omitevski

######
#It uses the cross entropy error function:
f = -(1/batchSize)*sum(data*log(recon) + (1-data)*log(1-recon))

#compute derivatives for the last layer
Ix_n2 = (1/float(batchSize))*(recon-data)
# outer product
# w_n2_probs the outputs at the layer before the last

dwLast = dot(w_n2_probs.t, Ix_n2)

omitevski

""" 
Author: Oliver Mitevski

References:
A Generalized Linear Model for Principal Component Analysis of Binary Data, 
Andrew I. Schein; Lawrence K. Saul; Lyle H. Ungar

The code was translated and adapted from Jakob Verbeek's 
"Hidden Markov models and mixtures for Binary PCA" implementation in MATLAB

omitevski

""" 
Author: Oliver Mitevski

References:
A Generalized Linear Model for Principal Component Analysis of Binary Data, 
Andrew I. Schein; Lawrence K. Saul; Lyle H. Ungar

The code was translated and adapted from Jakob Verbeek's 
"Hidden Markov models and mixtures for Binary PCA" implementation in MATLAB

omitevski

""" 
Author: Oliver Mitevski

References:
A Generalized Linear Model for Principal Component Analysis of Binary Data, 
Andrew I. Schein; Lawrence K. Saul; Lyle H. Ungar

The code was translated and adapted from Jakob Verbeek's 
"Hidden Markov models and mixtures for Binary PCA" implementation in MATLAB