I was thinking about doing some new practice application and found this cool sequencer where you can click a grid to create music.
So the data in that application could be treated like an “image”, and one could set up a rating system where a beat could get a score lets say between 1 and 5. So what if we want to deploy a web interface to collect user feedback and retrain the model continuously? I did a sketch guessing what that could look like
When presented with a new beat the user can either rate it right away, or “correct it” and then rate. After a while the server retrains the model and hopefully the user will get better beats over time.
I still need to do some research to pull this idea off. If you want to help me out I have a few questions listed below.
- Identifying the problem. Would this be a regression problem? Im thinking to make predictions to floats between 0 and 1 (later I can bin that into 1-5 or whatever score when talking with the user)
- Given a model, I can predict unknown items, but how to SPAWN new beats from the model?
- What kind of deployment solution would you use for this problem? I have a PHP server with a working fast-ai installation but that was a bit of a pain to integrate with python and it lacks GPU. Can you recommend some service that could supply both web server and GPU out of the box for maybe a flask/django API? Im very open to suggestion, feel lost about this one.
Thank you for your interest!