Brief comparison of PyTorch and Keras when working with Recurrent Neural Networks for sentiment classification in the English language.

├── scripts/
│   ├── data.csv
│   ├── keras_nlp.py
│   └── pytorch_nlp.py
└── docs/

The data are Twitter disaster feeds and therefore a binary classification problem. The last RNN version which was used (and uploaded here) are Gated Recurrent Units or GRUs.

Vanilla RNN	LSTM	GRU