Multi-layer Recurrent Neural Networks (LSTM, RNN) for word-level language models in Python using TensorFlow.
Branch: master
Clone or download
Pull request Compare This branch is 4 commits ahead of hunkim:master.
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
data/tinyshakespeare
logs
save
tests
.gitignore
.travis.yml
LICENSE.md
README.md
beam.py
model.py
sample.py
train.py
utils.py

README.md

This has been forked directly from https://github.com/hunkim/word-rnn-tensorflow with the explicit intention of creating a 'Joe Posset Review AI'.

For full documentation and possible updates check out the source repo above.

Training Data

The training data, in data/tinyshakespeare/input.txt, consists of Joe's reviews from the Radio Free Midwich music blog. They were selected by searching "joe murray on", "posset on", and "rfm on" with false positives removed. These searches are unlikely to be perfect so some reviews may have been missed. The copy/pasted text was then edited to remove titles, headers, and band/purchase links. What's left is purely Joe's review text.

Usage Rights

The text was sourced from https://radiofreemidwich.wordpress.com, with permission for this specific use granted by Joe personally. No further rights to this text are granted or implied. Feel free to play around but don't be a c_nt.

The rest of the code falls under the MIT license as per the original authors.

Requirements

Basic Usage

To train with default parameters on the tinyshakespeare corpus, run:

python train.py

I have been generating text with the following command:

python sample.py --prime "Starting text" -n 250 --pick 2 --width 4

This gives the opening words (in prime), with a length of 250 words and uses 'beam search' (check out the original project for more info/options).

Sample output

# python sample.py --prime "Guns N' Roses - Appetite for Destruction" -n 150 --pick 2 --width 4

Guns N' Roses - Appetite for Destruction

Vibe Miles perfected on Kind of Blue. On ‘What all the kind of mille plateaux-shudder
to create a fitter, leaner guest-blogger. I was sprouting.” appear this is a
meta-narrative of sound squeal. A purgatory of sunshine. The inflammation of that
warms up the Hall & Oates that have originally functions and play it as a sound
of the ‘Spin/Off’ is no one of the world’s longest pendulum tape east to west across
the teeth. Track two is a audience burr and dry and isolated. If you just I can
be the sort of mille plateaux-shudder to be a fitter, leaner guest-blogger.
I was sprouting.” “That’s a fan that makes me all crying into my ears as the fan
of the ion drive, the picture. The prolific artist at the kind of traffic – they
baffle up for tea, jab this all I can.