@karpathy @fchollet @sedielem @_mayfa @j26774 comparing char-rnn models, very different results for keras & lasagne!pic.twitter.com/dkynepPH9e
-
switched backend to tf, no significant change. but this other tf implementation has the correct behavior https://github.com/sherjilozair/char-rnn-tensorflow …
there is necessarily something wrong with your model definition in that case.
-
-
i totally agree. i need to understand where parameters/functions are missing in keras/lasagne vs tf/torch/chainer implementations
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
@sedielem fwiw, i noticed disagreement about rmsprop divisor. lasagne&keras use sqrt(a+1e-6). torch's sqrt(a)+1e-8 seems better? -
not big enough to account for this bug, but it does make a consistent difference across a few tests.
End of conversation
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.