RNNLyrics

###Recurrent Neural Networks for song lyrics prediction. Comparison of Character vs Word-level RNNs.

Authors: Anhad Mohananey, Yashovardhan Chaturvedi

In this project, Character-level RNNs have been compared to Word-level RNNs in terms of the task of song lyrics prediction. The Recurrent Neural Network is implemented in Python using the Keras Library.

For help in running and testing out the code, two sample lyrics corpus have been included: beatles_corpus.txt(Beatle's song lyrics) and acdc_corpus.txt(ACDC' song lyrics).

Parameters in the script :

• run_on_words : True : For running on words , False : For running on charecters

run_on_words = False 

• Parameters for script : Epochs : for defining the number of epochs to run , batch_size_value : For batch size for random sampling for training from the input , max_len : vector size for window sliding on input.

# Parameters to run the script :
epochs = 60  # define the number of epochs to fit the model on
batch_size_value = 128 #  values could be from 120 to len(token_list) - choosing 128 as default for running while on cpu
max_len = 20 # set a fixed vector size so that we can look at specific windows of characters

• corpus_file : defining the input file , change to ' acdc_corpus.txt ' to run on the provided ACDC songs file

corpus_file = open("beatles_corpus.txt", "r")