Treebank sentiment RNN. Dataset is from http://nlp.stanford.edu/sentiment/.
Implementation is based off the paper "Recursive Deep Models for Semantic Compositionality Over a Sentiment Treebank" by Socher et. al that is available at http://nlp.stanford.edu/~socherr/EMNLP2013_RNTN.pdf.
This implementation is done with Python and NumPy. It requires: -- Python >= 2.7.9 -- NumPy >= 1.9
Training parameters (the learning rate alpha; strength of regularization lambda; and the number of training epochs) are specified in the main function of rnn.py. To train, type "python rnn.py". This will save a model under the folder RNN/Models for the given parameters.
Then update evaluation.py with this folder name and run "python evaluation.py" to evaluate the learned model on the test set.