Era1 Train Set Not IID

[pastorep]

Note: Many train set examples are collected from the same 1min audio clips and share many properties; thus, the examples are not independent (a.k.a. non-IID). The learning of these interdependencies does not translate well to the more generalized validation set. I'm currently working to make the train set examples IID by (1) bringing in Era0 train set and (2) minimizing number of examples from any one 1min audio clip (detection)