argmin 0.0.9

// Copyright 2018 Stefan Kroboth
//
// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
// http://apache.org/licenses/LICENSE-2.0> or the MIT license <LICENSE-MIT or
// http://opensource.org/licenses/MIT>, at your option. This file may not be
// copied, modified, or distributed except according to those terms.

//! Optimizaton toolbox
//!
//! TODOs
//!
//! * Stopping criterions which can be stacked, also return the reason why a computation terminated
//! * keep track of cost function values
//! * count the number of cost function / gradient evaluations and return them

#![recursion_limit = "1024"]
#![cfg_attr(feature = "clippy", feature(plugin))]
#![cfg_attr(feature = "clippy", plugin(clippy))]
#![warn(missing_docs)]
#![feature(trace_macros)]
#[macro_use]
extern crate error_chain;
extern crate ndarray;
extern crate ndarray_linalg;
extern crate num;
extern crate rand;

/// Macros
#[macro_use]
pub mod macros;

use std::default::Default;
use num::{Bounded, ToPrimitive};
use errors::*;
use parameter::ArgminParameter;

/// Trait for cost function values
/// TODO: Do this with trait aliases once they work in rust.
pub trait ArgminCostValue: Bounded + ToPrimitive + Copy + Default + PartialOrd {}
impl<T> ArgminCostValue for T
where
    T: Bounded + ToPrimitive + Copy + Default + PartialOrd,
{
}

/// Trait every solve needs to implement (in the future)
pub trait ArgminSolver<'a> {
    /// Parameter vector
    type Parameter: ArgminParameter;
    /// Cost value
    type CostValue: ArgminCostValue;
    /// Hessian
    type Hessian;
    /// Initial parameter(s)
    type StartingPoints;
    /// Type of Problem
    type ProblemDefinition;

    /// Initializes the solver and sets the state to its initial state
    // fn init(&mut self, &'a Self::ProblemDefinition, &Self::StartingPoints) -> Result<()>;
    fn init(&mut self, Self::ProblemDefinition, &Self::StartingPoints) -> Result<()>;

    /// Moves forward by a single iteration
    fn next_iter(&mut self) -> Result<ArgminResult<Self::Parameter, Self::CostValue>>;

    /// Run initialization and iterations at once
    fn run(
        &mut self,
        // &'a Self::ProblemDefinition,
        Self::ProblemDefinition,
        &Self::StartingPoints,
    ) -> Result<ArgminResult<Self::Parameter, Self::CostValue>>;

    /// Handles the stopping criteria
    fn terminate(&self) -> TerminationReason;
}

/// Definition of all relevant traits
pub mod prelude;

/// Definition of the return type of the solvers
pub mod result;

/// Traits for implementing parameter vectors
pub mod parameter;

/// Problem formulation
pub mod problem;

/// Operator
pub mod operator;

/// Termination reasons
pub mod termination;

/// A set of test functions like Rosenbrock's function and so on.
pub mod testfunctions;

/// Backtracking line search
pub mod backtracking;

/// Simulated Annealing
pub mod sa;

/// Gradient Descent
pub mod gradientdescent;

/// Nelder Mead method
pub mod neldermead;

/// Newton method
pub mod newton;

/// Landweber algorithm
pub mod landweber;

/// Conjugate Gradient method
pub mod cg;

/// Errors using `error-chain`
mod errors;

/// Bring some structs into scope to make them easier to access
pub use problem::ArgminProblem;
pub use operator::ArgminOperator;
pub use result::ArgminResult;
pub use sa::SimulatedAnnealing;
pub use sa::SATempFunc;
pub use newton::Newton;
pub use cg::ConjugateGradient;
pub use gradientdescent::GradientDescent;
pub use gradientdescent::GDGammaUpdate;
pub use backtracking::BacktrackingLineSearch;
pub use landweber::Landweber;
pub use neldermead::NelderMead;
pub use termination::TerminationReason;

// #[cfg(test)]
// mod tests {
//     #[test]
//     fn it_works() {
//         assert_eq!(2 + 2, 4);
//     }
// }