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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068
use crate::*;
use crate::rewl::*;
use crate::glue_helper::*;
use rand::{Rng, SeedableRng, prelude::SliceRandom};
use std::{num::NonZeroUsize, sync::*, cmp::*};
use rayon::{iter::ParallelIterator, prelude::*};
#[cfg(feature = "sweep_time_optimization")]
use std::cmp::Reverse;
#[cfg(feature = "serde_support")]
use serde::{Serialize, Deserialize};
/// # Efficient replica exchange Wang landau
/// * use this to quickly build your own parallel replica exchange wang landau simulation
/// ## Tipp
/// Use the short hand [`Rewl`](crate::Rewl)
/// ## Citations
/// * the following paper were used to progamm this - you should cite them, if you use
/// this library for a publication!
///
/// > Y. W. Li, T. Vogel, T. Wüst and D. P. Landau,
/// > “A new paradigm for petascale Monte Carlo simulation: Replica exchange Wang-Landau sampling,”
/// > J. Phys.: Conf. Ser. **510** 012012 (2014), DOI [10.1088/1742-6596/510/1/012012](https://doi.org/10.1088/1742-6596/510/1/012012)
///
/// > T. Vogel, Y. W. Li, T. Wüst and D. P. Landau,
/// > “Exploring new frontiers in statistical physics with a new, parallel Wang-Landau framework,”
/// > J. Phys.: Conf. Ser. **487** 012001 (2014), DOI [10.1088/1742-6596/487/1/012001](https://doi.org/10.1088/1742-6596/487/1/012001)
///
/// > T. Vogel, Y. W. Li, T. Wüst and D. P. Landau,
/// > “Scalable replica-exchange framework for Wang-Landau sampling,”
/// > Phys. Rev. E **90**: 023302 (2014), DOI [10.1103/PhysRevE.90.023302](https://doi.org/10.1103/PhysRevE.90.023302)
///
/// > R. E. Belardinelli and V. D. Pereyra,
/// > “Fast algorithm to calculate density of states,”
/// > Phys. Rev. E **75**: 046701 (2007), DOI [10.1103/PhysRevE.75.046701](https://doi.org/10.1103/PhysRevE.75.046701)
///
/// > F. Wang and D. P. Landau,
/// > “Efficient, multiple-range random walk algorithm to calculate the density of states,”
/// > Phys. Rev. Lett. **86**, 2050–2053 (2001), DOI [10.1103/PhysRevLett.86.2050](https://doi.org/10.1103/PhysRevLett.86.2050)
#[derive(Debug)]
#[cfg_attr(feature = "serde_support", derive(Serialize, Deserialize))]
pub struct ReplicaExchangeWangLandau<Ensemble, R, Hist, Energy, S, Res>{
pub(crate) chunk_size: NonZeroUsize,
pub(crate) ensembles: Vec<RwLock<Ensemble>>,
pub(crate) walker: Vec<RewlWalker<R, Hist, Energy, S, Res>>,
pub(crate) log_f_threshold: f64,
pub(crate) replica_exchange_mode: bool,
}
/// Short for [`ReplicaExchangeWangLandau`](crate::rewl::ReplicaExchangeWangLandau),
/// which you can look at for citations
pub type Rewl<Ensemble, R, Hist, Energy, S, Res> = ReplicaExchangeWangLandau<Ensemble, R, Hist, Energy, S, Res>;
impl<Ensemble, R, Hist, Energy, S, Res> Rewl<Ensemble, R, Hist, Energy, S, Res>
{
/// # Read access to internal rewl walkers
/// * each of these walkers independently samples an interval.
/// * see paper for more infos
pub fn walkers(&self) -> &Vec<RewlWalker<R, Hist, Energy, S, Res>>
{
&self.walker
}
/// # Iterator over ensembles
/// If you do not know what `RwLockReadGuard<'a, Ensemble>` is - do not worry.
/// you can just pretend it is `&Ensemble` and everything should work out fine,
/// since it implements [`Deref`](https://doc.rust-lang.org/std/ops/trait.Deref.html).
/// Of cause, you can also take a look at [`RwLockReadGuard`](https://doc.rust-lang.org/std/sync/struct.RwLockReadGuard.html)
pub fn ensemble_iter<'a>(&'a self) -> impl Iterator<Item=RwLockReadGuard<'a, Ensemble>>
{
self.ensembles
.iter()
.map(|e| e.read().unwrap())
}
/// # read access to your ensembles
/// * `None` if `index` out of range
/// * If you do not know what `RwLockReadGuard<Ensemble>` is - do not worry.
/// you can just pretend it is `&Ensemble` and everything will work out fine,
/// since it implements [`Deref`](https://doc.rust-lang.org/std/ops/trait.Deref.html).
/// Of cause, you can also take a look at [`RwLockReadGuard`](https://doc.rust-lang.org/std/sync/struct.RwLockReadGuard.html)
pub fn get_ensemble(&self, index: usize) -> Option<RwLockReadGuard<Ensemble>>
{
self.ensembles
.get(index)
.map(|e| e.read().unwrap())
}
/// # Mutable iterator over ensembles
/// * if possible, prefer [`ensemble_iter`](Self::ensemble_iter)
/// * **unsafe** only use this if you know what you are doing
/// * it is assumed, that whatever you change has no effect on the
/// Markov Chain, the result of the energy function etc.
/// * might **panic** if a thread is poisened
pub unsafe fn ensemble_iter_mut(&mut self) -> impl Iterator<Item=&mut Ensemble>
{
self.ensembles
.iter_mut()
.map(|item| item.get_mut().unwrap())
}
/// # mut access to your ensembles
/// * if possible, prefer [`get_ensemble`](Self::get_ensemble)
/// * **unsafe** only use this if you know what you are doing
/// * it is assumed, that whatever you change has no effect on the
/// Markov Chain, the result of the energy function etc.
/// * None if `index` out of range
/// * might **panic** if a thread is poisened
pub unsafe fn get_ensemble_mut(&mut self, index: usize) -> Option<&mut Ensemble>
{
self.ensembles
.get_mut(index)
.map(|e| e.get_mut().unwrap())
}
/// # Get the number of intervals present
pub fn num_intervals(&self) -> usize
{
self.walker.len() / self.chunk_size.get()
}
/// Returns number of walkers per interval
pub fn walkers_per_interval(&self) -> NonZeroUsize
{
self.chunk_size
}
/// # Change step size for markov chain of walkers
/// * changes the step size used in the sweep
/// * changes step size of all walkers in the nth interval
/// * returns Err if index out of bounds, i.e., the requested interval does not exist
/// * interval counting starts at 0, i.e., n=0 is the first interval
pub fn change_step_size_of_interval(&mut self, n: usize, step_size: usize) -> Result<(), ()>
{
let start = n * self.chunk_size.get();
let end = start + self.chunk_size.get();
if self.walker.len() < end {
Err(())
} else {
let slice = &mut self.walker[start..start+self.chunk_size.get()];
slice.iter_mut()
.for_each(|entry| entry.step_size_change(step_size));
Ok(())
}
}
/// # Get step size for markov chain of walkers
/// * returns `None` if index out of bounds, i.e., the requested interval does not exist
/// * interval counting starts at 0, i.e., n=0 is the first interval
pub fn get_step_size_of_interval(&self, n: usize) -> Option<usize>
{
let start = n * self.chunk_size.get();
let end = start + self.chunk_size.get();
if self.walker.len() < end {
None
} else {
let slice = &self.walker[start..start+self.chunk_size.get()];
let step_size = slice[0].step_size();
slice[1..]
.iter()
.for_each(|w|
assert_eq!(
step_size, w.step_size(),
"Fatal Error encountered; ERRORCODE 0x9 - \
Sweep sizes of intervals do not match! \
This should be impossible! if you are using the latest version of the \
'sampling' library, please contact the library author via github by opening an \
issue! https://github.com/Pardoxa/sampling/issues"
)
);
Some(step_size)
}
}
/// # Change sweep size for markov chain of walkers
/// * changes the sweep size used in the sweep
/// * changes sweep size of all walkers in the nth interval
/// * returns Err if index out of bounds, i.e., the requested interval does not exist
/// * interval counting starts at 0, i.e., n=0 is the first interval
pub fn change_sweep_size_of_interval(&mut self, n: usize, sweep_size: NonZeroUsize) -> Result<(), ()>
{
let start = n * self.chunk_size.get();
let end = start + self.chunk_size.get();
if self.walker.len() < end {
Err(())
} else {
let slice = &mut self.walker[start..start+self.chunk_size.get()];
slice.iter_mut()
.for_each(|entry| entry.sweep_size_change(sweep_size));
Ok(())
}
}
/// # Get sweep size for markov chain of walkers
/// * returns `None` if index out of bounds, i.e., the requested interval does not exist
/// * interval counting starts at 0, i.e., n=0 is the first interval
pub fn get_sweep_size_of_interval(&self, n: usize) -> Option<NonZeroUsize>
{
let start = n * self.chunk_size.get();
let end = start + self.chunk_size.get();
if self.walker.len() < end {
None
} else {
let slice = &self.walker[start..start+self.chunk_size.get()];
let sweep_size = slice[0].sweep_size();
slice[1..]
.iter()
.for_each(|w|
assert_eq!(
sweep_size, w.sweep_size(),
"Fatal Error encountered; ERRORCODE 0xA - \
Sweep sizes of intervals do not match! \
This should be impossible! if you are using the latest version of the \
'sampling' library, please contact the library author via github by opening an \
issue! https://github.com/Pardoxa/sampling/issues"
)
);
Some(sweep_size)
}
}
}
impl<Ensemble, R, Hist, Energy, S, Res> Rewl<Ensemble, R, Hist, Energy, S, Res>
where R: Send + Sync + Rng + SeedableRng,
Hist: Send + Sync + Histogram + HistogramVal<Energy>,
Energy: Send + Sync + Clone,
Ensemble: MarkovChain<S, Res>,
Res: Send + Sync,
S: Send + Sync
{
/// # Perform the Replica exchange wang landau simulation
/// * will simulate until **all** walkers have factors `log_f`
/// that are below the threshold you chose
pub fn simulate_until_convergence<F>(
&mut self,
energy_fn: F
)
where
Ensemble: Send + Sync,
R: Send + Sync,
F: Fn(&mut Ensemble) -> Option<Energy> + Copy + Send + Sync
{
while !self.is_finished()
{
self.sweep(energy_fn);
}
}
/// # Perform the Replica exchange wang landau simulation
/// * will simulate until **all** walkers have factors `log_f`
/// that are below the threshold you chose **or**
/// * until condition returns false
pub fn simulate_while<F, C>(
&mut self,
energy_fn: F,
mut condition: C
)
where
Ensemble: Send + Sync,
R: Send + Sync,
F: Fn(&mut Ensemble) -> Option<Energy> + Copy + Send + Sync,
C: FnMut(&Self) -> bool
{
while !self.is_finished() && condition(&self)
{
self.sweep(energy_fn);
}
}
/// # Sanity check
/// * checks if the stored (i.e., last) energy(s) of the system
/// match with the result of energy_fn
pub fn check_energy_fn<F>(
&mut self,
energy_fn: F
) -> bool
where Energy: PartialEq,
F: Fn(&mut Ensemble) -> Option<Energy> + Copy + Send + Sync,
Ensemble: Sync + Send
{
let ensembles = self.ensembles.as_slice();
self.walker
.par_iter()
.all(|w| w.check_energy_fn(ensembles, energy_fn))
}
/// # Sweep
/// * Performs one sweep of the Replica exchange wang landau simulation
/// * You can make a complete simulation, by repeatatly calling this method
/// until `self.is_finished()` returns true
pub fn sweep<F>(&mut self, energy_fn: F)
where Ensemble: Send + Sync,
R: Send + Sync,
F: Fn(&mut Ensemble) -> Option<Energy> + Copy + Send + Sync
{
let slice = self.ensembles.as_slice();
#[cfg(not(feature = "sweep_time_optimization"))]
let walker = &mut self.walker;
#[cfg(feature = "sweep_time_optimization")]
let mut walker =
{
let mut walker = Vec::with_capacity(self.walker.len());
walker.extend(
self.walker.iter_mut()
);
walker.par_sort_unstable_by_key(|w| Reverse(w.duration()));
walker
};
walker
.par_iter_mut()
.for_each(|w| w.wang_landau_sweep(slice, energy_fn));
self.walker
.par_chunks_mut(self.chunk_size.get())
.filter(|chunk|
{
chunk.iter()
.all(RewlWalker::all_bins_reached)
}
)
.for_each(
|chunk|
{
chunk.iter_mut()
.for_each(RewlWalker::refine_f_reset_hist);
merge_walker_prob(chunk);
}
);
// replica exchange
if self.walkers_per_interval().get() > 1 {
let exchange_m = self.replica_exchange_mode;
self.walker
.par_chunks_mut(self.chunk_size.get())
.for_each(
|chunk|
{
let mut shuf = Vec::with_capacity(chunk.len());
if let Some((first, rest)) = chunk.split_first_mut(){
shuf.extend(
rest.iter_mut()
);
shuf.shuffle(&mut first.rng);
shuf.push(first);
let s = if exchange_m {
&mut shuf
} else {
&mut shuf[1..]
};
s.chunks_exact_mut(2)
.for_each(
|c|
{
let ptr = c.as_mut_ptr();
unsafe{
let a = &mut *ptr;
let b = &mut *ptr.offset(1);
replica_exchange(a, b);
}
}
);
}
}
);
}
let walker_slice = if self.replica_exchange_mode
{
&mut self.walker
} else {
&mut self.walker[self.chunk_size.get()..]
};
self.replica_exchange_mode = !self.replica_exchange_mode;
let chunk_size = self.chunk_size;
walker_slice
.par_chunks_exact_mut(2 * self.chunk_size.get())
.for_each(
|walker_chunk|
{
let (slice_a, slice_b) = walker_chunk.split_at_mut(chunk_size.get());
let mut slice_b_shuffled: Vec<_> = slice_b.iter_mut().collect();
slice_b_shuffled.shuffle(&mut slice_a[0].rng);
for (walker_a, walker_b) in slice_a.iter_mut()
.zip(slice_b_shuffled.into_iter())
{
replica_exchange(walker_a, walker_b);
}
}
)
}
/// returns largest value of factor log_f present in the walkers
pub fn largest_log_f(&self) -> f64
{
self.walker
.iter()
.map(|w| w.log_f())
.fold(std::f64::NEG_INFINITY, |acc, x| x.max(acc))
}
/// # Log_f factors of the walkers
/// * the log_f's will be reduced towards 0 during the simulation
pub fn log_f_vec(&self) -> Vec<f64>
{
self.walker
.iter()
.map(|w| w.log_f())
.collect()
}
/// # Is the simulation finished?
/// checks if **all** walkers have factors `log_f`
/// that are below the threshold you chose
pub fn is_finished(&self) -> bool
{
self.walker
.iter()
.all(|w| w.log_f() < self.log_f_threshold)
}
/// # Result of the simulations!
/// This is what we do the simulation for!
///
/// It returns the log10 of the normalized (i.e. sum=1 within numerical precision) probability density and the
/// histogram, which contains the corresponding bins.
///
/// Failes if the internal histograms (invervals) do not align. Might fail if
/// there is no overlap between neighboring intervals
pub fn merged_log10_prob(&self) -> Result<(Hist, Vec<f64>), HistErrors>
where Hist: HistogramCombine
{
let (e_hist, mut log_prob) = self.merged_log_prob()?;
// switch base of log
ln_to_log10(&mut log_prob);
Ok((e_hist, log_prob))
}
/// # Results of the simulation
///
/// This is what we do the simulation for!
///
/// It returns histogram, which contains the corresponding bins and
/// the logarithm with base 10 of the normalized (i.e. sum=1 within numerical precision)
/// probability density. Lastly it returns the vector of the aligned probability estimates (also log10) of the
/// different intervals. This can be used to see, how good the simulation worked,
/// e.g., by plotting them to see, if they match
///
/// ## Notes
/// Failes if the internal histograms (invervals) do not align. Might fail if
/// there is no overlap between neighboring intervals
pub fn merged_log10_prob_and_aligned(&self) -> Result<(Hist, Vec<f64>, Vec<Vec<f64>>), HistErrors>
where Hist: HistogramCombine
{
let (e_hist, mut log_prob, mut aligned) = self.merged_log_prob_and_aligned()?;
ln_to_log10(&mut log_prob);
aligned.par_iter_mut()
.for_each(
|slice|
{
ln_to_log10(slice);
}
);
Ok(
(e_hist, log_prob, aligned)
)
}
/// # Result of the simulations!
/// This is what we do the simulation for!
///
/// It returns the natural logarithm of the normalized (i.e. sum=1 within numerical precision) probability density and the
/// histogram, which contains the corresponding bins.
///
/// Failes if the internal histograms (invervals) do not align. Might fail if
/// there is no overlap between neighboring intervals
pub fn merged_log_prob(&self) -> Result<(Hist, Vec<f64>), HistErrors>
where Hist: HistogramCombine
{
let (mut log_prob, e_hist) = self.merged_log_probability()?;
norm_ln_prob(&mut log_prob);
Ok((e_hist, log_prob))
}
/// # Results of the simulation
///
/// This is what we do the simulation for!
///
/// It returns histogram, which contains the corresponding bins and
/// the natural logarithm of the normalized (i.e. sum=1 within numerical precision)
/// probability density. Lastly it returns the vector of the aligned probability estimates (also ln) of the
/// different intervals. This can be used to see, how good the simulation worked,
/// e.g., by plotting them to see, if they match
///
/// ## Notes
/// Failes if the internal histograms (invervals) do not align. Might fail if
/// there is no overlap between neighboring intervals
pub fn merged_log_prob_and_aligned(&self) -> Result<(Hist, Vec<f64>, Vec<Vec<f64>>), HistErrors>
where Hist: HistogramCombine
{
let (e_hist, mut log_prob, mut aligned) = self.merged_log_probability_and_align()?;
let shift = norm_ln_prob(&mut log_prob);
aligned.par_iter_mut()
.for_each(
|aligned|
{
aligned.iter_mut()
.for_each(|val| *val -= shift)
}
);
Ok(
(
e_hist,
log_prob,
aligned
)
)
}
fn merged_log_probability(&self) -> Result<(Vec<f64>, Hist), HistErrors>
where Hist: HistogramCombine
{
let (hists, log_probs) = self.get_log_prob_and_hists();
let (merge_points, alignment, log_prob, e_hist) =
self.merged_log_probability_helper2(log_probs, hists)?;
Ok(
only_merged(
merge_points,
alignment,
log_prob,
e_hist
)
)
}
fn merged_log_probability_and_align(&self) -> Result<(Hist, Vec<f64>, Vec<Vec<f64>>), HistErrors>
where Hist: HistogramCombine
{
let (hists, log_probs) = self.get_log_prob_and_hists();
let (merge_points, alignment, log_prob, e_hist) =
self.merged_log_probability_helper2(log_probs, hists)?;
merged_and_aligned(
self.walker.iter()
.step_by(self.walkers_per_interval().get())
.map(|v| v.hist()),
merge_points,
alignment,
log_prob,
e_hist
)
}
fn get_log_prob_and_hists(&self) -> (Vec<&Hist>, Vec<Vec<f64>>)
{
// get the log_probabilities - the walkers over the same intervals are merged
let log_prob: Vec<_> = self.walker
.par_chunks(self.chunk_size.get())
.map(get_merged_walker_prob)
.collect();
let hists: Vec<_> = self.walker.iter()
.step_by(self.chunk_size.get())
.map(|w| w.hist())
.collect();
(hists, log_prob)
}
fn merged_log_probability_helper2(
&self,
mut log_prob: Vec<Vec<f64>>,
hists: Vec<&Hist>
) -> Result<(Vec<usize>, Vec<usize>, Vec<Vec<f64>>, Hist), HistErrors>
where Hist: HistogramCombine
{
// get the log_probabilities - the walkers over the same intervals are merged
log_prob
.par_iter_mut()
.for_each(
|v|
{
subtract_max(v);
}
);
// get the derivative, for merging later
let derivatives: Vec<_> = log_prob.par_iter()
.map(|v| derivative_merged(v))
.collect();
let e_hist = Hist::encapsulating_hist(&hists)?;
let alignment = hists.iter()
.zip(hists.iter().skip(1))
.map(|(&left, &right)| left.align(right))
.collect::<Result<Vec<_>, _>>()?;
let merge_points = calc_merge_points(&alignment, &derivatives);
Ok(
(
merge_points,
alignment,
log_prob,
e_hist
)
)
}
/// # Get Ids
/// This is an indicator that the replica exchange works.
/// In the beginning, this will be a sorted vector, e.g. \[0,1,2,3,4\].
/// Then it will show, where the ensemble, which the corresponding walkers currently work with,
/// originated from. E.g. If the vector is \[3,1,0,2,4\], Then walker 0 has a
/// ensemble originating from walker 3, the walker 1 is back to its original
/// ensemble, walker 2 has an ensemble originating form walker 0 and so on.
pub fn get_id_vec(&self) -> Vec<usize>
{
self.walker
.iter()
.map(|w| w.id())
.collect()
}
/// # read access to the internal histograms used by the walkers
pub fn hists(&self) -> Vec<&Hist>
{
self.walker.iter()
.map(|w| w.hist())
.collect()
}
/// # read access to internal histogram
/// * None if index out of range
pub fn get_hist(&self, index: usize) -> Option<&Hist>
{
self.walker
.get(index)
.map(|w| w.hist())
}
/// # Convert into Rees
/// This creates a Replica exchange entropic sampling simulation
/// from this Replica exchange wang landau simulation
pub fn into_rees(self) -> Rees<(), Ensemble, R, Hist, Energy, S, Res>
{
self.into()
}
/// # Convert into Rees
/// * similar to [into_rees](`crate::rewl::Rewl::into_rees`), though now we can store extra information.
/// The extra information can be anything, e.g., files in which
/// each walker should later write information every nth step or something
/// else entirely.
///
/// # important
/// * The vector `extra` must be exactly as long as the walker slice and
/// each walker is assigned the corresponding entry from the vector `extra`
/// * You can look at the walker slice with the [walkers](`crate::rewl::Rewl::walkers`) method
pub fn into_rees_with_extra<Extra>(self, extra: Vec<Extra>) -> Result<Rees<Extra, Ensemble, R, Hist, Energy, S, Res>, (Self, Vec<Extra>)>
{
if extra.len() != self.walker.len()
{
Err((self, extra))
} else {
let mut walker = Vec::with_capacity(self.walker.len());
walker.extend(
self.walker
.into_iter()
.map(|w| w.into())
);
let rees =
Rees{
walker,
ensembles: self.ensembles,
replica_exchange_mode: self.replica_exchange_mode,
extra,
chunk_size: self.chunk_size
};
Ok(
rees
)
}
}
}
/// # Merge probability density of multiple rewl simulations
/// * Will calculate the merged log (base 10) probability density. Also returns the corresponding histogram.
/// * If an interval has multiple walkers, their probability will be merged before all probabilities are aligned
/// * `rewls` does not need to be sorted in any way
/// ## Errors
/// * will return `HistErrors::EmptySlice` if the `rees` slice is empty
/// * will return other HistErrors if the intervals have no overlap
pub fn merged_log10_prob<Ensemble, R, Hist, Energy, S, Res>(rewls: &[Rewl<Ensemble, R, Hist, Energy, S, Res>]) -> Result<(Vec<f64>, Hist), HistErrors>
where Hist: HistogramVal<Energy> + HistogramCombine + Send + Sync,
Energy: PartialOrd
{
let mut res = merged_log_prob(rewls)?;
ln_to_log10(&mut res.0);
Ok(res)
}
/// # Merge probability density of multiple rewl simulations
/// * Will calculate the merged log (base e) probability density. Also returns the corresponding histogram.
/// * If an interval has multiple walkers, their probability will be merged before all probabilities are aligned
/// * `rewls` does not need to be sorted in any way
/// ## Errors
/// * will return `HistErrors::EmptySlice` if the `rees` slice is empty
/// * will return other HistErrors if the intervals have no overlap
pub fn merged_log_prob<Ensemble, R, Hist, Energy, S, Res>(rewls: &[Rewl<Ensemble, R, Hist, Energy, S, Res>]) -> Result<(Vec<f64>, Hist), HistErrors>
where Hist: HistogramVal<Energy> + HistogramCombine + Send + Sync,
Energy: PartialOrd
{
if rewls.is_empty() {
return Err(HistErrors::EmptySlice);
}
let merged_prob = merged_probs(rewls);
let container = combine_container(rewls, &merged_prob, true);
let (merge_points, alignment, log_prob, e_hist) = align(&container)?;
Ok(
only_merged(
merge_points,
alignment,
log_prob,
e_hist
)
)
}
/// # Results of the simulation
/// This is what we do the simulation for!
///
/// * similar to [merged_log10_prob_and_aligned](`crate::rewl::Rewl::merged_log10_prob_and_aligned`)
/// * the difference is, that the logarithms are now calculated to base 10
pub fn merged_log10_probability_and_align<Ensemble, R, Hist, Energy, S, Res>(
rewls: &[Rewl<Ensemble, R, Hist, Energy, S, Res>]
) -> Result<(Hist, Vec<f64>, Vec<Vec<f64>>), HistErrors>
where Hist: HistogramCombine + HistogramVal<Energy> + Send + Sync,
Energy: PartialOrd
{
merged_log10_probability_and_align_ignore(rewls, &[])
}
/// # Results of the simulation
/// This is what we do the simulations for!
///
/// * similar to [merged_log10_probability_and_align](`crate::rewl::merged_log10_probability_and_align`)
/// * Now, however, we have a slice called `ignore`. It should contain the indices
/// of all walkers, that should be ignored for the alignment and merging into the
/// final probability density function. The indices do not need to be sorted, though
/// duplicates will be ignored and indices, which are out of bounds will also be ignored
pub fn merged_log10_probability_and_align_ignore<Ensemble, R, Hist, Energy, S, Res>(
rewls: &[Rewl<Ensemble, R, Hist, Energy, S, Res>],
ignore: &[usize]
) -> Result<(Hist, Vec<f64>, Vec<Vec<f64>>), HistErrors>
where Hist: HistogramCombine + HistogramVal<Energy> + Send + Sync,
Energy: PartialOrd
{
let mut res = merged_log_probability_and_align_ignore(rewls, ignore)?;
ln_to_log10(&mut res.1);
res.2.par_iter_mut()
.for_each(|slice| ln_to_log10(slice));
Ok(res)
}
/// # Results of the simulation
/// This is what we do the simulations for!
///
/// * similar to [log_probability_and_align](`crate::rewl::log_probability_and_align`)
/// * Here, however, all logarithms are base 10
pub fn log10_probability_and_align<Ensemble, R, Hist, Energy, S, Res>(rewls: &[Rewl<Ensemble, R, Hist, Energy, S, Res>]) -> Result<(Hist, Vec<f64>, Vec<Vec<f64>>), HistErrors>
where Hist: HistogramCombine + HistogramVal<Energy> + Send + Sync,
Energy: PartialOrd
{
log10_probability_and_align_ignore(rewls, &[])
}
/// # Results of the simulation
/// This is what we do the simulations for!
///
/// * similar to [log10_probability_and_align](`crate::rewl::log10_probability_and_align`)
/// * Now, however, we have a slice called `ignore`. It should contain the indices
/// of all walkers, that should be ignored for the alignment and merging into the
/// final probability density function. The indices do not need to be sorted, though
/// duplicates will be ignored and indices, which are out of bounds will also be ignored
pub fn log10_probability_and_align_ignore<Ensemble, R, Hist, Energy, S, Res>(rewls: &[Rewl<Ensemble, R, Hist, Energy, S, Res>], ignore: &[usize]) -> Result<(Hist, Vec<f64>, Vec<Vec<f64>>), HistErrors>
where Hist: HistogramCombine + HistogramVal<Energy> + Send + Sync,
Energy: PartialOrd
{
let mut res = log_probability_and_align_ignore(rewls, ignore)?;
ln_to_log10(&mut res.1);
res.2.par_iter_mut()
.for_each(|slice| ln_to_log10(slice));
Ok(res)
}
/// # Results of the simulation
/// This is what we do the simulations for!
///
/// * `rewls` a slice of all replica exchange simulations you which to merge
/// to create a final probability density estimate for whatever you sampled.
/// Note, that while the slice `rewls` does not need to be ordered,
/// there should not be no gaps between the intervals that were sampled.
/// Also, the overlap of adjacent intervals should be large enough.
///
/// # Result::Ok
/// * The Hist is only useful for the interval, i.e., it tells you which bins
/// correspond to the entries of the probability density function - it does not count how often the bins were hit.
/// It is still the encapsulating interval, for which the probability density function was calculated
/// * The `Vec<f64>` is the logarithm (base e) of the probability density function,
/// which you wanted to get!
/// * `Vec<Vec<f64>> these are the aligned probability estimates (also logarithm base e)
/// of the different intervals.
/// This can be used to see, how good the simulation worked, e.g., by plotting them to see, if they match
///
/// # Failures
/// Failes if the internal histograms (intervals) do not align.
/// Might fail if there is no overlap between neighboring intervals
///
/// # Notes
/// The difference between this function and
/// [log_probability_and_align](`crate::rewl::log_probability_and_align`) is,
/// that, if there are multiple walkers in the same interval, they **will** be merged by
/// averaging their probability estimates in this function, while they are **not** averaged in
/// [log_probability_and_align](`crate::rewl::log_probability_and_align`)
pub fn merged_log_probability_and_align<Ensemble, R, Hist, Energy, S, Res>
(
rewls: &[Rewl<Ensemble, R, Hist, Energy, S, Res>]
) -> Result<(Hist, Vec<f64>, Vec<Vec<f64>>), HistErrors>
where Hist: HistogramCombine + HistogramVal<Energy> + Send + Sync,
Energy: PartialOrd
{
merged_log_probability_and_align_ignore(rewls, &[])
}
/// # Result of the simulation
/// This is what you were looking for!
///
/// * similar to [merged_log_probability_and_align](`crate::rewl::merged_log_probability_and_align`)
/// * Now, however, we have a slice called `ignore`. It should contain the indices
/// of all walkers, that should be ignored for the alignment and merging into the
/// final probability density function. The indices do not need to be sorted, though
/// duplicates will be ignored and indices, which are out of bounds will also be ignored
pub fn merged_log_probability_and_align_ignore<Ensemble, R, Hist, Energy, S, Res>(
rewls: &[Rewl<Ensemble, R, Hist, Energy, S, Res>],
ignore: &[usize]
) -> Result<(Hist, Vec<f64>, Vec<Vec<f64>>), HistErrors>
where Hist: HistogramCombine + HistogramVal<Energy> + Send + Sync,
Energy: PartialOrd
{
if rewls.is_empty(){
return Err(HistErrors::EmptySlice);
}
let merged_prob = merged_probs(rewls);
let mut container = combine_container(rewls, &merged_prob, true);
ignore_fn(&mut container, ignore);
let (merge_points, alignment, log_prob, e_hist) = align(&container)?;
merged_and_aligned(
container.iter()
.map(|c| c.1),
merge_points,
alignment,
log_prob,
e_hist
)
}
/// # Results of the simulation
/// This is what we do the simulations for!
///
/// * `rewls` a slice of all replica exchange simulations you which to merge
/// to create a final probability density estimate for whatever you sampled.
/// Note, that while the slice `rewls` does not need to be ordered,
/// there should not be no gaps between the intervals that were sampled.
/// Also, the overlap of adjacent intervals should be large enough.
///
/// # Result::Ok
/// * The Hist is only useful for the interval, i.e., it tells you which bins
/// correspond to the entries of the probability density function - it does not count how often the bins were hit.
/// It is still the encapsulating interval, for which the probability density function was calculated
/// * The `Vec<f64>` is the logarithm (base e) of the probability density function,
/// which you wanted to get!
/// * `Vec<Vec<f64>> these are the aligned probability estimates (also logarithm base e)
/// of the different intervals.
/// This can be used to see, how good the simulation worked, e.g., by plotting them to see, if they match
///
/// # Failures
/// Failes if the internal histograms (intervals) do not align.
/// Might fail if there is no overlap between neighboring intervals
///
/// # Notes
/// The difference between this function and
/// [merged_log_probability_and_align](`crate::rewl::merged_log_probability_and_align`) is,
/// that, if there are multiple walkers in the same interval, they will **not** be merged by
/// averaging their probability estimates in this function, while they **are averaged** in
/// [merged_log_probability_and_align](`crate::rewl::merged_log_probability_and_align`)
pub fn log_probability_and_align<Ensemble, R, Hist, Energy, S, Res>(rewls: &[Rewl<Ensemble, R, Hist, Energy, S, Res>]) -> Result<(Hist, Vec<f64>, Vec<Vec<f64>>), HistErrors>
where Hist: HistogramCombine + HistogramVal<Energy> + Send + Sync,
Energy: PartialOrd
{
log_probability_and_align_ignore(rewls, &[])
}
/// # Results of the simulation
/// This is what we do the simulations for!
///
/// * similar to [log_probability_and_align](`crate::rewl::log_probability_and_align`)
/// * Now, however, we have a slice called `ignore`. It should contain the indices
/// of all walkers, that should be ignored for the alignment and merging into the
/// final probability density function. The indices do not need to be sorted, though
/// duplicates will be ignored and indices, which are out of bounds will also be ignored
pub fn log_probability_and_align_ignore<Ensemble, R, Hist, Energy, S, Res>(rewls: &[Rewl<Ensemble, R, Hist, Energy, S, Res>], ignore: &[usize]) -> Result<(Hist, Vec<f64>, Vec<Vec<f64>>), HistErrors>
where Hist: HistogramCombine + HistogramVal<Energy> + Send + Sync,
Energy: PartialOrd
{
if rewls.is_empty(){
return Err(HistErrors::EmptySlice);
}
let probs = probs(rewls);
let mut container = combine_container(rewls, &probs, false);
ignore_fn(&mut container, ignore);
let (merge_points, alignment, log_prob, e_hist) = align(&container)?;
merged_and_aligned(
container.iter()
.map(|c| c.1),
merge_points,
alignment,
log_prob,
e_hist
)
}
/// Helper to ignore specific intervals/walkers
pub(crate) fn ignore_fn<T>(container: &mut Vec<T>, ignore: &[usize])
{
let mut ignore = ignore.to_vec();
// sorting in reverse, to remove correct indices later on
ignore.sort_unstable_by_key(|&e| Reverse(e));
// remove duplicates
ignore.dedup();
// remove indices
ignore.into_iter()
.for_each(
|i|
{
if i < container.len(){
let _ = container.remove(i);
}
}
);
}
fn merged_probs<Ensemble, R, Hist, Energy, S, Res>
(
rewls: &[Rewl<Ensemble, R, Hist, Energy, S, Res>]
) -> Vec<Vec<f64>>
{
let merged_probs: Vec<_> = rewls.iter()
.flat_map(
|rewl|
{
rewl.walkers()
.chunks(rewl.walkers_per_interval().get())
.map(get_merged_walker_prob)
}
).collect();
merged_probs
}
fn probs<Ensemble, R, Hist, Energy, S, Res>
(
rewls: &[Rewl<Ensemble, R, Hist, Energy, S, Res>]
) -> Vec<Vec<f64>>
{
rewls.iter()
.flat_map(
|rewl|
{
rewl.walkers()
.iter()
.map(
|w|
w.log_density().into()
)
}
).collect()
}
fn combine_container<'a, Ensemble, R, Hist, Energy, S, Res>
(
rewls: &'a [Rewl<Ensemble, R, Hist, Energy, S, Res>],
log_probabilities: &'a [Vec<f64>],
merged: bool
) -> Vec<(&'a [f64], &'a Hist)>
where Hist: HistogramVal<Energy> + HistogramCombine,
Energy: PartialOrd
{
let mut step_by = NonZeroUsize::new(1).unwrap();
let hists: Vec<_> = rewls.iter()
.flat_map(
|rewl|
{
if merged {
step_by = rewl.walkers_per_interval();
}
rewl.walkers()
.iter()
.step_by(step_by.get())
.map(|w| w.hist())
}
).collect();
assert_eq!(hists.len(), log_probabilities.len());
let mut container: Vec<_> = log_probabilities
.iter()
.zip(hists.into_iter())
.map(|(prob, hist)| (prob.as_slice(), hist))
.collect();
container
.sort_unstable_by(
|a, b|
{
a.1.first_border()
.partial_cmp(&b.1.first_border())
.unwrap_or(Ordering::Equal)
}
);
container
}