Function neuroflow::estimators::widrows

pub fn widrows(architecture: &[i32], allowed_error: f64) -> f64

Widrow's rule of thumb

This is an empirical rule that shows the size of training sample in order to get good generalization.

Example

For network architecture [2, 1] and allowed error 0.1 (10%) the size of training sample must exceed the amount of free network parameters in 10 times