Functions to fit subunit grid models to spiking data from retinal ganglion cells.
This repository is a supplement to the manuscript "Nonlinear receptive fields evoke redundant retinal coding of natural scenes".
- MATLAB >= R2016b
- MATLAB Toolboxes:
- Parallel Computing Toolbox
- Signal Processing Toolbox
- Statistics and Machine Learning Toolbox
- An NVIDIA GPU and CUDA capabilities
Model optimization happens on the GPU for performance reasons, but a GPU is not strictly required. Replacing gpuArray arrays with standard single or double arrays will do the job.
To "install" the repository, just clone it or download and unpack a ZIP version of it (<1 min).
Data can be downloaded from our G-node repository. First you need to download and unpack the datasets in the same path as the repository. Demos are expected to produce informative figures at their end and are expected to run for up to a few minutes for subunit model fitting (on a normal desktop computer with a GPU). Fitting is slower for flickering gratings.
The script demo_flashed_gratings.m will guide you through data loading and model fitting. This script also uses the fitted model to generate predictions for natural images.
You will need datasets for which gratingflashes_data.m is available. You can change the path of the dataset at the beginning of the script. Model components can be visualized during the fitting process by setting the parameter options.showfig to True.
Similar to the flashed gratings, the script demo_flickering_gratings.m will guide you through data loading, model fitting, and generating model predictions for natural movies.
You will need datasets for which gratingflicker_data.m is available. You can change the path of the dataset at the beginning of the script.
The script demo_read_model_params.m contains code for reading the parameters of fitted models for marmoset and mouse retinal ganglion cells. These parameters were used in the manuscript and are stored as plain text files in this repository.
You can easily reuse the code for your own recordings, as long as the data are formatted in a similar manner.
gfdata = load('gratingflashes_data')
icell = 1
xx = stiminfo(presentOrder,:); %stimulus values
yy = trialCounts(icell,:)';