(method='default')
    prediction = uncertainty_model.predict(sample)  # same as model.predict
    uncertainty = uncertainty_model.uncertainty(sample)  # uncertainty value

    print("Let's check for insights...")
    print('\n'.join(pytrust.insights))
    print("Done!")


if __name__ == '__main__':
    run()
    plt.show()