13.3. Mean Squared Error (MSE)
- class fhez.nn.loss.mse.MSE
Loss function to node wrapper.
- backward(gradient)
Calculate MSE gradient with respect to \(\hat{y}\).
This will take a single gradient value, and return the average gradient with respect to \(\hat{y}\). If \(\hat{y}\) is more than 1 dim it will return a multidimensional array of values which are the average gradients in those dims.
\(\frac{d}{d\hat{y}}(\text{MSE})=\sum_{i=0}^{N-1} -2(y-\hat{y})\)
- property cost
Get computational cost of the forward function.
- forward(signal=None, y: Optional[numpy.ndarray] = None, y_hat: Optional[numpy.ndarray] = None)
Calculate the loss of the output given the ground truth.
This will take multiple values for both \(y\) and \(\hat{y}\), and return a single value that is the mean of their absolute difference.
\(\text{MSE}=\frac{\sum_{i=0}^{N-1} (y-\hat{y})^2 }{N}\)
- update()
Do nothing as there are no parameters to update.
- updates()
Do nothing as there are no parameters to update.