neuronika 0.2.0

use super::{
    assert_almost_equals, new_backward_input, new_input, new_tensor, Backward, Data, Forward,
    Gradient, Multiplication, MultiplicationBackward, MultiplicationBackwardUnary, Overwrite,
    Tensor,
};

mod forward {
    use super::{
        assert_almost_equals, new_input, new_tensor, Data, Forward, Multiplication, Tensor,
    };

    #[test]
    fn creation() {
        let left = new_input((3, 3), vec![-4., -3., -2., -1., 0., 1., 2., 3., 4.]);
        let right = new_input((3, 3), vec![-1.; 9]);
        let node = Multiplication::new(left, right);

        assert_eq!(*node.data(), Tensor::from_elem((3, 3), 0.));
        assert_eq!(*node.data_mut(), Tensor::from_elem((3, 3), 0.));
        assert!(!node.was_computed());
    }

    #[test]
    fn computation_was_computed_transition() {
        let left = new_input((3, 3), vec![-4., -3., -2., -1., 0., 1., 2., 3., 4.]);
        let right = new_input((3, 3), vec![-1.; 9]);
        let node = Multiplication::new(left, right);

        node.forward();
        assert!(node.was_computed());

        node.forward();
        assert!(node.was_computed());

        node.reset_computation();
        assert!(!node.was_computed());

        node.reset_computation();
        assert!(!node.was_computed());
    }

    #[test]
    fn forward() {
        let left = new_input((3, 3), vec![-4., -3., -2., -1., 0., 1., 2., 3., 4.]);
        let right = new_input((3, 3), vec![-1.; 9]);
        let node = Multiplication::new(left, right.clone());

        // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ First Evaluation ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
        node.forward();
        assert_almost_equals(
            &*node.data(),
            &new_tensor((3, 3), vec![4., 3., 2., 1., 0., -1., -2., -3., -4.]),
        );

        // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ No Second Evaluation ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
        *right.data_mut() = new_tensor((3, 3), vec![2.; 9]);
        assert_almost_equals(&*right.data(), &new_tensor((3, 3), vec![2.; 9]));

        node.forward();
        assert_almost_equals(
            &*node.data(),
            &new_tensor((3, 3), vec![4., 3., 2., 1., 0., -1., -2., -3., -4.]),
        );

        // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Second Evaluation ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
        node.reset_computation();
        node.forward();
        assert_almost_equals(
            &*node.data(),
            &new_tensor((3, 3), vec![-8., -6., -4., -2., 0., 2., 4., 6., 8.]),
        );
    }

    #[test]
    fn left_broadcast_forward() {
        let left = new_input((1, 3), vec![-1., 0., 1.]);
        let right = new_input((2, 2, 3), vec![-2.; 12]);
        let node = Multiplication::new(left, right);

        assert_eq!(*node.data(), Tensor::from_elem((2, 2, 3), 0.));
        node.forward();
        assert_almost_equals(
            &*node.data(),
            &new_tensor(
                (2, 2, 3),
                vec![2., 0., -2., 2., 0., -2., 2., 0., -2., 2., 0., -2.],
            ),
        );
    }

    #[test]
    fn right_broadcast_forward() {
        let left = new_input((2, 2, 3), vec![-2.; 12]);
        let right = new_input((1, 3), vec![-1., 0., 1.]);
        let node = Multiplication::new(left, right);

        assert_eq!(*node.data(), Tensor::from_elem((2, 2, 3), 0.));
        node.forward();
        assert_almost_equals(
            &*node.data(),
            &new_tensor(
                (2, 2, 3),
                vec![2., 0., -2., 2., 0., -2., 2., 0., -2., 2., 0., -2.],
            ),
        );
    }

    #[test]
    fn debug() {
        let left = new_input(1, vec![0.]);
        let right = new_input(1, vec![0.]);
        let node = Multiplication::new(left, right);

        let output = "Multiplication { data: [0.0], shape=[1], strides=[1], layout=CFcf (0xf), const ndim=1, computed: false }";

        assert_eq!(output, format!("{:?}", node));
    }

    #[test]
    fn display() {
        let left = new_input(1, vec![0.]);
        let right = new_input(1, vec![0.]);
        let node = Multiplication::new(left, right);

        assert_eq!(format!("{}", node.data()), format!("{}", node));
    }
}
mod backward {
    use super::{
        assert_almost_equals, new_backward_input, new_input, new_tensor, Backward, Gradient,
        MultiplicationBackward, MultiplicationBackwardUnary, Overwrite, Tensor,
    };

    #[test]
    fn creation() {
        let node = MultiplicationBackward::new(
            new_input((3, 3), vec![3.; 9]),
            new_backward_input((3, 3), vec![0.; 9]),
            new_input((3, 3), vec![5.; 9]),
            new_backward_input((3, 3), vec![0.; 9]),
        );

        assert_eq!(*node.gradient(), Tensor::from_elem((3, 3), 0.));
        assert_eq!(*node.gradient_mut(), Tensor::from_elem((3, 3), 0.));
        assert!(node.can_overwrite());
    }

    #[test]
    fn computation_state_transition() {
        let lhs = new_backward_input((3, 3), vec![0.; 9]);
        let rhs = new_backward_input((3, 3), vec![0.; 9]);
        let node = MultiplicationBackward::new(
            new_input((3, 3), vec![3.; 9]),
            lhs.clone(),
            new_input((3, 3), vec![5.; 9]),
            rhs.clone(),
        );

        node.backward();
        assert!(node.can_overwrite());
        assert!(!lhs.can_overwrite());
        assert!(!rhs.can_overwrite());

        node.backward();
        assert!(node.can_overwrite());
        assert!(!lhs.can_overwrite());
        assert!(!rhs.can_overwrite());

        lhs.set_overwrite(true);
        assert!(node.can_overwrite());
        assert!(lhs.can_overwrite());
        assert!(!rhs.can_overwrite());

        lhs.set_overwrite(true);
        assert!(node.can_overwrite());
        assert!(lhs.can_overwrite());
        assert!(!rhs.can_overwrite());

        rhs.set_overwrite(true);
        assert!(node.can_overwrite());
        assert!(lhs.can_overwrite());
        assert!(rhs.can_overwrite());

        rhs.set_overwrite(true);
        assert!(node.can_overwrite());
        assert!(lhs.can_overwrite());
        assert!(rhs.can_overwrite());

        node.set_overwrite(false);
        assert!(!node.can_overwrite());
        assert!(lhs.can_overwrite());
        assert!(rhs.can_overwrite());

        node.set_overwrite(false);
        assert!(!node.can_overwrite());
        assert!(lhs.can_overwrite());
        assert!(rhs.can_overwrite());

        node.backward();
        assert!(!node.can_overwrite());
        assert!(!lhs.can_overwrite());
        assert!(!rhs.can_overwrite());

        node.backward();
        assert!(!node.can_overwrite());
        assert!(!lhs.can_overwrite());
        assert!(!rhs.can_overwrite());
    }

    #[test]
    fn backward() {
        let lhs = new_backward_input((3, 3), vec![0.; 9]);
        let rhs = new_backward_input((3, 3), vec![0.; 9]);
        let node = MultiplicationBackward::new(
            new_input((3, 3), vec![3.; 9]),
            lhs.clone(),
            new_input((3, 3), vec![5.; 9]),
            rhs.clone(),
        );

        // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Seed Gradient ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
        *node.gradient_mut() = new_tensor((3, 3), vec![1.; 9]);
        assert_almost_equals(&*node.gradient(), &new_tensor((3, 3), vec![1.; 9]));

        // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ First Evaluation ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
        node.backward();
        assert_almost_equals(&*lhs.gradient(), &new_tensor((3, 3), vec![5.; 9]));
        assert_almost_equals(&*rhs.gradient(), &new_tensor((3, 3), vec![3.; 9]));

        // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Second Evaluation ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
        node.backward();
        assert_almost_equals(&*lhs.gradient(), &new_tensor((3, 3), vec![10.; 9]));
        assert_almost_equals(&*rhs.gradient(), &new_tensor((3, 3), vec![6.; 9]));

        // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Third Evaluation ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
        lhs.set_overwrite(true);
        rhs.set_overwrite(true);
        node.backward();
        assert_almost_equals(&*lhs.gradient(), &new_tensor((3, 3), vec![5.; 9]));
        assert_almost_equals(&*rhs.gradient(), &new_tensor((3, 3), vec![3.; 9]));
    }

    #[test]
    fn backward_broadcast_left() {
        let lhs = new_backward_input(3, vec![0.; 3]);
        let rhs = new_backward_input((3, 3), vec![0.; 9]);
        let node = MultiplicationBackward::new(
            new_input(3, vec![3.; 3]),
            lhs.clone(),
            new_input((3, 3), vec![5.; 9]),
            rhs.clone(),
        );

        // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Seed Gradient ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
        *node.gradient_mut() = new_tensor((3, 3), vec![1.; 9]);
        assert_almost_equals(&*node.gradient(), &new_tensor((3, 3), vec![1.; 9]));

        // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ First Evaluation ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
        node.backward();
        assert_almost_equals(&*lhs.gradient(), &new_tensor(3, vec![15.; 3]));
        assert_almost_equals(&*rhs.gradient(), &new_tensor((3, 3), vec![3.; 9]));

        // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Second Evaluation ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
        node.backward();
        assert_almost_equals(&*lhs.gradient(), &new_tensor(3, vec![30.; 3]));
        assert_almost_equals(&*rhs.gradient(), &new_tensor((3, 3), vec![6.; 9]));

        // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Third Evaluation ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
        lhs.set_overwrite(true);
        rhs.set_overwrite(true);
        node.backward();
        assert_almost_equals(&*lhs.gradient(), &new_tensor(3, vec![15.; 3]));
        assert_almost_equals(&*rhs.gradient(), &new_tensor((3, 3), vec![3.; 9]));
    }

    #[test]
    fn backward_broadcast_right() {
        let lhs = new_backward_input((3, 3), vec![0.; 9]);
        let rhs = new_backward_input((1, 3), vec![0.; 3]);
        let node = MultiplicationBackward::new(
            new_input((3, 3), vec![3.; 9]),
            lhs.clone(),
            new_input((1, 3), vec![5.; 3]),
            rhs.clone(),
        );

        // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Seed Gradient ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
        *node.gradient_mut() = new_tensor((3, 3), vec![1.; 9]);
        assert_almost_equals(&*node.gradient(), &new_tensor((3, 3), vec![1.; 9]));

        // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ First Evaluation ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
        node.backward();
        assert_almost_equals(&*lhs.gradient(), &new_tensor((3, 3), vec![5.; 9]));
        assert_almost_equals(&*rhs.gradient(), &new_tensor((1, 3), vec![9.; 3]));

        // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Second Evaluation ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
        node.backward();
        assert_almost_equals(&*lhs.gradient(), &new_tensor((3, 3), vec![10.; 9]));
        assert_almost_equals(&*rhs.gradient(), &new_tensor((1, 3), vec![18.; 3]));

        // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Third Evaluation ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
        lhs.set_overwrite(true);
        rhs.set_overwrite(true);
        node.backward();
        assert_almost_equals(&*lhs.gradient(), &new_tensor((3, 3), vec![5.; 9]));
        assert_almost_equals(&*rhs.gradient(), &new_tensor((1, 3), vec![9.; 3]));
    }

    #[test]
    fn backward_unary() {
        let diff = new_backward_input((3, 3), vec![0.; 9]);
        let node = MultiplicationBackwardUnary::new(diff.clone(), new_input((3, 3), vec![5.; 9]));

        // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Seed Gradient ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
        *node.gradient_mut() = new_tensor((3, 3), vec![1.; 9]);
        assert_almost_equals(&*node.gradient(), &new_tensor((3, 3), vec![1.; 9]));

        // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ First Evaluation ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
        node.backward();
        assert_almost_equals(&*diff.gradient(), &new_tensor((3, 3), vec![5.; 9]));

        // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Second Evaluation ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
        node.backward();
        assert_almost_equals(&*diff.gradient(), &new_tensor((3, 3), vec![10.; 9]));

        // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Third Evaluation ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
        diff.set_overwrite(true);
        node.backward();
        assert_almost_equals(&*diff.gradient(), &new_tensor((3, 3), vec![5.; 9]));
    }

    #[test]
    fn backward_unary_broadcast() {
        let diff = new_backward_input(3, vec![0.; 3]);
        let node = MultiplicationBackwardUnary::new(diff.clone(), new_input((3, 3), vec![5.; 9]));

        // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Seed Gradient ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
        *node.gradient_mut() = new_tensor((3, 3), vec![1.; 9]);
        assert_almost_equals(&*node.gradient(), &new_tensor((3, 3), vec![1.; 9]));

        // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ First Evaluation ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
        node.backward();
        assert_almost_equals(&*diff.gradient(), &new_tensor(3, vec![15.; 3]));

        // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Second Evaluation ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
        node.backward();
        assert_almost_equals(&*diff.gradient(), &new_tensor(3, vec![30.; 3]));

        // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Third Evaluation ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
        diff.set_overwrite(true);
        node.backward();
        assert_almost_equals(&*diff.gradient(), &new_tensor(3, vec![15.; 3]));
    }

    #[test]
    fn no_grad() {
        // MultiplicationBackward
        let node = MultiplicationBackward::new(
            new_input((3, 3), vec![0.; 9]),
            new_backward_input((3, 3), vec![0.; 9]),
            new_input((3, 3), vec![0.; 9]),
            new_backward_input((3, 3), vec![0.; 9]),
        );

        node.no_grad();
        assert!(node.gradient.borrow().is_none());

        node.with_grad();
        assert_eq!(&*node.gradient(), Tensor::zeros(node.shape));

        // MultiplicationBackwardUnary
        let node = MultiplicationBackwardUnary::new(
            new_backward_input((3, 3), vec![0.; 9]),
            new_input((3, 3), vec![0.; 9]),
        );

        node.no_grad();
        assert!(node.gradient.borrow().is_none());

        node.with_grad();
        assert_eq!(&*node.gradient(), Tensor::zeros(node.shape));
    }

    #[test]
    fn debug() {
        {
            let node = MultiplicationBackward::new(
                new_input(1, vec![0.]),
                new_backward_input(1, vec![0.]),
                new_input(1, vec![0.]),
                new_backward_input(1, vec![0.]),
            );

            let output = "MultiplicationBackward { gradient: Some([0.0], shape=[1], strides=[1], layout=CFcf (0xf), const ndim=1), overwrite: true }";
            assert_eq!(output, format!("{:?}", node));
        }
    }

    #[test]
    fn debug_unary() {
        let node = MultiplicationBackwardUnary::new(
            new_backward_input(1, vec![0.]),
            new_input(1, vec![0.]),
        );

        let output = "MultiplicationBackwardUnary { gradient: Some([0.0], shape=[1], strides=[1], layout=CFcf (0xf), const ndim=1), overwrite: true }";
        assert_eq!(output, format!("{:?}", node));
    }

    #[test]
    fn display() {
        {
            let node = MultiplicationBackward::new(
                new_input(1, vec![0.]),
                new_backward_input(1, vec![0.]),
                new_input(1, vec![0.]),
                new_backward_input(1, vec![0.]),
            );
            assert_eq!(format!("{}", node.gradient()), format!("{}", node));
        }
    }

    #[test]
    fn display_unary() {
        let node = MultiplicationBackwardUnary::new(
            new_backward_input(1, vec![0.]),
            new_input(1, vec![0.]),
        );
        assert_eq!(format!("{}", node.gradient()), format!("{}", node));
    }
}