minimaxir / Char Embeddings
Licence: mit
A repository containing 300D character embeddings derived from the GloVe 840B/300D dataset, and uses these embeddings to train a deep learning model to generate Magic: The Gathering cards using Keras
Stars: ✭ 199
Programming Languages
python
139335 projects - #7 most used programming language
char-embeddings
char-embeddings is a repository containing 300D character embeddings derived from the GloVe 840B/300D dataset, and uses these embeddings to train a deep learning model to generate Magic: The Gathering cards using Keras. The generation and model construction is heavily modified after the automatic text generation Keras example by François Chollet.
Usage
This repository contains a number of Python 3 scripts:
-
create_embeddings.py
: Converts a pretrained word embeddings file into a character embeddings file by averaging the per-character vectors. -
create_magic_text.py
: Converts an MTG JSON card dump into a one-per-line card encoding. -
text_generator_keras.py
: Constructs and trains the Keras model, producing Magic cards at each epoch. -
text_generator_keras_sample.py
: Using the text file and Keras model generated from the previous two scripts, generate a large amount of Magic cards.
The output
folder contains Magic card output at each epoch, a log of losses at every 50th batch, the learned character embeddings at the last epoch, the trained model itself, and a large sample of generated Magic cards.
Requirements
keras, tensorflow, h5py, scikit-learn
Maintainer
Max Woolf (@minimaxir)
License
MIT
Note that the project description data, including the texts, logos, images, and/or trademarks,
for each open source project belongs to its rightful owner.
If you wish to add or remove any projects, please contact us at [email protected].