1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116

// 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.

use crate::{ArgminOp, Error};
use serde::{Deserialize, Serialize};

/// This wraps an operator and keeps track of how often the cost, gradient and Hessian have been
/// computed and how often the modify function has been called. Usually, this is an implementation
/// detail unless a solver is needed within another solver (such as a line search within a gradient
/// descent method), then it may be necessary to wrap the operator in an OpWrapper.
#[derive(Serialize, Deserialize, Clone, Debug)]
pub struct OpWrapper<O: ArgminOp> {
    op: O,
    pub cost_func_count: u64,
    pub grad_func_count: u64,
    pub hessian_func_count: u64,
    pub modify_func_count: u64,
}

impl<O: ArgminOp> OpWrapper<O> {
    /// Constructor
    pub fn new(op: &O) -> Self {
        OpWrapper {
            op: op.clone(),
            cost_func_count: 0,
            grad_func_count: 0,
            hessian_func_count: 0,
            modify_func_count: 0,
        }
    }

    /// Calls the `apply` method of `op` and increments `cost_func_count`.
    pub fn apply(&mut self, param: &O::Param) -> Result<O::Output, Error> {
        self.cost_func_count += 1;
        self.op.apply(param)
    }

    /// Calls the `gradient` method of `op` and increments `gradient_func_count`.
    pub fn gradient(&mut self, param: &O::Param) -> Result<O::Param, Error> {
        self.grad_func_count += 1;
        self.op.gradient(param)
    }

    /// Calls the `hessian` method of `op` and increments `hessian_func_count`.
    pub fn hessian(&mut self, param: &O::Param) -> Result<O::Hessian, Error> {
        self.hessian_func_count += 1;
        self.op.hessian(param)
    }

    /// Calls the `modify` method of `op` and increments `modify_func_count`.
    pub fn modify(&mut self, param: &O::Param, extent: f64) -> Result<O::Param, Error> {
        self.modify_func_count += 1;
        self.op.modify(param, extent)
    }

    /// Consumes an operator by increasing the function call counts of `self` by the ones in
    /// `other`.
    pub fn consume_op(&mut self, other: OpWrapper<O>) {
        self.cost_func_count += other.cost_func_count;
        self.grad_func_count += other.grad_func_count;
        self.hessian_func_count += other.hessian_func_count;
        self.modify_func_count += other.modify_func_count;
    }

    /// Reset the cost function counts to zero
    pub fn reset(mut self) -> Self {
        self.cost_func_count = 0;
        self.grad_func_count = 0;
        self.hessian_func_count = 0;
        self.modify_func_count = 0;
        self
    }

    /// Returns the operator `op` by taking ownership of `self`.
    pub fn get_op(self) -> O {
        self.op
    }

    /// Returns a clone of the operator `op`.
    pub fn clone_op(&self) -> O {
        self.op.clone()
    }

    /// Creates a new `OpWrapper<O>` from another `OpWrapper<O>` by cloning the `op` and
    /// initializing all counts with `0`.
    pub fn new_from_op(op: &OpWrapper<O>) -> Self {
        Self::new(&op.clone_op())
    }
}

/// The OpWrapper<O> should behave just like any other `ArgminOp`
impl<O: ArgminOp> ArgminOp for OpWrapper<O> {
    type Param = O::Param;
    type Output = O::Output;
    type Hessian = O::Hessian;

    fn apply(&self, param: &Self::Param) -> Result<Self::Output, Error> {
        self.op.apply(param)
    }

    fn gradient(&self, param: &Self::Param) -> Result<Self::Param, Error> {
        self.op.gradient(param)
    }

    fn hessian(&self, param: &Self::Param) -> Result<Self::Hessian, Error> {
        self.op.hessian(param)
    }

    fn modify(&self, param: &Self::Param, extent: f64) -> Result<Self::Param, Error> {
        self.op.modify(param, extent)
    }
}