Closed am-gill closed 8 years ago
thanks for pointing this out. I myself do not have easy access to machines that can run this. Do you think this would be an easy fix? and can submit a pull request? thanks!
@asgill The bug is caused by this line: https://github.com/uoguelph-mlrg/theano_alexnet/blob/master/train.py#L92
You can fix this by moving this line after
load_momentums(vels, config['weights_dir'], epoch)
Also replace epoch
in the load_momentums
arguments to load_epoch
.
@hma02 thanks for looking into this! can u do a quick pull request?
@gwding just get a chance to fix this, see #24
When loading the saved state of the network, momentum is loaded for from one epoch higher than the loaded epoch