|
1 //! Type definitions and re-exports |
|
2 |
|
3 use numeric_literals::replace_float_literals; |
|
4 |
|
5 use colored::ColoredString; |
|
6 use serde::{Serialize, Deserialize}; |
|
7 use clap::ValueEnum; |
|
8 use alg_tools::iterate::LogRepr; |
|
9 use alg_tools::euclidean::Euclidean; |
|
10 use alg_tools::norms::{Norm, L1}; |
|
11 |
|
12 pub use alg_tools::types::*; |
|
13 pub use alg_tools::loc::Loc; |
|
14 pub use alg_tools::sets::Cube; |
|
15 |
|
16 use crate::measures::DiscreteMeasure; |
|
17 |
|
18 /// [`Float`] with extra display and string conversion traits such that [`clap`] doesn't choke up. |
|
19 pub trait ClapFloat : Float |
|
20 + std::str::FromStr<Err=std::num::ParseFloatError> |
|
21 + std::fmt::Display {} |
|
22 impl ClapFloat for f32 {} |
|
23 impl ClapFloat for f64 {} |
|
24 |
|
25 /// Structure for storing iteration statistics |
|
26 #[derive(Debug, Clone, Serialize)] |
|
27 pub struct IterInfo<F : Float, const N : usize> { |
|
28 /// Function value |
|
29 pub value : F, |
|
30 /// Number of speaks |
|
31 pub n_spikes : usize, |
|
32 /// Number of iterations this statistic covers |
|
33 pub this_iters : usize, |
|
34 /// Number of spikes removed by merging since last IterInfo statistic |
|
35 pub merged : usize, |
|
36 /// Number of spikes removed by pruning since last IterInfo statistic |
|
37 pub pruned : usize, |
|
38 /// Number of inner iterations since last IterInfo statistic |
|
39 pub inner_iters : usize, |
|
40 /// Current tolerance |
|
41 pub ε : F, |
|
42 /// Strict tolerance update if one was used |
|
43 pub maybe_ε1 : Option<F>, |
|
44 /// Solve fin.dim problem for this measure to get the optimal `value`. |
|
45 pub postprocessing : Option<DiscreteMeasure<Loc<F, N>, F>>, |
|
46 } |
|
47 |
|
48 impl<F, const N : usize> LogRepr for IterInfo<F, N> where F : LogRepr + Float { |
|
49 fn logrepr(&self) -> ColoredString { |
|
50 let eqsign = match self.maybe_ε1 { |
|
51 Some(ε1) if ε1 < self.ε => '≛', |
|
52 _ => '=', |
|
53 }; |
|
54 format!("{}\t| N = {}, ε {} {:.8}, inner_iters_mean = {}, merged+pruned_mean = {}+{}", |
|
55 self.value.logrepr(), |
|
56 self.n_spikes, |
|
57 eqsign, |
|
58 self.ε, |
|
59 self.inner_iters as float / self.this_iters as float, |
|
60 self.merged as float / self.this_iters as float, |
|
61 self.pruned as float / self.this_iters as float, |
|
62 ).as_str().into() |
|
63 } |
|
64 } |
|
65 |
|
66 /// Branch and bound refinement settings |
|
67 #[derive(Clone, Copy, Eq, PartialEq, Serialize, Deserialize, Debug)] |
|
68 #[serde(default)] |
|
69 pub struct RefinementSettings<F : Float> { |
|
70 /// Function value tolerance multiplier for bisection tree refinement in |
|
71 /// [`alg_tools::bisection_tree::BTFN::maximise`] and related functions. |
|
72 pub tolerance_mult : F, |
|
73 /// Maximum branch and bound steps |
|
74 pub max_steps : usize, |
|
75 } |
|
76 |
|
77 #[replace_float_literals(F::cast_from(literal))] |
|
78 impl<F : Float> Default for RefinementSettings<F> { |
|
79 fn default() -> Self { |
|
80 RefinementSettings { |
|
81 tolerance_mult : 0.1, |
|
82 max_steps : 50000, |
|
83 } |
|
84 } |
|
85 } |
|
86 |
|
87 /// Data term type |
|
88 #[derive(Clone, Copy, Eq, PartialEq, Serialize, Deserialize, Debug, ValueEnum)] |
|
89 pub enum DataTerm { |
|
90 /// $\\|z\\|\_2^2/2$ |
|
91 L2Squared, |
|
92 /// $\\|z\\|\_1$ |
|
93 L1, |
|
94 } |
|
95 |
|
96 impl DataTerm { |
|
97 /// Calculate the data term value at residual $z=Aμ - b$. |
|
98 pub fn value_at_residual<F : Float, E : Euclidean<F> + Norm<F, L1>>(&self, z : E) -> F { |
|
99 match self { |
|
100 Self::L2Squared => z.norm2_squared_div2(), |
|
101 Self::L1 => z.norm(L1), |
|
102 } |
|
103 } |
|
104 } |
|
105 |