def loss_function(
self, inputs: Dict[str, tf.Tensor],
outputs: Dict[str,
tf.Tensor]) -> Tuple[tf.Tensor, Dict[str, tf.Tensor]]:
sequence_mask = rk.utils.convert_sequence_length_to_sequence_mask(
inputs['primary'], inputs['protein_length'])
valid_mask = tf.cast(inputs['valid_mask'], tf.float32)
sequence_mask = tf.cast(sequence_mask, tf.float32) * valid_mask
angle_loss = self.compute_angle_loss(inputs['phi'], inputs['psi'],
outputs['phi_pred'],
outputs['psi_pred'], valid_mask)
rsa_loss = tf.losses.mean_squared_error(inputs['rsa'],
outputs['rsa_pred'],
sequence_mask)
disorder_loss = tf.losses.sigmoid_cross_entropy(
inputs['disorder'], outputs['disorder_pred'], sequence_mask)
interface_loss = tf.losses.sigmoid_cross_entropy(
inputs['interface'], outputs['interface_pred'], sequence_mask)
ss3_loss, ss3_acc = classification_loss_and_accuracy(
inputs['ss3'], outputs['ss3_pred'], sequence_mask)
ss8_loss, ss8_acc = classification_loss_and_accuracy(
inputs['ss8'], outputs['ss8_pred'], sequence_mask)
loss = angle_loss + rsa_loss + disorder_loss + interface_loss + ss3_loss + ss8_loss
metrics = {'SS3ACC': ss3_acc, 'SS8ACC': ss8_acc}
return loss, metrics
def loss_function(
self, inputs: Dict[str, tf.Tensor],
outputs: Dict[str,
tf.Tensor]) -> Tuple[tf.Tensor, Dict[str, tf.Tensor]]:
labels = inputs[self._label_name]
logits = outputs[self._output_name]
if self._mask_name != 'sequence_mask':
mask = outputs[self._mask_name]
else:
mask = rk.utils.convert_sequence_length_to_sequence_mask(
labels, inputs['protein_length'])
loss, accuracy = classification_loss_and_accuracy(labels, logits, mask)
ece = tf.exp(loss)
probs = tf.nn.softmax(logits)
logp = tf.nn.log_softmax(logits)
perplexity = tf.exp(-tf.reduce_sum(probs * logp, -1))
weights = tf.ones_like(perplexity) * tf.cast(mask, perplexity.dtype)
perplexity = tf.reduce_sum(
perplexity * weights) / (tf.reduce_sum(weights) + 1e-10)
metrics = {
self.key_metric: accuracy,
'ECE': ece,
'Perplexity': perplexity
}
return loss, metrics
def loss_function(self,
inputs: Dict[str, tf.Tensor],
outputs: Dict[str, tf.Tensor]) -> Tuple[tf.Tensor, Dict[str, tf.Tensor]]:
label = inputs[self._label]
prediction = outputs[self._output_name]
loss, accuracy = classification_loss_and_accuracy(label, prediction)
metrics = {self.key_metric: accuracy}
return loss, metrics
def loss_function(
self, inputs: Dict[str, tf.Tensor],
outputs: Dict[str,
tf.Tensor]) -> Tuple[tf.Tensor, Dict[str, tf.Tensor]]:
labels = inputs[self._label_name]
logits = outputs[self._output_name]
if self._mask_name != 'sequence_mask':
mask = outputs[self._mask_name]
else:
mask = rk.utils.convert_sequence_length_to_sequence_mask(
labels, inputs['protein_length'])
loss, accuracy = classification_loss_and_accuracy(labels, logits, mask)
metrics = {self.key_metric: accuracy}
return loss, metrics