I've added the "debug" option, that allows for output compatibility with Mikolov's rnnlm when it's value is set to 2, as in Mikolov's case.
This is slow because it doesn't compute probabilities directly, but it keeps the log probability computation and then does exponentiation.
I've added the "debug" option, that allows for output compatibility with Mikolov's rnnlm when it's value is set to 2, as in Mikolov's case. This is slow because it doesn't compute probabilities directly, but it keeps the log probability computation and then does exponentiation.