chrishkchris edited a comment on issue #685: URL: https://github.com/apache/singa/pull/685#issuecomment-617508136
One point to notice is that: Currently in SINGA, ReLU is stateless (except we record the input for backward propagation), so we can just use y = autograd.relu(y) without manually initializing a ReLU python object. On the other hand, if we initialize a ReLU instance manually, we have to the take the element [0] following https://github.com/apache/singa/blob/master/python/singa/autograd.py#L448 All in all, personally I suggest we directly use y = autograd.relu(y) instead ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected]
