CTBN 0.0.1

Conditional Tree Bayesian Network for multi-label classification

Conditional Tree Bayesian Network

This package can be used as a library to fit a Conditional Tree Bayesian Network (CTBN)[1].

The main module is the CTBN class.

1. Use the fit() method to fit the CTBN to your multi-label classification data [1]. This method generates an optimal CTBN which is an instance of DirectedGraph using the Chu-Liu-Edmond's algorithm for finding a maximum spanning arborescence [2].
2. The predict() method returns the most likely assignment to the class variables along with the probability of the assignment. The predict method uses the junction tree algorithm[3] to run the most likely explanation(MLE) query.

Usage

An example of using the package can be found in the jupyter notebook here.

# Import the CTBN class and assuming you have a dataset X_train and Y_train which are numpy arrays.
from ctbn import CTBN
model = CTBN()
model.fit(X_train, Y_train)

#Calling the fit method will generate an optimal CTBN graph of type
#DirectedGraph defined in src/graph_preliminaries.py

#Get predictions and the probability of a prediction on a single sample
#using the predict method. This method will in turn call the junction tree
#algorithm to run the max-sum algorithm on a test_sample.

max_log_prob, max_assignment = model.predict(test_sample)

References

[1] Batal, Iyad and Hong, Charmgil and Hauskrecht, Milos (2013). An Efficient Probabilistic Framework for Multi-Dimensional Classification. In Proceedings of the 22nd ACM International Conference on Information amp; Knowledge Management, CIKM ’13, New York, NY,USA, pp. 2417–2422. Association for Computing Machinery. https://doi.org/10.1145/2505515.2505594

[2] Chu, Y. J. and T. H. Liu, "On the Shortest Arborescence of a Directed Graph," Sci. Sinica, 14, 1965, pp. 1396-1400.

[3] Koller, D., & Friedman, N. (2009). Probabilistic Graphical Models: Principles and Techniques. MIT Press.