Club409
The best damn waste of time!
Musics
Shawn
Total posts: 1373
5/5/2017 10:47 PM
That visualization showing the model flow is pretty sexy.
I'd be curious how they went about processing and tagging the music data to train their model(s). That to me would be a big part of how well the model was certain it was classical/ragtime/... Not a trivial thing and I would guess the hardest part of their example. In my limited experience building ML models it is about 95% data cleaning/feature engineering and 5% model building. It's cool how good a model you can build with open source tools like keras in Python nowadays, but there is always an amount of outperformance to be gained from specific domain knowledge of the problem or better feature engineering (lots of depth here).
I'd be curious how they went about processing and tagging the music data to train their model(s). That to me would be a big part of how well the model was certain it was classical/ragtime/... Not a trivial thing and I would guess the hardest part of their example. In my limited experience building ML models it is about 95% data cleaning/feature engineering and 5% model building. It's cool how good a model you can build with open source tools like keras in Python nowadays, but there is always an amount of outperformance to be gained from specific domain knowledge of the problem or better feature engineering (lots of depth here).
Drew
Total posts: 5183
5/5/2017 9:32 PM
How much is the license? Like $20? Def got a role for it
mwinter
Total posts: 4364
5/5/2017 2:06 PM
Neural networks.. so hot right now