RegressionTrustee

Simple example on how to use the ClassificationTrustee class to extract a decision tree from a MLPRegressor (neural network) from scikit-learn.

# importing required libraries
# importing Scikit-learn library and datasets package
import graphviz

from sklearn import tree
from sklearn import datasets
from sklearn.neural_network import MLPRegressor
from sklearn.model_selection import train_test_split
from sklearn.metrics import r2_score

from trustee import RegressionTrustee

# Loading the diabetes dataset (regression)
diabetes = datasets.load_diabetes()
X, y = datasets.load_diabetes(return_X_y=True)
# Spliting arrays or matrices into random train and test subsets
# i.e. 70 % training dataset and 30 % test datasets
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.30)

# creating a MLP regressor
clf = MLPRegressor(solver="lbfgs", alpha=1e-5, hidden_layer_sizes=(100, 50), max_iter=500, random_state=1)
# Training the model on the training dataset
# fit function is used to train the model using the training sets as parameters
clf.fit(X_train, y_train)
# performing predictions on the test dataset
y_pred = clf.predict(X_test)

# Evaluate model accuracy
print("Model R2-score:")
print(r2_score(y_test, y_pred))

# Initialize Trustee and fit for classification models
trustee = RegressionTrustee(expert=clf)
trustee.fit(X_train, y_train, num_iter=50, num_stability_iter=10, samples_size=0.3, verbose=True)

# Get the best explanation from Trustee
dt, pruned_dt, agreement, reward = trustee.explain()
print(f"Model explanation training (agreement, fidelity): ({agreement}, {reward})")
print(f"Model Explanation size: {dt.tree_.node_count}")
print(f"Top-k Prunned Model explanation size: {pruned_dt.tree_.node_count}")

# Use explanations to make predictions
dt_y_pred = dt.predict(X_test)
pruned_dt_y_pred = pruned_dt.predict(X_test)

# Evaluate accuracy and fidelity of explanations
print("Model explanation global fidelity:")
print(r2_score(y_pred, dt_y_pred))
print("Top-k Model explanation global fidelity:")
print(r2_score(y_pred, pruned_dt_y_pred))

print("Model explanation R2-score:")
print(r2_score(y_test, dt_y_pred))
print("Top-k Model explanation R2-score:")
print(r2_score(y_test, pruned_dt_y_pred))

# Output decision tree to pdf
dot_data = tree.export_graphviz(
    dt,
    feature_names=diabetes.feature_names,
    filled=True,
    rounded=True,
    special_characters=True,
)
graph = graphviz.Source(dot_data)
graph.render("dt_explanation")

# Output pruned decision tree to pdf
dot_data = tree.export_graphviz(
    pruned_dt,
    feature_names=diabetes.feature_names,
    filled=True,
    rounded=True,
    special_characters=True,
)
graph = graphviz.Source(dot_data)
graph.render("pruned_dt_explation")