Fork Author: Kevin Paniagua Romero
BayesInterpret is a computational project focused on creating interactive dashboards for interpreting and visualizing machine learning models using Dash. This fork is part of the BayesInterpret initiative, where the main goal is to apply an intuitive interface to existing implementations, making model outputs and analyses easier to understand and use.
This work is based on ideas and research from the Computational Intelligence Group (CIG) and integrates interface design with machine learning tools to enhance interpretability in a simple and practical way. I'll be using Dash for the new interface.
In this repository we find the code developed for the master's thesis "Interpreting Bayesian Network-based Clustering". Code for solving the clustering problem with Bayesian Networks (BN) for categorical data is found. discrete_structure.py contains all the functions needed for such purpose, whereas customers.py is an applied example on how to use these functions for obtaining the desired BN for the customers dataset. This dataset can be found in the dataset folder, we also find in the network folder the learnt BN best_network_customers_2.pickle .
Code for analyzing the respective BNs obtained is implemented in discrete_analysis_hellinger.py and discrete_representation.py . In these files we find the functions implementing the methodology developed in the thesis for cluster characterization in order to solve cluster labeling. Finally, customers_analysis.py contains an example on how to apply these functions for the network obtained in customers.py .
Moreover, continuous_structure.py and continuous_analysis.py are files where the proposed methodology and structure learning are adapted for a particular continuous case presented in the thesis with dataset example (dataset folder). This can be applied to others datasets but for problems with external libraries it may not work.
Finally, gbf_example.py contains code for a particular example of the thesis and pybnesianCPT_to_df.py with radar_chart_discrete.py contain support functions needed for the implementations mentioned before.