k501

from pylearn2.datasets.csv_dataset import CSVDataset
import pickle

print 'convert: train.csv -> train.pkl'
pyln_data = CSVDataset("../R/out_for_pylearn2/train.csv", one_hot=True, delimiter=',')
pickle.dump( pyln_data, open( "data/train.pkl" , 'w' ) )

k501

""" Deep Auto-Encoder implementation
	
	An auto-encoder works as follows:

	Data of dimension k is reduced to a lower dimension j using a matrix multiplication:
	softmax(W*x + b)  = x'
	
	where W is matrix from R^k --> R^j

	A reconstruction matrix W' maps back from R^j --> R^k