https://github.com/dluman/char-rnn
A basic, single-layer character-level RNN based on Andrej Karpathy's "vanilla" minimal RNN. I am looking for suggestions for improvements!
https://github.com/dluman/char-rnn
Last synced: 3 months ago
JSON representation
A basic, single-layer character-level RNN based on Andrej Karpathy's "vanilla" minimal RNN. I am looking for suggestions for improvements!
- Host: GitHub
- URL: https://github.com/dluman/char-rnn
- Owner: dluman
- License: unlicense
- Created: 2017-06-07T15:20:26.000Z (about 8 years ago)
- Default Branch: master
- Last Pushed: 2017-06-21T17:20:06.000Z (almost 8 years ago)
- Last Synced: 2025-01-17T05:41:42.120Z (5 months ago)
- Language: Python
- Homepage:
- Size: 4.47 MB
- Stars: 0
- Watchers: 2
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# char-rnn
A basic, single-layer character-level RNN based on Andrej Karpathy's "vanilla" minimal RNN. This is largely a class-based transcription of the 100-line gist provided here: https://gist.github.com/karpathy/d4dee566867f8291f086.
I am very intersted in making this code much better, even adapting it to a mutliple-layer RNN. While I will eventually have to go to a framework (like Tensorflow), I am interested in working framework-free for a while.
## Intention
I want to write a RNN that studies style and writes stylistically like a given source text. Right now, the routine hits a hard loss limit of about 39.0000, even after millions of iterations. This code is currently run on a VPS sans-GPU. As I wrote above, I am absolutely open to suggestions of how I can improve this up to the limit of needing a framework.