rapGod()

Inspiration: We wanted to learn tensorflow and the concepts of machine learning in a fun way

What it does: Writes rap songs using a trained neural net model

How we built it: We used python3 with the Spotify and Genius Lyrics APIs to scrape many rap song lyrics. From there we created a LSTM Recurrent Neural Net in python and keras. We packaged our code and sent it to the GCP ML API for learning. After learning, we used python to run the model.

Challenges we ran into: Optimizer functions continually driving loss up and accuracy down during learning. Learning the correct syntax to run GCP properly. Making sure our data set did not contain any contamination.

Accomplishments that we're proud of: After training for 25 epochs the model can write semi-intelligible sentences with clear 'rap themes'

What we learned: How to set up a LSTM RNN with keras. How to integrate data and computation with GCP using the storage and ML APIs

What's next for rapGod(): More epochs learning and tweaking of the optimizer function to develop a more 'human like' model

Built With

• google-cloud
• python