Expectation maximization for fitting agent-based models to behavioural data
This repository is ongoing work on buulding robust algorithms for fitting agent models to data acquired from human and animal behavioural tasks. It is based on the Expectation Maximization family of algorithms and the specifics of the implementation can be found in the notes "expectation_maximization.pdf". Future work will focus on improving the Laplace approximation for each subject's posterior distribution and potentially replacing it with a sample-based algorithm. Adapting the code to work well with the Turing.jl probabilistic programming framework is another priority.
Please check the notes for implementation details.