Trained on the mnist data-set in tensorflow..
Accuracy on training set is 0.98
Accuracy on test set is 0.9725
Trained on mnist data-set in tensorlfow..
Accuracy on training set is 0.99
Accuracy on test set is 0.9892
Trained on all the plays of Shakespeare using tensorlfow..
The RNN has Gated Recurrent Unit(GRU) cells..
The training is done at character level and text is encoded using a 98 element vector..
As the rnn trains, the random output plays become more and more sophisticated and realistic..
The Random plays generated by the network at various stages in training are stored here...