def calculate_loss(self, ff_dict):
###
# Compute the loss.
###
real_targets = ff_dict['real_targets']
cate_targets = ff_dict['cate_targets']
out = ff_dict['out_params']
loss_type = self.loss_type
if loss_type == 'ce':
loss = loss_func.ce_loss(out, cate_targets)
elif loss_type == 'mol':
quant_chann = self.quant_chann
loss = loss_func.mol_loss(out, real_targets, quant_chann)
else:
raise ValueError('[{}] loss is not supported.'.format(loss_type))
return {'loss': loss}
def calculate_loss(self, ff_dict):
###
# Compute the loss.
###
real_targets = ff_dict['real_targets']
cate_targets = ff_dict['cate_targets']
out = ff_dict['out_params']
loss_type = self.loss_type
if loss_type == 'ce':
loss = loss_func.ce_loss(out, cate_targets)
elif loss_type == 'mol':
quant_chann = self.quant_chann
loss = loss_func.mol_loss(out, real_targets, quant_chann)
elif loss_type == 'gauss':
loss = loss_func.gauss_loss(out, real_targets)
if DETAIL_LOG:
mean, std = loss_func.mean_std_from_out_params(
out, use_log_scales=True)
tf.summary.histogram('mean', mean)
tf.summary.histogram('std', std)
tf.summary.histogram('log_std', tf.log(std))
else:
raise ValueError('[{}] loss is not supported.'.format(loss_type))
return {'loss': loss}