===========

Inferring binary features for neural populations.

This code implements a version of the Gamma-Poisson model on a pseudopopulation of independently recorded neurons. Details of the model and inference are in gamma_poisson_notes.tex.

• gamma_model.py: Implements the Gamma model for firing rates with HMM state space dynamics.
• helpers.py: miscellaneous functions useful for setting up the model and evaulating outputs.

• forward_backward.py: Forward-Backward algorithm implemented using Numba's JIT facilities to provide much faster inference.
• hsmm_forward_backward.py: Forward-Backward inference for the hidden semi-Markov model. Again uses Numba to provide just-in-time compilation.

Deprecated files from semi-Markov model in gamma poisson model and lognormal model. The archive folder contains the codes that are not used in our current settings, but kept for recording purposes.

The docs folder contains notes, papers, and other documentation for the algorithms.

The experiments folder contains case studies for the application of the algorithm to test data sets.

The algorithm is defined by a graphical model with nodes in a directed graph corresponding to each variable. The nodes folder contains code defining odes for several common distribution types.

Unit tests are located in the tests folder. These can be run in their entirety by

nosetests

or as modules by

nosetests tests/name_of_test_module.py

• Correlated noise across neurons (log-Normal model does this)
• Response sparsity within populations, not just features
• Smarter feature sparsity
• Noise autocorrelation in time (hard)