Closed githubnemo closed 8 years ago
Related: https://github.com/sherjilozair/ctc/issues/2 Test file: https://github.com/githubnemo/theano_ctc/blob/gradients/tests/test_ctc_gradients.py
Any computation on the gradient of the CTC loss is ignored:
tGrad = T.grad(T.mean(tCost), tsActs) tGrad2x = T.grad(2 * T.mean(tCost), tsActs)
Both operations will yield the same result.
Good catch, thanks. Fixed.
Related: https://github.com/sherjilozair/ctc/issues/2 Test file: https://github.com/githubnemo/theano_ctc/blob/gradients/tests/test_ctc_gradients.py
Any computation on the gradient of the CTC loss is ignored:
Both operations will yield the same result.