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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381
use itertools::{izip, Itertools};
use rand::distributions::uniform::SampleUniform;
use std::{
convert::TryInto,
f64, fmt,
ops::{Add, AddAssign, Div, Mul, Range, Sub},
sync::{
atomic::{AtomicBool, AtomicU64, AtomicU8, Ordering},
Arc, Mutex,
},
thread,
time::Duration,
};
use crate::util::{poll, Polling};
/// Castes all given ranges to `f64` values and calls [`grid_search()`].
/// ```
/// use std::{sync::Arc,time::Duration};
/// use simple_optimization::{grid_search, Polling};
/// fn simple_function(list: &[f64; 3], _: Option<Arc<()>>) -> f64 { list.iter().sum() }
/// let best = grid_search!(
/// (0f64..10f64, 5u32..15u32, 10i16..20i16), // Value ranges.
/// simple_function, // Evaluation function.
/// None, // No additional evaluation data.
/// // Polling every `10ms`, printing progress (`true`), exiting early if `15.` or less is reached, and not printing thread execution data (`false`).
/// Some(Polling { poll_rate: Duration::from_millis(5), printing: true, early_exit_minimum: Some(15.), thread_execution_reporting: false }),
/// None, // We don't specify the number of threads.
/// // Take `10` samples along range `0` (`0..10`), `11` along range `1` (`5..15`)
/// // and `12` along range `2` (`10..20`).
/// // In total taking `10*11*12=1320` samples.
/// [10,11,12],
/// );
/// assert_eq!(simple_function(&best, None), 15.);
/// ```
/// Due to specific design the `threads` parameter is excluded for now.
#[macro_export]
macro_rules! grid_search {
(
// Generic
($($x:expr),*),
$f: expr,
$evaluation_data: expr,
$polling: expr,
$threads: expr,
// Specific
$points: expr,
) => {
{
use num::ToPrimitive;
let mut ranges = [
$(
$x.start.to_f64().unwrap()..$x.end.to_f64().unwrap(),
)*
];
simple_optimization::grid_search(
ranges,
$f,
$evaluation_data,
$polling,
$threads,
$points,
)
}
};
}
/// [Grid search](https://en.wikipedia.org/wiki/Hyperparameter_optimization#Grid_search)
///
/// Evaluate all combinations of values from the 3 values where:
/// - Value 1 covers `10` values at equal intervals from `0..10` (`0,1,2,3,4,5,6,7,8,9`).
/// - Value 2 covers `11` values at equal intervals from `5..15`.
/// - Value 3 covers `12` values at equal intervals from `10..20`.
///
/// Printing progress every `10ms` and exiting early if a value is found which is less than or equal to `15.`.
/// ```
/// use std::{sync::Arc,time::Duration};
/// use simple_optimization::{grid_search, Polling};
/// fn simple_function(list: &[f64; 3], _: Option<Arc<()>>) -> f64 { list.iter().sum() }
/// let best = grid_search(
/// [0f64..10f64, 5f64..15f64, 10f64..20f64], // Value ranges.
/// simple_function, // Evaluation function.
/// None, // No additional evaluation data.
/// // Polling every `10ms`, printing progress (`true`), exiting early if `15.` or less is reached, and not printing thread execution data (`false`).
/// Some(Polling { poll_rate: Duration::from_millis(5), printing: true, early_exit_minimum: Some(15.), thread_execution_reporting: false }),
/// None, // We don't specify the number of threads.
/// // Take `10` samples along range `0` (`0..10`), `11` along range `1` (`5..15`)
/// // and `12` along range `2` (`10..20`).
/// // In total taking `10*11*12=1320` samples.
/// [10,11,12],
/// );
/// assert_eq!(simple_function(&best, None), 15.);
/// ```
/// Due to specific design the `threads` parameter is excluded for now.
pub fn grid_search<
A: 'static + Send + Sync,
T: 'static
+ Copy
+ Send
+ Sync
+ Default
+ fmt::Debug
+ SampleUniform
+ PartialOrd
+ AddAssign
+ Add<Output = T>
+ Sub<Output = T>
+ Div<Output = T>
+ Mul<Output = T>
+ num::FromPrimitive,
const N: usize,
>(
// Generics
ranges: [Range<T>; N],
f: fn(&[T; N], Option<Arc<A>>) -> f64,
evaluation_data: Option<Arc<A>>,
polling: Option<Polling>,
threads: Option<usize>,
// Specifics
points: [u64; N],
) -> [T; N] {
// Gets cpu number
let cpus = crate::cpus!(threads);
// 1 cpu is used for polling (this one), so we have -1 cpus for searching.
let search_cpus = cpus - 1;
// Computes points per thread
let mut remainder = [Default::default(); N];
let mut per = [Default::default(); N];
for i in 0..N {
remainder[i] = points[i] % search_cpus as u64;
per[i] = std::cmp::max(points[i] / search_cpus as u64, 1);
}
// println!("remainder: {:?}, per: {:?}",remainder,per);
// Points ranges for remainder
// ---------------------------------------
let remainder_ranges: [Range<u64>; N] = remainder
.iter()
.map(|&r| 0..r)
.collect::<Vec<_>>()
.try_into()
.unwrap();
// Points ranges per thread
// ---------------------------------------
// If at any point the threads need to evaluate more than 1 value.
let some_thread_work = per.iter().any(|&x| x > 1);
// We effectively fold over our threads for each point range
let per_ranges_opt: Option<Vec<[Range<u64>; N]>> = some_thread_work.then(|| {
let mut offset = [Default::default(); N];
// Initial offset is after remainder
for i in 0..N {
offset[i] = remainder_ranges[i].end;
}
(0..search_cpus)
.map(|_| {
(0..N)
.map(|i| {
let new = offset[i]..offset[i] + per[i];
offset[i] = new.end;
new
})
.collect::<Vec<_>>()
.try_into()
.unwrap()
})
.collect::<Vec<_>>()
});
// println!("remainder_ranges: {:?}, per_ranges_opt: {:?}",remainder_ranges,per_ranges_opt);
// Checks ranges
// ---------------------------------------
// Number of evaluations all the threads do.
let mut iterations = 0;
// Number of evaluations the remainder does.
let mut remainder = 0;
for i in 0..N {
// Gets points covered by remainder
let remainder_point_sum = remainder_ranges[i].end - remainder_ranges[i].start;
remainder += remainder_point_sum;
// Gets points covered by threads
let point_sum = per_ranges_opt.as_ref().map_or(0, |per_ranges| {
per_ranges
.iter()
.fold(0, |acc, x| acc + x[i].end - x[i].start)
});
iterations += point_sum;
// Checks sum
assert_eq!(
remainder_point_sum + point_sum,
points[i],
"remainder: {:?}, threads: {:?}",
remainder_ranges,
per_ranges_opt
);
}
// Compute step sizes
// ---------------------------------------
let mut steps = [Default::default(); N];
for (r, k, s) in izip!(ranges.iter(), points.iter(), steps.iter_mut()) {
*s = (r.end - r.start) / T::from_u64(*k).unwrap();
}
// Covers remainder section
// ---------------------------------------
let ranges_arc = Arc::new(ranges);
let (best_value, mut best_params) = search(
// Generics
ranges_arc.clone(),
f,
evaluation_data.clone(),
// Since we are doing this on the same thread, we don't need to use these
Arc::new(AtomicU64::new(Default::default())),
Arc::new(Mutex::new(Default::default())),
Arc::new(AtomicBool::new(false)),
Arc::new(AtomicU8::new(0)),
Arc::new([]),
// Specifics
remainder_ranges,
steps,
);
// println!("completed remainder: {}",best_value);
// Threads
// ---------------------------------------
if let Some(per_ranges) = per_ranges_opt {
let thread_exit = Arc::new(AtomicBool::new(false));
let (handles, links): (Vec<_>, Vec<_>) = (0..search_cpus)
.zip(per_ranges.into_iter())
.map(|(_, per_ranges)| {
let ranges_clone = ranges_arc.clone();
let counter = Arc::new(AtomicU64::new(0));
let thread_best = Arc::new(Mutex::new(f64::MAX));
let thread_execution_position = Arc::new(AtomicU8::new(0));
let thread_execution_time = Arc::new([]);
let counter_clone = counter.clone();
let thread_best_clone = thread_best.clone();
let thread_exit_clone = thread_exit.clone();
let evaluation_data_clone = evaluation_data.clone();
let thread_execution_position_clone = thread_execution_position.clone();
let thread_execution_time_clone = thread_execution_time.clone();
(
thread::spawn(move || {
search(
// Generics
ranges_clone,
f,
evaluation_data_clone,
counter_clone,
thread_best_clone,
thread_exit_clone,
thread_execution_position_clone,
thread_execution_time_clone,
// Specifics
per_ranges,
steps,
)
}),
(
counter,
(
thread_best,
(thread_execution_position, thread_execution_time),
),
),
)
})
.unzip();
let (counters, links): (Vec<Arc<AtomicU64>>, Vec<_>) = links.into_iter().unzip();
let (thread_bests, links): (Vec<Arc<Mutex<f64>>>, Vec<_>) = links.into_iter().unzip();
let (thread_execution_positions, thread_execution_times) = links.into_iter().unzip();
if let Some(poll_data) = polling {
poll(
poll_data,
counters,
remainder,
iterations,
thread_bests,
thread_exit,
thread_execution_positions,
thread_execution_times,
);
}
// Joins all handles and folds across extracting the best value and best points.
let (new_best_value, new_best_params) = handles
.into_iter()
.map(|h| h.join().unwrap())
.fold((best_value, best_params), |(bv, bp), (v, p)| {
if v < bv {
(v, p)
} else {
(bv, bp)
}
});
// If the best value from threads is better than the value from remainder
if new_best_value < best_value {
best_params = new_best_params
}
}
return best_params;
fn search<
A: 'static + Send + Sync,
T: 'static
+ Copy
+ Send
+ Sync
+ Default
+ fmt::Debug
+ SampleUniform
+ PartialOrd
+ AddAssign
+ Add<Output = T>
+ Sub<Output = T>
+ Div<Output = T>
+ Mul<Output = T>
+ num::FromPrimitive,
const N: usize,
>(
// Generics
ranges: Arc<[Range<T>; N]>,
f: fn(&[T; N], Option<Arc<A>>) -> f64,
evaluation_data: Option<Arc<A>>,
counter: Arc<AtomicU64>,
best: Arc<Mutex<f64>>,
thread_exit: Arc<AtomicBool>,
_thread_execution_position: Arc<AtomicU8>,
_thread_execution_times: Arc<[Mutex<(Duration, u64)>; 0]>,
// Specifics
point_ranges: [Range<u64>; N],
steps: [T; N],
) -> (f64, [T; N]) {
let (mut best_value, mut best_points) = (f64::MAX, [Default::default(); N]);
let mut start_point = [Default::default(); N];
for i in 0..N {
start_point[i] = ranges[i].start;
}
// println!("start_point: {:?}",start_point);
for cartesian_product in point_ranges
.iter()
.map(|r| r.clone())
.multi_cartesian_product()
{
// Gets new point
let mut point = start_point;
// print!("[");
for i in 0..N {
// print!("{:?}*{:?}=",steps[i],T::from_u64(cartesian_product[i]).unwrap());
point[i] += steps[i] * T::from_u64(cartesian_product[i]).unwrap();
// print!("{:?},",point[i]);
}
// println!("] = {:?}",point);
let new = f(&point, evaluation_data.clone());
// println!("{:?} -> {:?}",point,new);
if new < best_value {
best_value = new;
best_points = point;
*best.lock().unwrap() = best_value;
}
counter.fetch_add(1, Ordering::SeqCst);
// Checks early exit
if thread_exit.load(Ordering::SeqCst) {
break;
}
}
(best_value, best_points)
}
}