def forward(self, prediction: ndarray, target: ndarray) -> float:
"""Computes the actual loss value."""
assert_same_shape(prediction, target)
self.prediction = prediction
self.target = target
loss_value = self._output()
return loss_value
def backward(self, output_grad: ndarray) -> ndarray:
"""
Calls the self._input_grad() function.
Checks that the appropriate shapes match.
"""
assert_same_shape(self.output, output_grad)
self.input_grad = self._input_grad(output_grad)
assert_same_shape(self.input_, self.input_grad)
return self.input_grad
def backward(self, output_grad: ndarray) -> ndarray:
"""Passes output_grad backward through a series of operations."""
assert_same_shape(self.output, output_grad)
for operation in reversed(self.operations):
output_grad = operation.backward(output_grad)
input_grad = output_grad
self._param_grads()
return input_grad
def backward(self, output_grad: ndarray) -> ndarray:
"""
Calls self._input_grad and self._param_grad.
Checks appropriate shapes.
"""
assert_same_shape(self.output, output_grad)
self.input_grad = self._input_grad(output_grad)
self.param_grad = self._param_grad(output_grad)
assert_same_shape(self.input_, self.input_grad)
return self.input_grad
def backward(self) -> ndarray:
"""Computes gradient of the loss value with respect to the input to the loss function."""
self.input_grad = self._input_grad()
assert_same_shape(self.prediction, self.input_grad)
return self.input_grad