A Recurrent Neural Network built with Tensorflow - generates H.P. Lovecraft-ian text using a model trained on the British horror author's corpus
scrape.py script grabs corpus (105 fiction titles) from:
https://hplovecraft.com - thank you!
lovecraft_charRNN_colab.ipynb is the notebook that I wrote and ran on google colab cloud platform (my own NVIDIA GPU at home has only compute ability 5.0, and does not train Gated Recurrent Unit models nearly as quickly as the GPUs/TPUs on the colab platform), to format the data (corpus) and train the model. This script then saves the model in .h5 format.
generate_char_based.py script uses the trained model (loaded from .h5
format), to generate text.
This Character-based Recurrent Neural Network is based on ideas and code from: "Hands-on Machine Learning with Scikit-Learn, Keras & Tensorflow", by Aurélien Géron
ricky@DOUGLAS-ENVY:~/Documents/code/lovecraft$ ./generate_char_based.py
the used, itsaeposing kind which vadiably resembled a
sane boys’ fragments, ahd thead fear church
upon memork; he saw the fellew about the clothing: “dr. rays 1857—the dra some chroarence
fear—south of castle and was paper stoning
it tuems on, and appro
Invoke with python3 generate_char_based.py to generate text. The she-bang at
top of my scripts points right to python3.9 - this will likely not work unless
you have that specific version installed. You can of course amend the she-bang
line and invoke directly i.e. ./generate_char_based.py. For convenience I
have saved the entire corpus (all.txt), and the trained-model in .h5 format, so
one does not need to run the scrape.py and lovecraft_charRNN_colab.ipynb
scripts prior to running generate_char_based.py. Please note there is a
word-based version I tried initially in the train.py script. Thanks! -Eric