class-based implementation of nonlinear input model
Builds on the original "NIMtoolbox" which can be downloaded here: http://neurotheory.umd.edu/nimcode
See also: McFarland JM, Cui Y, Butts DA (2013) Inferring nonlinear neuronal computation based on physiologically plausible inputs. PLoS Computational Biology 9(7): e1003142.
In addition to being object-oriented, there are several additions/changes.
- The stimulus Xmat now must be a cell array, and each subunit has a field which specifies which element of the stimulus cell array it 'acts on'.
- There is now an option to fit an 'offset' term along with the filters.
- Optional inputs are now provided through 'option-flag/value' pairs
- The likelihood function is now specified separately from the noise distribution model.
- There are several new options for the upstream NLs, and spkNL, including 'rectified power-law functions'
and many more minor changes... See the help docs for each function.
The fit_filters function has also had some significant revamping that greatly speeds up estimation of high-dimensional filters, particularly for models with lots of subunits