remoteml / songnet Goto Github PK

Pick a genre and a decade to train an LSTM. If there just isn't enough data, use the entire corpus.

Perhaps instead of this, use transfer learning as per Slack...

Pick a matrix, any short matrix that already exists or that you make, and use the cell values to train the lyrics generator outputs.

Will this ask the generator to create a new lyric if a generated lyric doesn't work? How will this method modify the lyrics generator to let it know the requested word is the wrong choice? Does this become Reinforcement Learning? Is the SongSim like a Q matrix? Hmmm...

I have enabled Travis CI for this repo. To use it, create a .travis.yml file which you can use to run tests, linters and so on