3 ms·
What I would love is for you to through the music through SampleCNN to reduce it to a vector representation, and then use that for smoothing. This could help sm
by bravura 7y ago
What I would love is for you to through the music through SampleCNN to reduce it to a vector representation, and then use that for smoothing. This could help smooth the cold start problem.
- jacobobryant 7y agoI'll look into that! I'm no machine learning expert, but I'll certainly be digging into it more as the project progresses. edit: so after a quick search for "what is smoothing in machine learning", I think I'm doing something similar, though basic, already. Basically each value in the matrix is a tuple of (completions + j, skips + k), where j and k are constants. e.g. "In listening sessions where A was completed, B was completed 3 times and skipped 4 times." But if j and k are 2, then the effective values would be (5, 6). I got this idea from my bayesian textbook.