serendipity-matrix 1.0.0

Serendipity Matrix

Serendipity Matrix

The package provides the methods to provide the serendipity matrix and for visualizing the serendipity matrix in a horizontal bar chart. The serendipity matrix is an innovative method to understand the behavior of a prediction model for classification problems.

This matrix provides information about the degree of certainty that the classifier has in its own predictions, indicating whether it is robust and reliable or uncertain and doubtful. This method has two variants: the class-independent serendipity matrix and the class-specific serendipity matrix depending on the kind of analysis required.

By analyzing the data provided by them, our goal is to improve the reliability and explainability of prediction models and to provide users with a clearer understanding of why a model is certain or uncertain about its predictions.

Installation

Serendipity Matrix can be installed from PyPI

pip install serendipity_matrix

Or you can clone the repository and run:

pip install .

Sample usage

from serendipity_matrix import class_indep_matrix, class_spec_matrix, plot_class_spec
from sklearn.naive_bayes import GaussianNB
from ucimlrepo import fetch_ucirepo

# Loads the dataset
wine_quality  = fetch_ucirepo(id=186) 
X, y = wine_quality.data.features, wine_quality.data.targets.squeeze()

# Training and predict
model = GaussianNB().fit(X, y)
result = model.predict_proba(X)

# Calculates and prints the class-independent serendipity matrix
ci_matrix = class_indep_matrix(y, result)
print(ci_matrix)

# Calculates and prints the class-specific serendipity matrix
cs_matrix = class_spec_matrix(y, result)
print(cs_matrix)

# Plots the class-specific serendipity matrix
plot_class_spec(y, result)

Result sample

Class-independent serendipity matrix

Reliability	Overconfidence	Underconfidence	Serendipity
0.26401	0.29623	0.318717	0.021864

Class-specific serendipity matrix

CLASS_NAME	Reliability	Overconfidence	Underconfidence	Serendipity
3	0.072567	0.493307	0.412262	0.021864
4	0.054267	0.310598	0.597766	0.037369
5	0.383453	0.308486	0.198692	0.109368
6	0.272532	0.264399	0.267862	0.195207
7	0.227578	0.400299	0.308644	0.063479
8	0.010690	0.017746	0.902782	0.068782
9	0.031477	0.137502	0.822256	0.008765

Class-specific serendipity matrix horizontal bar chart

Citation

The methodology is described in detail in:

[1] J. S. Aguilar-Ruiz and A. García Conde, “”