A Python implementation of the Replicated Softmax model.

This is an implementation of R. Salakhutdinov and G.E. Hinton's Replicated Softmax model,
see http://www.mit.edu/~rsalakhu/papers/repsoft.pdf
You will need Numpy and SciPy installed. I don't provide the dataset.
Also feel free to change the initialization of the biases, it just worked for me.
It is part of my M.Sc. Thesis, advised by Christian Osendorfer at TUM, I6.
You can reach me here: joerg [-at-] fylance [-.-] de

• rsm.py: contains the RSM implementation encapsulated in a class.
• rsm_driver_20newsgroups.py: shows an exemplary usage of the class.

Download: rsm.tar.gz