-
Notifications
You must be signed in to change notification settings - Fork 3
/
rockstar.pl
884 lines (722 loc) · 35.1 KB
/
rockstar.pl
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
#!C:\perl\bin\perl
# ROCK* Tuner
# Copyright (C) 2009-2015 Lyudmil Antonov and Joona Kiiski
#
# ROCK* Tuner is free software: you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation, either version 3 of the License, or
# (at your option) any later version.
#
# ROCK* Tuner is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this program. If not, see <http://www.gnu.org/licenses/>.
#use diagnostics;
use strict;
use warnings;
use threads;
use threads::shared;
use Time::HiRes qw(time);
use IPC::Open2;
use IO::Select;
use Config::Tiny;
use Text::CSV;
#use Math::Complex;
use Math::CDF qw(qbeta);
#use Math::Random qw(random_multivariate_normal);
use Math::Round qw(nearest nearest_floor);
use Math::MatrixReal;
use Statistics::Descriptive;
use Statistics::Distributions qw(chisqrdistr);
use List::Util qw(min max sum);
use IO::Handle;
use AutoLoader qw(AUTOLOAD);
STDOUT->autoflush(1);
STDERR->autoflush(1);
my $IS_WINDOWS = ($^O eq 'MSWin32');
### SECTION. Settings (Static data during execution)
my $ConfigFile = $ARGV[0] || die "You must pass the name of config file as parameter!";
my $Config = Config::Tiny->new;
$Config = Config::Tiny->read($ConfigFile) || die "Unable to read configuration file '" . $ConfigFile . "'";
my $simulate = $Config->{Main}->{Simulate} ; defined($simulate) || die "Simulate not defined!";
my $variables_path = $Config->{Main}->{Variables} ; defined($variables_path) || die "Variables not defined!";
my $log_path = $Config->{Main}->{Log} ; defined($log_path) || die "Log not defined!";
my $gamelog_path = $Config->{Main}->{GameLog} ; defined($gamelog_path) || die "GameLog not defined!";
my $iterations = $Config->{Main}->{Iterations} ; defined($iterations) || die "Iterations not defined!";
### SECTION. Strategic parameters for ROCK*
my $sigma = $Config->{Main}->{Sigma} ; defined($sigma) || die "Sigma not defined!";
my $n_parameters = $Config->{Main}->{n_parameters} ; defined($n_parameters) || die "n_parameters not defined!";
my $lambda = $Config->{Main}->{lambda} ; defined($lambda) || die "lambda not defined!";
my $lambdaMD = $Config->{Main}->{lambdaMD} ; defined($lambdaMD) || die "lambdaMD not defined!";
my $games_in_match = $Config->{Main}->{games_in_match} ; defined($games_in_match) || die "games_in_match not defined!";
my $sigma_cur = $Config->{Main}->{sigma_cur} ; defined($sigma_cur) || die "sigma_cur not defined!";
my $near_bins_size = $Config->{Main}->{near_bins_size} ; defined($near_bins_size) || die "near_bins_size not defined!";
my $determinant = $Config->{Main}->{determinant} ; defined($determinant) || die "determinant not defined!";
my $best_cost = $Config->{Main}->{best_cost} ; defined($best_cost) || die "best_cost not defined!";
my $matches_evaluated = $Config->{Main}->{matches_evaluated} ; defined($matches_evaluated) || die "matches_evaluated not defined!";
my $initial_exp = $Config->{Main}->{initial_exp} ; defined($initial_exp) || die "initial_exp not defined!";
my $initial_sd = $Config->{Main}->{initial_sd} ; defined($initial_sd) || die "initial_sd not defined!";
my $constraint_penalty = $Config->{Main}->{constraint_penalty} ; defined($constraint_penalty)|| die "constraint_penalty not defined!";
### SECTION. Game parameters for ROCK* (same as in SPSA)
my $eng1_path = $Config->{Engine}->{Engine1} ; defined($eng1_path) || $simulate || die "Engine1 not defined!";
my $eng2_path = $Config->{Engine}->{Engine2} ; defined($eng2_path) || $simulate || die "Engine2 not defined!";
my $epd_path = $Config->{Engine}->{EPDBook} ; defined($epd_path) || $simulate || die "EPDBook not defined!";
my $base_time = $Config->{Engine}->{BaseTime} ; defined($base_time) || $simulate || die "BaseTime not defined!";
my $inc_time = $Config->{Engine}->{IncTime} ; defined($inc_time) || $simulate || die "IncTime not defined!";
my $threads = $Config->{Engine}->{Concurrency} ; defined($threads) || $simulate || die "Concurrency not defined!";
my $draw_score_limit = $Config->{Engine}->{DrawScoreLimit} ; defined($draw_score_limit) || $simulate || die "DrawScoreLimit not defined!";
my $draw_move_limit = $Config->{Engine}->{DrawMoveLimit} ; defined($draw_move_limit) || $simulate || die "DrawMoveLimit not defined!";
my $win_score_limit = $Config->{Engine}->{WinScoreLimit} ; defined($win_score_limit) || $simulate || die "WinScoreLimit not defined!";
my $win_move_limit = $Config->{Engine}->{WinMoveLimit} ; defined($win_move_limit) || $simulate || die "WinMoveLimit not defined!";
$threads = 1 if ($simulate);
### SECTION. Variable CSV-file columns. (Static data during execution)
my $VAR_NAME = 0; # Name
my $THETA = 1; # Start Value (theta_0)
my $POLICY = 2; # Policy
my $COST = 3; # Cost
my $VAR_SIMUL_ELO = 4; # Simulation: Elo loss from 0 (optimum) to +-100)
my $VAR_END = 5; # Nothing
# Extra calculated COLUMNS (SPSA parameters)
my $VAR_C = 6; # c
my $VAR_A_END = 7; # a in the last iteration
my $VAR_A = 8; # a
### SECTION. Variable definitions. (Static data during execution)
my @variables;
my %variableIdx;
### SECTION. Log file handle (Static data during execution)
local (*LOG);
### SECTION. Shared data (volatile data during the execution)
my $shared_lock :shared;
my $shared_iter :shared; # Iteration counter
my $shared_policy_old :shared; # Current policy by variable name
my $shared_mean_minus_min :shared; # Current policy by variable name
my %shared_theta :shared; # Current policy by variable name
my $var_eng1 :shared;
### SECTION. Helper functions
# Function to safely read in a standard CSV-file.
sub read_csv
{
my ($csvfile, $rows) = @_;
my ($CSV, $row);
open(INFILE, '<', $csvfile) || die "Could not open file '$csvfile' for reading!";
binmode(INFILE);
$CSV = Text::CSV->new();
while($row = $CSV->getline(\*INFILE))
{
push(@$rows, $row);
}
$CSV->eof || die "CSV-file parsing error: " . $CSV->error_diag();
close(INFILE);
}
# STEP. Calculated strategic ROCK* parameters.
my $cost2policy_cov_factor = chisqrdistr($n_parameters,0.05) * -0.5 / log($lambda);
my $range = chisqrdistr($n_parameters,0.05) * $lambdaMD;
my $expansion_factor_sigma = 1.3 ** (1 / log($n_parameters + 2.5)) - 1;
my $cc = 3 / ($n_parameters + 6) / log($n_parameters + 6);
my $ccov = 6 / ($n_parameters + 7) / log($n_parameters + 7);
my $chiN = $n_parameters ** 0.5 * (1 - 1 / (4 * $n_parameters) + 1 / (21 * $n_parameters ** 2));
my $imp_factor = 1.3;
my $best_policy;
# STEP. Initial ROCK* zero vectors and matrices.
my $initial_theta = new Math::MatrixReal(1,$n_parameters);
my $initial_policy = new Math::MatrixReal(1,$n_parameters);
my $cost_history = new Math::MatrixReal(1,$iterations);
my $theta_history = new Math::MatrixReal($iterations,$n_parameters);
my $policy_history = new Math::MatrixReal($iterations,$n_parameters);
# STEP. Init random generator
srand((time() ^ (time() % $])) ^ exp(length($0))**$$);
### SECTION. Log file handle (Static data during execution)
local (*LOG);
### SECTION. Execution preparation code ("main" function)
{
my $row;
# STEP. Open shared log file.
open(LOG, '>', $log_path) || die "Could not open file '$log_path' for writing!";
LOG->autoflush(1);
# STEP. Read in variable data.
read_csv($variables_path, \@variables);
# STEP. Validate variable data.
foreach $row (@variables)
{
die "Wrong number of columns!" if (scalar(@$row) != $VAR_END);
die "Invalid name: '$row->[$VAR_NAME]'" if ($row->[$VAR_NAME] !~ /^\w+$/);
die "Invalid current: '$row->[$THETA]'" if ($row->[$THETA] !~ /^[-+]?[0-9]*\.?[0-9]+$/);
die "Invalid max: '$row->[$COST]'" if ($row->[$COST] !~ /^[-+]?[0-9]*\.?[0-9]+$/);
die "Invalid min: '$row->[$POLICY]'" if ($row->[$POLICY] !~ /^[-+]?[0-9]*\.?[0-9]+$/);
die "Invalid simul ELO: '$row->[$VAR_SIMUL_ELO]'" if ($row->[$VAR_SIMUL_ELO] !~ /^[-+]?[0-9]*\.?[0-9]+$/);
}
# STEP. Prepare shared data
$shared_iter = 0;
# STEP. Launch ROCK* threads
my @thr;
for (my $i = 1; $i <= $threads; $i++)
{
$thr[$i] = threads->create(\&run_rockstar, $i);
# HACK: Under Windows the combination of starting new threads and
# calling open2() at the same time seems to be problematic.
# So wait for 3 seconds to make sure each new thread has cleanly
# started the engine process before starting a new thread.
sleep(3) if $IS_WINDOWS;
}
# STEP. Join threads
for (my $i = 1; $i <= $threads; $i++)
{
$thr[$i]->join();
}
# STEP. Close Log file
close(LOG);
# STEP. Quit
exit 0;
}
### SECTION. ROCK*
local (*GAMELOG);
sub run_rockstar
{
my ($threadId) = @_;
my $row;
my $covar = new Math::MatrixReal($n_parameters,$n_parameters);
$covar = $covar->one();
my $covar_inv = $covar->inverse();
my $pc = new Math::MatrixReal($n_parameters,1);
my $c_normalized = $covar;
my $result_inc = 0;
# STEP. Open thread specific log file
my $path = $gamelog_path;
my $from = quotemeta('$THREAD');
my $to = $threadId;
$path =~ s/$from/$to/g;
open(GAMELOG, '>', $path) || die "Could not open file '$path' for writing!";
# STEP. Init engines
engine_init() if (!$simulate);
# Generate random vector and initialise engine variables
#$my $random = Math::MatrixReal->new_random(1,$n_parameters,{bounded_by=>[-1,1]});
my $policy = $initial_policy;# + $random; # print "$policy\n";
#$theta = $iter > 1 ? $theta : $initial_theta;
#my $var_eng1 = $initial_theta + $policy * $sigma;
my ($cost, %var_eng2, %var_value, $shared_theta2, $var_eng1);
my $theta = $initial_theta;
# STEP. Create variable index for easy access.
my $count = 1;
foreach $row (@variables)
{
$variableIdx{$row->[$VAR_NAME]} = $row;
$shared_theta{$row->[$VAR_NAME]} = $row->[$THETA];
$initial_theta->assign(1,$count,$row->[$THETA]);
$count++;
}
# STEP. Calculate the names and variables for the second engine.
foreach $row (@variables)
{
my $name = $row->[$VAR_NAME];
$var_value{$name} = $shared_theta{$name};
$var_eng2{$name} = $var_value{$name};
}
while(1)
{
# SPSA coefficients indexed by variable.
# my ($theta, $cost, $var_eng1, %var_eng2);
my $iter;
my $optimization_done = 0;
{
lock($shared_lock);
# STEP. Increase the shared interation shared_iter
if (++$shared_iter > $iterations)
{
engine_quit() if (!$simulate);
return;
}
$iter = $shared_iter; # print "$iter \n";
my $random = Math::MatrixReal->new_random(1,$n_parameters,{bounded_by=>[-1,1]});
my $policy_eps = $policy + $random; # print "$policy_eps\n";
$var_eng1 = $theta + $policy_eps * $sigma;
# STEP. Play two games (with alternating colors) and obtain the score from eng1 perspective.
my $result = ($simulate ? simulate_2games(\$var_eng1, \%var_eng2) : engine_2games(\%var_eng2));
$theta = $var_eng1;# print "$theta \n";
$result_inc = $result_inc + $result;
$cost = 1 - 1 / (1 + 10 ** (-$result_inc / 400)); print "$cost \n";
# Record the policy, theta, and cost
$policy_history->assign_row($iter,$policy);# print "$policy_history \n";
$theta_history->assign_row($iter,$theta); # print "$theta_history \n";
$cost_history->assign(1,$iter,$cost);# print "$cost_history \n";
if ($cost < $best_cost) {
$best_cost = $cost;
$best_policy = $policy_history->row($iter);# print "$best_policy \n"
}
### Take a sample of near policies
my $temp_coef = 1.0;
my $counter = 0;
my $min_minus_mean = $shared_mean_minus_min;
my $policy_old = $shared_policy_old;
my $near_policies = Math::MatrixReal->new($n_parameters,$n_parameters);
my $near_policy_costs = Math::MatrixReal->new(1,$n_parameters);
my $cur_policy_new = $initial_policy;
my $cur_policy_new2 = $initial_policy;
while(1){
for (my $sample=max($iter-$n_parameters,1);$sample<$iter;$sample++) {
if ($temp_coef*$range > ($policy_history->row($sample) - $policy) * $covar_inv * ~($policy_history->row($sample) - $policy))
{$counter++;
$near_policies->assign_row($counter,$policy_history->row($sample));# print "$near_policies\n";
$near_policy_costs->assign(1,$counter,$cost_history->element(1,$sample)); #print "$near_policy_costs\n";#print "$policy_history\n";
}
$temp_coef = $temp_coef * 3.0;
}
if ($counter > 1) {
$near_bins_size = $counter;# print "Near bins:", "$near_bins_size\n";#print "Near policies:", "$near_policies \n";print "Near policy costs:", "$near_policy_costs\n";
last; }
else {$near_bins_size = 1; last;}
} #end while loop
my $near_costs_block = Math::MatrixReal->new(1,$near_bins_size);
for (my $i=1;$i<$near_bins_size+1;$i++) {
$near_policy_costs->element(1,$i) > 0 ? $near_costs_block->assign(1,$i,$near_policy_costs->element(1,$i)) : $near_costs_block->assign(1,$i,1);
}
### Determine the minimal cost and the minimal index
my $min_cost = min($near_costs_block->as_list());# print "$near_costs_block\n"; print "$min_cost\n";
# my $min_cost = $near_costs_block->minimum();# print "$near_costs_block\n"; print "$min_cost\n";
my $min_index;
for (my $i=1;$i<$near_bins_size+1;$i++) {
$min_index = $i unless $near_costs_block->element(1,$i) > $min_cost;
}
### Determine the minimal cost with ordinary vector (not MatrixReal one)
# my @near_costs_block = ();
# for (my $i=1;$i<$near_bins_size+1;$i++) {
# $near_policy_costs->element(1,$i) > 0 ? push @near_costs_block, $near_policy_costs->element(1,$i) : push @near_costs_block,1;
# }
# my $min_cost = min(@near_costs_block); print "@near_costs_block\n"; print "$min_cost\n";
# my $min_index;
# for (my $i=1;$i<$near_bins_size+1;$i++) {
# $min_index = $i unless $near_costs_block->element(1,$i) > $min_cost;
# }
# my $min_index;
# for (my $i=0;$i<$near_bins_size;$i++) {
# $min_index = $i+1 unless $near_costs_block[$i] > $min_cost;
# }
# print "$min_index\n";
### Separate the near policies in a block and remove the zero vectors
my $near_policy_block = Math::MatrixReal->new($near_bins_size,$n_parameters);
for (my $i=1;$i<$near_bins_size+1;$i++) {
$near_policy_block->assign_row($i,$near_policies->row($i));
}
# print "$near_policy_block\n";
### Determine the mean cost and subtract minimal cost from it
# my $mean_cost = sum(@near_costs_block) / $near_bins_size; print "$mean_cost\n";
my $mean_cost = sum($near_costs_block->as_list()) / $near_bins_size;# print "$mean_cost\n";
# my $mean_cost = $near_costs_block->norm_sum() / $near_bins_size; print "$mean_cost\n";
my $best_within_near_policies = $near_policy_block->row($min_index);# print "$best_within_near_policies\n";
$shared_mean_minus_min = $mean_cost - $min_cost; #print "$shared_mean_minus_min \n";
my $mean_minus_min = $shared_mean_minus_min; #print "$mean_minus_min\n";
### Create a matrix for use in gradient descent instead of vector for loop
# my $diffoftheta = $near_policy_block - $policy_block; print "$diffoftheta\n";
# my $diffscalar = ~$near_policy_block * $policy_block; print "$diffscalar\n";
# my @sum = $diffscalar->as_list; print "$sum\n";
### Choose current policies and costs depending on the results from gradient descent
# my $cur_policy_new = $initial_policy;
# my $cur_policy_new2 = $initial_policy;
if ($mean_minus_min != 0) {
my ($e_cost,$cur_policy_new) = gradient_descent($near_costs_block,$near_policy_block,$mean_cost,$near_bins_size,$policy,$covar,$covar_inv);# print "$e_cost\n";
#print "$e_cost\n";print "$cur_policy_new\n";
my ($e_cost2,$cur_policy_new2) = gradient_descent($near_costs_block,$near_policy_block,$mean_cost,$near_bins_size,$best_within_near_policies,$covar,$covar_inv);# print "$e_cost\n";
if ($e_cost > $e_cost2) {
$cur_policy_new = $cur_policy_new2;
$e_cost = $e_cost2;
# print "$e_cost\n";#print "$cur_policy_new\n";
}
($e_cost2,$cur_policy_new2) = gradient_descent($near_costs_block,$near_policy_block,$mean_cost,$near_bins_size,$policy_eps,$covar,$covar_inv);# print "$e_cost\n";
if ($e_cost > $e_cost2) {
$cur_policy_new = $cur_policy_new2;
$e_cost = $e_cost2;
#print "$e_cost\n";print "$cur_policy_new\n";
}
if ($min_cost != $best_cost) {
($e_cost2,$cur_policy_new2) = gradient_descent($near_costs_block,$near_policy_block,$mean_cost,$near_bins_size,$best_policy,$covar,$covar_inv);# print "$e_cost\n";
if ($e_cost > $e_cost2) {
$cur_policy_new = $cur_policy_new2;
# print "$e_cost\n";#print "$cur_policy_new\n";
}
}
# }
### Adjust sigma depending on the previous cost
$sigma = $sigma * (1 + $expansion_factor_sigma) if $result < 0;
$sigma = $sigma / ((1 + $expansion_factor_sigma) ** $imp_factor) if $result > 0;
print "$sigma\n"; # print "$expansion_factor_sigma\n";
# STEP. Apply the result
{
lock($shared_lock);
my $logLine = "$iter";
my $logLine1;
my $logLine2;
foreach $row (@variables)
{
my $name = $row->[$VAR_NAME];
$logLine .= " ,$var_value{$name}";
# $logLine1 .= ",";
# $logLine2 .= ", ";
}
$logLine2 = "$var_eng1";
print LOG "$logLine \n $logLine2 \n";
}
### Adjust the covariant matrix
if (($cur_policy_new-$policy) * $covar_inv * ~($cur_policy_new-$policy) < ($chiN * 1.5)**2) {
$pc = (1.0 - $cc) * $pc + $cc * ~($cur_policy_new-$policy) / $sigma;
$c_normalized = (1.0 - $ccov) * $c_normalized + $pc * ~$pc * $ccov;
my $determinant = $c_normalized->det();
$c_normalized = $c_normalized * 1.0 / $determinant / $n_parameters if $determinant != 0; #print $c_normalized->det();
}
### Enforcing symmetry
# for(my $k=1;$k<$n_parameters+1;$k++) {
# for (my $i=$k+1;$i<$n_parameters+1;$i++) {
# $c_normalized->element($k,$i) = $c_normalized->element($i,$k);
# }}
$c_normalized = 0.5 * ($c_normalized + ~$c_normalized);
### Update the covariant matrix, its inverse and the current policy
$covar = $c_normalized * $sigma * $sigma;# print "$covar\n";
$covar_inv = $covar->inverse();# print "$covar_inv\n";
$policy = $cur_policy_new; print "$cur_policy_new\n";
$optimization_done = 1 if ($sigma < 1.0e-8);
}
}
# print "Iteration: $iter\n, variable: $name, cost: $var_eng1{$name}, theta: $var_eng2{$name}, policy: $policy{$name}, policy_cur: $policy_cur{$name}\n";
}
# STEP. Close log
close(GAMELOG);
}
### SECTION. Simulating a game
sub simulate_ELO
{
my ($var) = @_;
my $ELO = 0.0;
foreach my $key (keys(%$var))
{
my $a = -0.0001 * $variableIdx{$key}[$VAR_SIMUL_ELO];
$ELO += $a * $var->{$key} ** 2;
}
return $ELO;
}
sub simulate_winPerc
{
my ($ELO_A, $ELO_B) = @_;
my $Q_A = 10 ** ($ELO_A / 400);
my $Q_B = 10 ** ($ELO_B / 400);
return $Q_A / ($Q_A + $Q_B);
}
sub simulate_2games
{
my ($var_eng1, $var_eng2) = @_;
my $eng1_elo = simulate_ELO($var_eng1);
my $eng2_elo = simulate_ELO($var_eng2);
my $eng1_winperc = simulate_winPerc($eng1_elo, $eng2_elo);
return $eng1_winperc * rand() * (rand() < $eng1_winperc ? 1 : -1);
# return (rand() < $eng1_winperc ? 1 : -1) + (rand() < $eng1_winperc ? 1 : -1);
}
### SECTION. Playing a game
my @fenlines;
my ($eng1_pid, $eng2_pid);
local (*Eng1_Reader, *Eng1_Writer);
local (*Eng2_Reader, *Eng2_Writer);
sub engine_init
{
# STEP. Read opening book.
open(INPUT, "<$epd_path");
binmode(INPUT);
my @lines;
(@lines) = <INPUT>;
@fenlines = grep {/\w+/} @lines; # Filter out empty lines
close (INPUT);
die "epd read failure!" if ($#fenlines == -1);
# STEP. Launch engines.
$eng1_pid = open2(\*Eng1_Reader, \*Eng1_Writer, $eng1_path);
$eng2_pid = open2(\*Eng2_Reader, \*Eng2_Writer, $eng2_path);
# STEP. Init engines
my $line;
print Eng1_Writer "uci\n";
print Eng2_Writer "uci\n";
while(engine_readline(\*Eng1_Reader) ne "uciok") {}
while(engine_readline(\*Eng2_Reader) ne "uciok") {}
}
sub engine_quit
{
print Eng1_Writer "quit\n";
print Eng2_Writer "quit\n";
waitpid($eng1_pid, 0);
waitpid($eng2_pid, 0);
}
sub engine_readline
{
my ($Reader) = @_;
local $/ = $IS_WINDOWS ? "\r\n" : "\n";
my $line = <$Reader>;
chomp $line;
return $line;
}
sub engine_2games
{
my ($var_eng1,$var_eng2) = @_;
my $result = 0;
my $score = 0;
my $line;
my $cost;
my $score_count = 0;
my $sigmoid = 0;
# my $var_eng1 = $shared_var_eng1;
# STEP. Choose a random opening
my $rand_i = int(rand($#fenlines + 1));
my @tmparray = split(/\;/, $fenlines[$rand_i]);
my $fenline = $tmparray[0];
@tmparray = split(/ /, $fenline);
my $side_to_start = $tmparray[1]; #'b' or 'w'
# STEP. Send rounded values to engines
my $column = 1;
foreach my $var (keys(%$var_eng2))
{
my $val1 = nearest(1, $var_eng1->element(1,$column));
my $val2 = nearest(1, $var_eng2->{$var});
print Eng1_Writer "setoption name $var value $val1\n";# print $val2;
print Eng2_Writer "setoption name $var value $val2\n";
$column++;
}
# STEP. Play two games
for (my $eng1_is_white = 0; $eng1_is_white < 2; $eng1_is_white++)
{
# STEP. Tell engines to prepare for a new game
print Eng1_Writer "ucinewgame\n";
print Eng2_Writer "ucinewgame\n";
print Eng1_Writer "isready\n";
print Eng2_Writer "isready\n";
# STEP. Wait for engines to be ready
while(engine_readline(\*Eng1_Reader) ne "readyok") {}
while(engine_readline(\*Eng2_Reader) ne "readyok") {}
# STEP. Init Thinking times
my $eng1_time = $base_time;
my $eng2_time = $base_time;
# STEP. Check which engine should start?
my $engine_to_move = ($eng1_is_white == 1 && $side_to_start eq 'w') || ($eng1_is_white == 0 && $side_to_start eq 'b') ? 1 : 2;
print GAMELOG "Starting game using opening fen: $fenline (opening line $rand_i). Engine to start: $engine_to_move\n";
# STEP. Init game variabless
my $moves = '';
my $winner = 0;
my $draw_shared_iter = 0;
my @win_shared_iter = (0, 0, 0);
GAME: while(1)
{
my $wtime = nearest_floor(1, $eng1_is_white == 1 ? $eng1_time : $eng2_time);
my $btime = nearest_floor(1, $eng1_is_white == 0 ? $eng1_time : $eng2_time);
my $Curr_Writer = ($engine_to_move == 1 ? \*Eng1_Writer : \*Eng2_Writer);
my $Curr_Reader = ($engine_to_move == 1 ? \*Eng1_Reader : \*Eng2_Reader);
# STEP. Send engine the current positionn
print $Curr_Writer "position fen $fenline" . ($moves ne '' ? " moves $moves" : "") . "\n";
print GAMELOG "Engine " . ($engine_to_move == 1 ? '1' : '2') . " starts thinking. Time: " .
sprintf("%d", $engine_to_move == 1 ? $eng1_time : $eng2_time) . " Moves: $moves \n";
# STEP. Let it go!
my $t0 = time;
print $Curr_Writer "go wtime $wtime btime $btime winc $inc_time binc $inc_time\n";
# STEP. Read output from engine until it prints the bestmove.
my $flag_mate = 0;
my $flag_stalemate = 0;
READ: while($line = engine_readline($Curr_Reader))
{
my @array = split(/ /, $line);
# When engine is done, it prints bestmove.
if ($#array >= 0 && $array[0] eq 'bestmove') {
$flag_stalemate = 1 if ($array[1] eq '(none)');
$moves = $moves . " " . $array[1];
last READ;
}
# Check for mate in one
if ($#array >= 9 && $array[0] eq 'info' && $array[1] eq 'depth' &&
$array[7] eq 'score' && $array[8] eq 'mate' && $array[9] eq '1')
{
$flag_mate = 1;
$winner = $engine_to_move;
}
# Record score
if ($#array >= 7 && $array[0] eq 'info' && $array[1] eq 'depth' &&
$array[7] eq 'score')
{
$score = $array[9] if ($array[8] eq 'cp');
# $score = +10000 if ($array[8] eq 'mate' && $array[9] > 0);
# $score = -10000 if ($array[8] eq 'mate' && $array[9] < 0);
$sigmoid = 0.5 - 1.0 / (1 + 10.0 ** -(0.7 * $score / 10000))
}
}
print GAMELOG "Score: $score\n" if defined($score);
# STEP. Update thinking times
my $elapsed = time - $t0;
$eng1_time = $eng1_time - ($engine_to_move == 1 ? $elapsed * 1000 - $inc_time : 0);
$eng2_time = $eng2_time - ($engine_to_move == 2 ? $elapsed * 1000 - $inc_time : 0);
# STEP. Check for mate and stalemate
if ($flag_mate)
{
$winner = $engine_to_move;
last GAME;
}
if ($flag_stalemate)
{
$winner = 0;
last GAME;
}
# STEP. Update draw shared_iter
$draw_shared_iter = (abs($score) <= $draw_score_limit ? $draw_shared_iter + 1 : 0);
print GAMELOG "Draw shared_iter: $draw_shared_iter / $draw_move_limit\n" if ($draw_shared_iter);
if ($draw_shared_iter >= $draw_move_limit)
{
$winner = 0;
last GAME;
}
# STEP. Update win shared_iters
my $us = $engine_to_move;
my $them = $engine_to_move == 1 ? 2 : 1;
$win_shared_iter[$us] = ($score >= +$win_score_limit ? $win_shared_iter[$us] + 1 : 0);
$win_shared_iter[$them] = ($score <= -$win_score_limit ? $win_shared_iter[$them] + 1 : 0);
$score = $engine_to_move == 1 ? $score : 0; # print "$score\n";
$score_count++;
print GAMELOG "Win shared_iter: $win_shared_iter[$us] / $win_move_limit\n" if ($win_shared_iter[$us]);
print GAMELOG "Loss shared_iter: $win_shared_iter[$them] / $win_move_limit\n" if ($win_shared_iter[$them]);
if ($win_shared_iter[$us] >= $win_move_limit)
{
$winner = $us;
last GAME;
}
if ($win_shared_iter[$them] >= $win_move_limit)
{
$winner = $them;
last GAME;
}
# STEP. Change turn
$engine_to_move = $them;
}
# STEP. Record the result
print GAMELOG "Winner: $winner\n";
### Here, the aim is to code the Texel formula for the objective function,
### with the result calculated AFTER the score. Not sure if correct.
### One way is to sum all scores in one sigmoid and divide by score_count (mean score)
### Another way is to sum all sigmoids and divide by score_count (mean sigmoid).
# print $score_count;
$result += ($winner == 1 ? 1 : $winner == 2 ? -1 : 0); #print "$result\n"
# $result = ($winner == 1 ? 1 : $winner == 2 ? 0 : 0.5); #print "$result\n"
$score += $score;
my $mean_score = $score / $score_count;
$cost = ($result - 1.0 / (1 + 10.0 ** -(0.7 * $mean_score / 100))) ** 2; #print "$cost \n";
}
# $score += $score; print "$score\n";
$cost += $cost;
return $result;
}
######################################################
#sub take_sample
#{ lock($shared_lock);
# my ($policy,$policy_history,$cost_history) = @_;
# my $temp_coef = 1.0;
# my $counter = 0;
# my $iter = $shared_iter;
# my $min_minus_mean = $shared_mean_minus_min;
# my $policy_old = $shared_policy_old;
# my $near_policies = Math::MatrixReal->new($n_parameters,$n_parameters);
# my $near_policy_costs = Math::MatrixReal->new(1,$n_parameters);
#while(1){
# for (my $sample=max($iter-$n_parameters,1);$sample<$iter;$sample++) {
# if ($temp_coef*$range > ($policy_history->row($sample) - $policy) * $covar_inv * ~($policy_history->row($sample) - $policy))
# {$counter++;
# $near_policies->assign_row($counter,$policy_history->row($sample)); #print "$near_policies\n";
# $near_policy_costs->assign(1,$counter,$cost_history->element(1,$sample)); #print "$near_policy_costs\n";#print "$policy_history\n";
# }
# $temp_coef = $temp_coef * 3.0;
# }
# if ($counter > 1) {
# $near_bins_size = $counter; print "Near bins:", "$near_bins_size\n";#print "Near policies:", "$near_policies \n";print "Near policy costs:", "$near_policy_costs\n";
# last; }
#} #end while loop
# my $near_costs_block = Math::MatrixReal->new(1,$near_bins_size);
# for (my $i=1;$i<$near_bins_size+1;$i++) {
# $near_policy_costs->element(1,$i) > 0 ? $near_costs_block->assign(1,$i,$near_policy_costs->element(1,$i)) : $near_costs_block->assign(1,$i,1);
# }
# my @near_costs_block = ();
# for (my $i=1;$i<$near_bins_size+1;$i++) {
# $near_policy_costs->element(1,$i) > 0 ? push @near_costs_block, $near_policy_costs->element(1,$i) : push @near_costs_block,1;
# }
# my $min_cost = min($near_costs_block->as_list());# print "$near_costs_block\n"; print "$min_cost\n";
# my $min_cost = $near_costs_block->minimum();# print "$near_costs_block\n"; print "$min_cost\n";
# my $min_index;
# for (my $i=1;$i<$near_bins_size+1;$i++) {
# $min_index = $i unless $near_costs_block->element(1,$i) > $min_cost;
# }
# my $min_cost = min(@near_costs_block); print "@near_costs_block\n"; print "$min_cost\n";
# my $min_index;
# for (my $i=1;$i<$near_bins_size+1;$i++) {
# $min_index = $i unless $near_costs_block->element(1,$i) > $min_cost;
# }
# my $min_index;
# for (my $i=0;$i<$near_bins_size;$i++) {
# $min_index = $i+1 unless $near_costs_block[$i] > $min_cost;
# }
# print "$min_index\n";
# my $near_policy_block = Math::MatrixReal->new($near_bins_size,$n_parameters);
# for (my $i=1;$i<$near_bins_size+1;$i++) {
# $near_policy_block->assign_row($i,$near_policies->row($i));
# }
# print "$near_policy_block\n";
# my $mean_cost = sum(@near_costs_block) / $near_bins_size; print "$mean_cost\n";
# my $mean_cost = sum($near_costs_block->as_list()) / $near_bins_size; #print "$mean_cost\n";
# my $mean_cost = $near_costs_block->norm_sum() / $near_bins_size; print "$mean_cost\n";
# my $best_within_near_policies = $near_policy_block->row($min_index);# print "$best_within_near_policies\n";
# $shared_mean_minus_min = $mean_cost - $min_cost; #print "$shared_mean_minus_min \n";
# my $mean_minus_min = $shared_mean_minus_min; #print "$mean_minus_min\n";
# my $policy_block = Math::MatrixReal->new($near_bins_size,$n_parameters);
# for (my $i=1;$i<$near_bins_size+1;$i++) {
# $policy_block->assign_row($i,$near_policies->row($i));
# }
# print "$policy_block\n";
# my $diffoftheta = $near_policy_block - $policy_block; print "$diffoftheta\n";
# my $diffscalar = ~$near_policy_block * $policy_block; print "$diffscalar\n";
# my @sum = $diffscalar->as_list; print "$sum\n";
# my ($e_cost,$cur_policy_new) = gradient_descent($near_costs_block,$near_policy_block,$mean_cost,$near_bins_size,$policy_old,$policy,$covar,$covar_inv);# print "$e_cost\n";
#print "$cur_policy_new\n";
# return ($near_bins_size,$near_policy_costs,$near_policies);
#} #end while
#} # end take_sample
#######################################################################
### Natural gradient descent
sub gradient_descent {
lock($shared_lock);
my $iter = $shared_iter;
my $mean_minus_min = $shared_mean_minus_min;
my ($near_costs_block,$near_policy_block,$mean_cost,$near_bins_size,$policy,$covar,$covar_inv) = @_;
my $op_cost;
#if ($mean_minus_min != 0) {
my ($Pprior,$a,$b,$expMD2,$probabilistic_distance,$op_policy,$residual_term,$adot,$bdot,$Jaco,$update,$diffoftheta);
my $policy_old = $policy;
### $Pprior = 2.0 / pow(sqrt(2.0 * M_PI), n_parameter);
$Pprior = 1.0;
my $terminate = 0;
### STEP: Autotune alpha according to 50% in every step criteria
# my $covar_factored = $covar * $cost2policy_cov_factor;
my $alpha = 10.0 * $covar * $cost2policy_cov_factor / $mean_minus_min;
# my $covar_inv = $covar->inverse();
for(my $i = 0; $i < 200; $i++){
$a = $Pprior * $mean_cost;
$b = $Pprior;
for(my $row=1;$row<$near_bins_size+1;$row++){
$diffoftheta = $near_policy_block->row($row) - $policy_old; #print "$diffoftheta\n";
$residual_term = 0.5 * $diffoftheta * $covar_inv / $cost2policy_cov_factor; #print "$residual_term\n";
my $diff_covar = ~(-0.25 * $diffoftheta * $covar_inv / $cost2policy_cov_factor); #print "$diff_covar\n";
my $MD2 = $diff_covar->scalar_product(~$diffoftheta);# print "$MD2\n";
$expMD2 = exp $MD2; #print "$expMD2\n";
$adot = $near_costs_block->element(1,$row) * $expMD2 * $residual_term; #print "$adot\n";
$bdot = $expMD2 * $residual_term; #print "$bdot\n";
$a += $near_costs_block->element(1,$row) * $expMD2; #print "$a\n";
$b += $expMD2; #print "$b\n";
last if ($terminate);
# $covar_inv = $covar->inverse();
$Jaco = $adot / $b - $bdot * $a / $b / $b; #print "$Jaco\n"; print "$alpha\n";
$update = $Jaco * $alpha; #print "$update\n";
$op_policy = $policy_old - $update; #print "$op_policy\n";
$probabilistic_distance = $update * $covar_inv / $cost2policy_cov_factor * ~$update;# print "$probabilistic_distance\n";
$policy_old = $op_policy;
$terminate = 1 if ($probabilistic_distance < 0.0001 || ($op_policy-$policy) * $covar_inv / $cost2policy_cov_factor * ~($op_policy-$policy) > 1.0);
} #end for inner clause
$op_cost = $a / $b;
} #end for outer clause
return ($op_cost,$policy_old);
#} # end if clause
} # end gradient_descent