/sp2cp

Imageboard bot with recurrent neural network (RNN, GRU)

Primary LanguagePythonMIT LicenseMIT

sp2cp

WIP

Imageboard bot with recurrent neural network (RNN, GRU)

Бот с нейросетью, обучающейся генерировать комментарии на датасете из тредов с 2ch.hk

JUST FOR LULZ

Screenshot

Screenshot

Screenshot

https://drive.google.com/drive/folders/1Xmm5DTxQ8sJwL2ygefWTPFWsTSviMPDP

How to run

Collect dataset

python3 ./grub_threads.py --board=b

Prepare dataset for training

python3 ./parse_dataset.py --lexicon_limit=100000 --sequence_max_len=50

It will save tokenized dataset and id2token map in ./dataset/parsed/ directory.

Train

python3 ./train.py --epochs=20 --batch_size=100

After training you can find trained models and id2token map in ./models/ directory.

Run autoposter

python3 ./autoposter.py --weights_file=./models/weights.h5 --id2token_file=./models/id2token.json \
  --board=b --passcode=passcode --max_threads=30 \
  --max_res_len=30