Brother Grim fairytale generated by Neural network that used all 209 Brothers Grim Fairytales for training.
I continue to train the network and post stories on every epoch 50 interval (First 50 then 100, 150...) to see how the text changes. I do replace the generated ' and \n symbols with their correct representation, meaning ' = ' and \n = newline. Other than that I keep the text the same. The story lenght is set at 6878 characters since this is the average length for grim fairy tails.
The Story filename represents at which epoch they were created .txt, ex. 50.txt was created at epoch 50.
The File Last epoch.txt contains the last epoch after a training session, ex. before training 100 and after training 150 which then gets updated in this file.
You can calculate the training time in minutes by just: LastEpoch / 50 * 50m
- Graphics Card: NVIDIA GTX 770
- Epochs per Training session: 50
- Time per Training session: 50min
python train.py --data_dir=./data/grim --batch_size=50 --seq_length=100 --rnn_size=256 --init_from=./save
python sample.py -n=6878 --sample=1 >> story.txt
- Blog about RNN: http://karpathy.github.io/2015/05/21/rnn-effectiveness/
- Multi-layer Recurrent Neural Networks (LSTM, RNN) for character-level language models in Python using Tensorflow: https://github.com/sherjilozair/char-rnn-tensorflow
- This book contains 209 tales collected by the brothers Grimm: https://www.cs.cmu.edu/~spok/grimmtmp/