This project proposes an efficient character based recurrent neural network. The RNN has word level input instead of standard character level input. The output of thhe RNN is conditioned by the previous words.
There are two models:
char-rnn-conditional
- This contains the character RNN which has conditional outputs
mixed-rnn
- This contains the mixed word/char level rnn code
the mixed-rnn code splits up the stream of characters into words using the character _ .
This code has been tested on Linux, but should work on any machine. The is no dependencies.
See the CONTRIBUTING file for how to help out.
Conditional character based RNN is BSD-licensed. We also provide an additional patent grant.