What this be?

We will use the ALARM (A Logical Alarm Reduction Mechanism) dataset to infer a consistent Bayesian Network underlying a complex medical dataset. This Bayes-Net can then be used for constructing a data-driven medical diagnostic system using inference algorithms, such as Belief-Propagation. Three type of variables are present in the ALARM dataset- diagnoses, measurements, and intermediate variables. After constructing a suitable probabilistic model, the resulting model can be used for automatically diagnosing a patient with a set of symptoms and test results. For details on the variables present in the dataset, please refer to the original paper and the webpage https://rdrr.io/cran/bnlearn/man/alarm. In this exercise, we will be estimating the most likely tree-structured probabilistic graphical model underlying the ALARM dataset.

• Go to PSET2-EE16B070.ipynb for problem 1.
• Tree Diagrams Obtained are at mi_10k.gv.pdf and mi_jvhw_10k.gv.pdf
• Collaborators: Siddharth Nayak

Requirements:

• graphviz
• numpy
• matplotlib
• pandas
• sklearn

For installing graphviz you have to install the graphviz package along with the python graphviz package.

For MacOSX: brew install graphviz

For Ubuntu: sudo apt-get install graphviz

After the above step,

• pip install graphviz or conda install graphviz

For other python libraries, just do pip install <package-name>

References:

• Prim's Algorithm: CoderByte tutorial on Minimum Spanning Tree
• JVHW Mutual Information Estimator: GitHub Repository