def get_current_consistency_weight(final_consistency_weight, epoch,
step_in_epoch, total_steps_in_epoch):
# Consistency ramp-up from https://arxiv.org/abs/1610.02242
epoch = epoch - args.consistency_rampup_starts
epoch = epoch + step_in_epoch / total_steps_in_epoch
return final_consistency_weight * ramps.sigmoid_rampup(
epoch, args.consistency_rampup_ends - args.consistency_rampup_starts)
def get_current_consistency_weight(self, epoch):
# Consistency ramp-up from https://arxiv.org/abs/1610.02242
#unsupervised weight ramp-up function
"""
we noticed that optimization tended to explode during the ramp-up period, and we
eventually found that using a lower value for Adam β2 parameter (e.g., 0.99 instead of 0.999) seems
to help in this regard.
In our implementation, the unsupervised loss weighting function w(t) ramps up, starting from zero,
along a Gaussian curve during the first 80 training epochs. See Appendix A for further details about
this and other training parameters. In the beginning the total loss and the learning gradients are thus
dominated by the supervised loss component, i.e., the labeled data only. We have found it to be
very important that the ramp-up of the unsupervised loss component is slow enough—otherwise,
the network gets easily stuck in a degenerate solution where no meaningful classification of the data
is obtained.
"""
return self.args.consistency * ramps.sigmoid_rampup(
epoch, self.args.consistency_rampup)
def get_current_consistency_weight(epoch):
# Consistency ramp-up from https://arxiv.org/abs/1610.02242
return args.consistency * ramps.sigmoid_rampup(epoch,
args.consistency_rampup)
def get_current_consistency_weight(epoch):
# Consistency ramp-up from https://arxiv.org/abs/1610.02242
return args.consistency * ramps.sigmoid_rampup(epoch, args.consistency_rampup)