pavanchaggar / networkinference

Implementing network models, comparing with data and performing Bayesian Inference

I am currently a DPhil student at the Mathematical Institute, University of Oxford, supervised by Prof. Alain Goriely and Prof. Saad Jbabdi.

I am working on mathematical models of neurodegeneration, specifically, Alzheimer's Disease. My work is very interdisciplinary and includes mechanistic modelling, Bayesian inference and brain image analysis.

• 📫 How to reach me: [email protected] or twitter: @ChaggarPavan

networkinference's Issues

Implement and test an MCMC

Implement an MCMC scheme to work on smaller models. This will be used to test the accuracy of vb and sbi. If possible, this should be implemented using a package (for e.g. pints, pymc3, numpyro).

Once complete, this should be tested on the same toy network model as other inference schemes.

• create and test class to initialise mcmc problem.
• unit test inference toy network model

Hierarchical Modelling

Add ability for hierarchical modelling. This will be useful for patient prediction. Specifically, how being able to draw parameters from a population model.

This could be linked to work on Alzheimer's subtypes and classifications of spread.

Testing vb

Refactor vb script for better flow of variables by setting up an vb model object. This should replace the use of params and priors in /_inference/_vb.py

Once this is complete. Unit test the vb framework:

• unit test inference setup api
• unit test vb functions
• unit test vb inference on a toy network model

Implement and test simulation based inference

Implement simulation based inference using the sbi package. Similarly to mcmc issue, this will require:

• implementing and testing a class to initially the sbi problem
• test the sbi inference software on simple toy problem (the same as vb and mcmc