use crate::backend::Backend;
use crate::cost::Cost;
use ndarray::Dimension;
use serde::{Deserialize, Serialize};
#[derive(Clone, Copy, Serialize, Deserialize)]
pub struct MSE;
impl<B: Backend> Cost<B> for MSE {
fn loss<D: Dimension>(&self, predicted: &B::Tensor<D>, target: &B::Tensor<D>) -> f32 {
let diff = B::sub(predicted, target);
B::mean(&B::mul(&diff, &diff)).unwrap_or(0.0)
}
fn gradient<D: Dimension>(
&self,
predicted: &B::Tensor<D>,
target: &B::Tensor<D>,
) -> B::Tensor<D> {
B::sub(predicted, target)
}
}