Chao Xing 2015-08-31

Last Week:

Finish Matrix Factorization for X - Z * U * ZT.

Test for different gradient descent function( mini-batch, sgd, adadelta) and different initial method (Non negative | gaussian distribution).

Fix bug for enormous increase for mini-batch & sgd by normalize gradient for each step.

This Week:

Finish test & knowledge enhancement.

Plan to do:

 min((X - Z * U * ZT) + (Z * ZT - Y))
 min((X + Y) - Z * U * ZT)
 min(concatenate(X, Y) - Z * U * ZT)

For this X is from raw text & Y is derive from knowledge.