elglm
matlab routines for demonstrating the expected log-likelihood
Description: Computes the maximum "expected log-likelihood" for standard regression and linear-non-linear Poisson spiking models. Demonstrates how this estimator can be used to approximate the maximum likelihood estimate.
Relevant publication : [Ramirez,A.D.; Paninski, L., "Fast inference in generalized linear models via expected log-likelihoods", Journal of Computational Neuroscience (36), 2014]
Dependencies
) Matlab
) Statistics and Machine Learning Toolbox
) Mark Schmidt's minFunc toolbox https://www.cs.ubc.ca/~schmidtm/Software/minFunc_2007.zip
For the spikeGLMDemo.m
) GLMspiketools from the Pillow lab https://github.com/pillowlab/GLMspiketools/archive/old_v1.zip
) Download and instal GLM Net https://web.stanford.edu/~hastie/glmnet_matlab/glmnet_matlab.zip
Installation
- Download the elglm zip file or clone the repository: "git clone https://github.com/alxdroR/elglm"
- Download and install the dependent code (minFunc and optionally GLMspiketools and glmnet). GLMspiketools requires mex file compilation. glmnet might as well depending on your version of Matlab and OS.
- Add elglm and dependent code to the Matlab path
Usage
Open and read the demo scripts in /demos/ to see simple examples of code usage. LNPdemo - compares the maximum likelihood estimator and maximum expected likelihood estimator with and without an L2 penalty on the likelihood.