/hclearn

Unitary coherent hippocampus model code, created by Charles Fox and Alan Saul and edited by Mathew Evans

Primary LanguagePython

hclearn

Unitary coherent hippocampus model code, created by Charles Fox and Alan Saul and edited by Mathew Evans

Below are notes on how I (M.E.) understand what the code does. This document has now been superseeded by the notebook, but a detailed list of functions might stay here.

rbm.py : Restricted Boltzmann Machine code, based on Hinton's TRBM work. Includes a number of functions for building the network, setting observed and hidden units, training weights and measuring the 'Energy'. List Of Functions boltzmannProbs(W,x) : Returns the probability of a node being on (is the weight*x above a half?)

	trainPriorBias(hids) : Normalises and concatenates the hidden values (I think! 30/4/14). Uses addBias(hids) to concatenate the hidden node values (their energy?). Also adds/removes tiny values to/from zeros/ones, and uses invsig to normalise the output.

	addBias(xs) : reshapes and concatenates input vector with a column of ones (I think! 30/4/14). 

cffun.py : More functions, probably from Charles Fox (hence CFfun).

	invsig(x) : takes neg log of 1/x - 1. Squashes x into the range 0:1, with infinitely large outputs near those values. (Try inputing -log((1/x)-1)) into google.