playground
playground copied to clipboard
Parameter sweep?
I know there are a lot of parameters to play with, but not too many to try a rough 'sweep' of all settings... I'm curious to try to categorise the results into working and non-working (converging and non-converging) to try to get some intuition about what works well in which situations.
Have you done this?
Dear @dbolser, by reading your issue it seems that you may appreciate the fork we've been working on !
https://coolearning.github.io/playground/
The Coolearning interface allows interactive sweeping of the parameters by using MIDI user interface during training, and much more ! Like individual learning rate, activation functions and regularization for each neuron...
See you there :)