|
|
|
@ -74,3 +74,63 @@ impl<LayerOutput, Target, Loss> NeuraGradientSolverTransient<LayerOutput>
|
|
|
|
|
(epsilon_out, combine_gradients(layer_gradient, rec_gradient))
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
#[cfg(test)]
|
|
|
|
|
mod test {
|
|
|
|
|
use approx::assert_relative_eq;
|
|
|
|
|
|
|
|
|
|
use super::*;
|
|
|
|
|
use crate::{prelude::*, derivable::{activation::Tanh, loss::Euclidean, NeuraDerivable}, utils::uniform_vector};
|
|
|
|
|
|
|
|
|
|
#[test]
|
|
|
|
|
fn test_backprop_epsilon_bias() {
|
|
|
|
|
// Checks that the epsilon term from backpropagation is well applied, by inspecting the bias terms
|
|
|
|
|
// of the neural network's gradient
|
|
|
|
|
|
|
|
|
|
for _ in 0..100 {
|
|
|
|
|
let network = neura_sequential![
|
|
|
|
|
neura_layer!("dense", 4, f64).activation(Tanh),
|
|
|
|
|
neura_layer!("dense", 2, f64).activation(Tanh)
|
|
|
|
|
].construct(NeuraShape::Vector(4)).unwrap();
|
|
|
|
|
|
|
|
|
|
let optimizer = NeuraBackprop::new(Euclidean);
|
|
|
|
|
let input = uniform_vector(4);
|
|
|
|
|
let target = uniform_vector(2);
|
|
|
|
|
|
|
|
|
|
let layer1_intermediary = &network.layer.weights * &input;
|
|
|
|
|
let layer2_intermediary = &network.child_network.layer.weights * layer1_intermediary.map(|x| x.tanh());
|
|
|
|
|
|
|
|
|
|
assert_relative_eq!(layer1_intermediary.map(|x| x.tanh()), network.clone().trim_tail().eval(&input));
|
|
|
|
|
|
|
|
|
|
let output = network.eval(&input);
|
|
|
|
|
|
|
|
|
|
let gradient = optimizer.get_gradient(&network, &input, &target);
|
|
|
|
|
|
|
|
|
|
let mut delta2_expected = Euclidean.nabla(&target, &output);
|
|
|
|
|
for i in 0..2 {
|
|
|
|
|
delta2_expected[i] *= Tanh.derivate(layer2_intermediary[i]);
|
|
|
|
|
}
|
|
|
|
|
let delta2_actual = gradient.1.0.1;
|
|
|
|
|
|
|
|
|
|
assert_relative_eq!(delta2_actual.as_slice(), delta2_expected.as_slice());
|
|
|
|
|
|
|
|
|
|
let gradient2_expected = &delta2_expected * layer1_intermediary.map(|x| x.tanh()).transpose();
|
|
|
|
|
let gradient2_actual = gradient.1.0.0;
|
|
|
|
|
|
|
|
|
|
assert_relative_eq!(gradient2_actual.as_slice(), gradient2_expected.as_slice());
|
|
|
|
|
|
|
|
|
|
let mut delta1_expected = network.child_network.layer.weights.transpose() * delta2_expected;
|
|
|
|
|
for i in 0..4 {
|
|
|
|
|
delta1_expected[i] *= Tanh.derivate(layer1_intermediary[i]);
|
|
|
|
|
}
|
|
|
|
|
let delta1_actual = gradient.0.1;
|
|
|
|
|
|
|
|
|
|
assert_relative_eq!(delta1_actual.as_slice(), delta1_expected.as_slice());
|
|
|
|
|
|
|
|
|
|
let gradient1_expected = &delta1_expected * input.transpose();
|
|
|
|
|
let gradient1_actual = gradient.0.0;
|
|
|
|
|
|
|
|
|
|
assert_relative_eq!(gradient1_actual.as_slice(), gradient1_expected.as_slice());
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|