A side project focused on generating text using a recurrent neural network
I wanted to see if I could generate a poem based on past poems I've written. A friend suggested I look into RNNs and I stumbled on this great package Following the instructions there, as well as this blog post I was able to get pretty decent results.
I built the model in google colab as suggested in the blogpost. The GPU there was a lot faster than my own CPU.
# Follow the install instructions for minimaxirs project
# Clone this repo
# Change the input textfile
python3
from textgenrnn import textgenrnn
textgen = textgenrnn()
textgen.train_from_file('test_input.txt', num_epochs=1)
textgen.generate()
If you want to import an already generated model (created in google colab for instance), use this code.
from textgenrnn import textgenrnn
textgen = textgenrnn(weights_path='colaboratory_weights.hdf5',
vocab_path='colaboratory_vocab.json',
config_path='colaboratory_config.json')