add thresholding parameter to classifiers

yaatehr / learnedbloomfilter

MIT dsail research project

8 stars 0 forks source link

Closed yaatehr closed 4 years ago

yaatehr commented 4 years ago

refactor classifiers so they can use log softmax, or sigmoid/tanh
gather statistics on the probablility of inputs being selected post threshold
make sure tau is a savable parameter, and that the forward method and prediction methods are different (loss should still be NLL)

yaatehr commented 4 years ago

commit 73a73fb6e63a9659e6690e0478036e743056557b