dainas-rnn

Multi-layer Recurrent Neural Networks (LSTM, RNN) for word-level language models in Python using TensorFlow. Mostly reused code from hunkim/word-rnn-tensorflow which was inspired from sherjilozair/char-rnn-tensorflow and Andrej Karpathy's char-rnn.

Example output

Jauni sēju Trīs dēli,
Jāņa ir pate ozolos.
Jau Bitītēm Abi tavā tīrumā:
Kam tev vakarā.

Ne asu Solīj'
Trūka, mirdz rillināja .
Vienam māti meita, čīkstēdama,
Tur Jānīša āboliņu.

Dependencies

• [Python 3] (https://www.python.org/)
• [Tensorflow] (https://www.tensorflow.org/)
• [NodeJS] (https://nodejs.org/en/)

Dataset extraction

• Run npm install in crawler directory
• Run node crawler.js in crawler directory to crawl for dainas
• Run node agregator.js in crawler directory to agregate dainas into one file
• Copy out.txt to main directory as input.txt

Training

Run python train.py --num_epochs=150 --data_dir=""

Generation

Run python sample.py