def _loss_function(model, prediction, target):
loss = _theano.T.nnet.categorical_crossentropy(prediction, target)
pred = T.argmax(prediction)
targ = T.argmax(target)
indices = T.and_(T.neq(pred, targ), T.eq(targ, 1.0))
loss_ = loss * 5.0
loss = T.where(indices, loss_, loss)
return loss
def _loss_function(model, prediction, target):
loss = _theano.T.nnet.categorical_crossentropy(prediction, target)
pred = T.argmax(prediction)
targ = T.argmax(target)
indices = T.and_(T.neq(pred, targ), T.eq(targ, 1.0))
loss_ = loss * 5.0
loss = T.where(indices, loss_, loss)
return loss
def _loss_function(model, prediction, target):
inner_loss = T.abs_(target - prediction)
outer_loss = 1.0 - inner_loss
combined_loss = T.stack([inner_loss, outer_loss], axis=1)
return T.min(combined_loss, axis=1)
def _loss_function(model, prediction, target):
inner_loss = T.abs_(target - prediction)
outer_loss = 1.0 - inner_loss
combined_loss = T.stack([inner_loss, outer_loss], axis=1)
return T.min(combined_loss, axis=1)
