Metrics and monitors
Metrics are used to keep track of model performance, while monitors are used to keep track of model parameters. This page provides an overview of the metrics (and thus losses) you can use for supervised evolution, as well as the monitors you can use for both supervised and unsupervised evolution. For information on how to use metrics and monitors, view the documentation of the
func_evolve functions here.
Mean squared error (metric)
mean(square(y_true - y_pred))
Mean absolute error (metric)
mean(abs(y_true - y_pred))
Mean absolute percentage error (metric)
100 * mean(abs(y_true - y_pred) / (abs(y_true) + epsilon))
A small positive \epsilon is added to avoid division by zero. However, it is still not recommended to use this metric if your target output contains zeros.
Model size (monitor)
Total number of neurons in the model, including input and output neurons.