-
Notifications
You must be signed in to change notification settings - Fork 0
/
main.jl
2256 lines (1696 loc) · 97.1 KB
/
main.jl
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
using LinearAlgebra, Statistics, Plots, Distributions, StatsBase, DelimitedFiles, CSV, Tables, Random, BenchmarkTools
using XGBoost
#using BayesianOptimization, GaussianProcesses, Distributions
using Flux
using BSON: @load
using StatsBase
using Suppressor
#BLAS.set_num_threads(1)
mutable struct exp4
policy_list::Vector{Function}
policy_probs::Vector{Float64}
#policy::Function
end
include("rollouts.jl")
include("policies.jl")
include("opt_alloc.jl")
include("coordinated_utilities.jl")
include("glm.jl")
include("mcmc.jl")
include("vb.jl")
include("gp.jl")
include("mab.jl")
#cnst mts = MersenneTwister.(1:Threads.nthreads())
# Parameters
const context_dim = 5
const context_mean = 0
const context_sd = .25
# set to false for gp
const context_constant = true
const obs_sd = 4
const bandit_count = 3
const bandit_prior_mean = 0
const bandit_prior_sd = 1
# MCMC parameters
const prior_mean = repeat([bandit_prior_mean], context_dim)
const prior_cov = diagm(repeat([bandit_prior_sd^2], context_dim))
const proposal_sd = .1
const n_burn = 200
const n_dup = 10
# Multi Action
const multi_count = 10
# SIMULATION HORIZON
const T = 100
# NUMBER OF GLOBAL SIMULATION EPISODES (PER INDEX JOB)
const n_episodes = 5
# DISCOUNT PARAMETER
const discount = .98
# PARAMETER FOR EPSILON GREEDY POLICY
const epsilon = .4
const decreasing = true
# PARAMETERS FOR ALL ROLLOUT METHODS
const rollout_length = 50# 20
const n_rollouts = 10000 # 100000
# PARAMETERS FOR SPSA OPTIMIZATION METHOD
const n_opt_rollouts = 10000# 10000hh0
const n_spsa_iter = 20000
## PARAMETERS FOR GRID OPTIMIZATION METHOD
const n_grid_iter = 7
const grid_ratio = 2
const grid_num = 6
const int_length = 2
const n_grid_rollouts = 50
### NON ADAPTIVE GRID (n_opt_rollouts used for opt rollouts)
#grid_margin_1 = [0, .5, 1, 2, 4]
#grid_margin_2 = [0, .5, 1, 2]
#grid_margin_3 = [.5, 1, 2]
grid_margin_1 = [0., 1.]
grid_margin_2 = [0., 1.]
grid_margin_3 = [1., 2.]
### NON STATIONARY PARAMETERS
const delta = 1.
### GP PARAMETERS
const kernel_scale = 10
const kernel_bandwidth = 10
### MAB PARAMETERS
const a_beta_prior = .5
const b_beta_prior = .5
### VB PARAMETER
const vb_rollout_mult = 5
const vb_rollout_tol = .00001
const vb_policy_tol = .01
### ADAPTIVE PARAM
const expected_regret_thresh = .0001
const action_expected_regret_thresh = .0001
const dp_expected_regret_thresh = .0001
## DP PARAM
const dp_rollout_count = 4000
const dp_within_rollout_count = 50
const num_round = 3
const dp_length = 3
const param = ["max_depth" => 3,
"eta" => .5,
"objective" => "reg:squarederror",
"verbosity" => 0]
const metrics = ["rmse"]
## SIMULATOR FUNCTION
const idx = Base.parse(Int, ENV["SLURM_ARRAY_TASK_ID"])
Random.seed!(idx + 42)
neural_net_list = []
scale_list = []
multi_neural_net_list = []
multi_scale_list = []
bern_neural_net_list = []
bern_scale_list = []
use_nn = false
### UNKNOWN CONTEXT DIST
const use_context_prior = true
const est_context_var_noise = .5
const ν = 2 / est_context_var_noise^2 + context_dim - context_constant + 3
const Ψ = context_sd^2 * (ν - context_dim - context_constant - 1) * Matrix(1.0I, context_dim - context_constant, context_dim - context_constant)
const λ = 1
const μ = zeros(context_dim -context_constant)
function generate_context(context_dim, context_mean, context_sd, context_constant)
if context_constant
return [1; context_mean .+ context_sd .* randn(context_dim - 1)]
else
return context_mean .+ context_sd .* randn(context_dim)
end
end
function generate_context_mvn(context_dim, context_mean, context_cov, context_constant)
if context_constant
return [1; rand(MvNormal(context_mean, context_cov))]
else
return rand(MvNormal(context_mean, context_cov))
end
end
function generate_context_mvn(sampler, context_constant)
if context_constant
return [1; rand(sampler)]
else
return rand(sampler)
end
end
if use_context_prior
const ground_truth_context_covs = rand(InverseWishart(ν, Ψ), n_episodes)
const ground_truth_context_means = [rand(MvNormal(zeros(context_dim - context_constant), iw ./ λ)) for iw in ground_truth_context_covs]
const ground_truth_context = [[generate_context_mvn(MvNormal(ground_truth_context_means[i], ground_truth_context_covs[i]), context_constant) for t in 1:T] for i in 1:n_episodes]
else
const ground_truth_context = [[generate_context(context_dim, context_mean, context_sd, context_constant) for t in 1:T] for i in 1:n_episodes]
end
#GC.enable_logging(true)
## NEED TO DECIDE WHETHER TRUE IS USED
if use_nn
for i in 1:T
@load "/hpc/home/jml165/rl/valneuralnets/valnn_$(i).bson" m
scales = CSV.File("/hpc/home/jml165/rl/neuralnetscales/scales_$(i).csv", header = false) |> Tables.matrix
push!(neural_net_list, m)
push!(scale_list, scales)
end
for i in 1:T
@load "/hpc/home/jml165/rl/multivalneuralnets/multivalnn_$(i).bson" m
scales = CSV.File("/hpc/home/jml165/rl/multineuralnetscales/multiscales_$(i).csv", header = false) |> Tables.matrix
push!(multi_neural_net_list, m)
push!(multi_scale_list, scales)
end
for i in 1:T
@load "/hpc/home/jml165/rl/bernvalneuralnets/bernvalnn_$(i).bson" m
scales = CSV.File("/hpc/home/jml165/rl/bernneuralnetscales/bernscales_$(i).csv", header = false) |> Tables.matrix
push!(bern_neural_net_list, m)
push!(bern_scale_list, scales)
end
end
upper_triangular_vec = function(M)
d = size(M)[1]
output = zeros(convert(Int64,(d+1) * d / 2))
count = 1
for i in 1:d
for j in i:d
output[count] = M[i, j]
count += 1
end
end
return output
end
const ground_truth_param = bandit_prior_mean .+ bandit_prior_sd .* randn(bandit_count, context_dim, n_episodes)
const ground_truth_obs = obs_sd .* randn(T, n_episodes)
const ground_truth_obs_bern = rand(T, n_episodes)
function update_context_posterior!(context, posterior)
if context_constant
posterior[4] .+= (posterior[2] / (1 + posterior[2])) .* ((context[2:end] .- posterior[1]) * (context[2:end] .- posterior[1])')
posterior[1] .*= (posterior[2] / (1 + posterior[2]))
posterior[1] .+= context[2:end] ./ (1 + posterior[2])
posterior[2] += 1
posterior[3] += 1
else
posterior[4] .+= (posterior[2] / (1 + posterior[2])) .* ((context .- posterior[1]) * (context .- posterior[1])')
posterior[1] .*= (posterior[2] / (1 + posterior[2]))
posterior[1] .+= context ./ (1 + posterior[2])
posterior[2] += 1
posterior[3] += 1
end
end
function ep_contextual_bandit_simulator(ep,action_function, T, rollout_length, n_episodes, n_rollouts, n_opt_rollouts, context_dim, context_mean,
context_sd, obs_sd, bandit_count, bandit_prior_mean, bandit_prior_sd, discount, epsilon, global_bandit_param)
bandit_posterior_means = zeros(bandit_count, context_dim)
bandit_posterior_covs = zeros(bandit_count, context_dim, context_dim)
bandit_param = copy(global_bandit_param)
true_bandit_param = copy(global_bandit_param)
EPREWARDS = zeros(T)
EPOPTREWARDS = zeros(T)
for i in 1:bandit_count
bandit_posterior_means[i, :] = repeat([bandit_prior_mean], context_dim)
bandit_posterior_covs[i, :, :] = Diagonal(repeat([bandit_prior_sd^2], context_dim))
end
for t in 1:T
#context = generate_context(context_dim, context_mean, context_sd, context_constant)
context = ground_truth_context[ep][t]
true_expected_rewards = true_bandit_param * context
#true_expected_rewards = bandit_posterior_means * context
action = action_function(ep, t, T, bandit_count, context, bandit_posterior_means, bandit_posterior_covs, discount, epsilon, rollout_length, n_rollouts, n_opt_rollouts, context_dim)
true_expected_reward = true_expected_rewards[action]
EPREWARDS[t] = true_expected_reward
EPOPTREWARDS[t] = maximum(true_expected_rewards)
#obs = randn() * sqrt(obs_sd^2 + dot(context, bandit_posterior_covs[action,:,:],context)) + true_expected_reward
obs = ground_truth_obs[t, ep] + true_expected_reward
old_cov = bandit_posterior_covs[action, :, :]
CovCon = old_cov * context ./ obs_sd
#bandit_posterior_covs[action, :, :] = inv(context * context' / obs_sd^2 + old_precision)
bandit_posterior_covs[action, :, :] = old_cov - CovCon * CovCon' ./ (1 + dot(context, old_cov, context) / obs_sd^2)
bandit_posterior_covs[action, :, :] = ((bandit_posterior_covs[action,:,:]) + bandit_posterior_covs[action,:,:]')/2
bandit_posterior_means[action, :] = (bandit_posterior_covs[action, :, :]) * (old_cov \ (bandit_posterior_means[action,:]) + context * obs / obs_sd^2)
println("Ep: ", ep, " - ", t, " of ", T, " for ", String(Symbol(action_function)))
flush(stdout)
#GC.gc()
end
return EPREWARDS, EPOPTREWARDS
end
function ep_contextual_bandit_simulator_context_mvn(ep,action_function, T, rollout_length, n_episodes, n_rollouts, n_opt_rollouts, context_dim, context_mean,
context_sd, obs_sd, bandit_count, bandit_prior_mean, bandit_prior_sd, discount, epsilon, global_bandit_param)
bandit_posterior_means = zeros(bandit_count, context_dim)
bandit_posterior_covs = zeros(bandit_count, context_dim, context_dim)
bandit_param = copy(global_bandit_param)
true_bandit_param = copy(global_bandit_param)
EPREWARDS = zeros(T)
EPOPTREWARDS = zeros(T)
context_posterior = [copy(μ), copy(λ), copy(ν), copy(Ψ)]
for i in 1:bandit_count
bandit_posterior_means[i, :] = repeat([bandit_prior_mean], context_dim)
bandit_posterior_covs[i, :, :] = Diagonal(repeat([bandit_prior_sd^2], context_dim))
end
for t in 1:T
#context = generate_context(context_dim, context_mean, context_sd, context_constant)
context = ground_truth_context[ep][t]
update_context_posterior!(context, context_posterior)
true_expected_rewards = true_bandit_param * context
#true_expected_rewards = bandit_posterior_means * context
action = action_function(ep, t, T, bandit_count, context, bandit_posterior_means, bandit_posterior_covs, discount, epsilon, rollout_length, n_rollouts, n_opt_rollouts, context_dim, context_posterior)
true_expected_reward = true_expected_rewards[action]
EPREWARDS[t] = true_expected_reward
EPOPTREWARDS[t] = maximum(true_expected_rewards)
#obs = randn() * sqrt(obs_sd^2 + dot(context, bandit_posterior_covs[action,:,:],context)) + true_expected_reward
obs = ground_truth_obs[t, ep] + true_expected_reward
old_cov = bandit_posterior_covs[action, :, :]
CovCon = old_cov * context ./ obs_sd
#bandit_posterior_covs[action, :, :] = inv(context * context' / obs_sd^2 + old_precision)
bandit_posterior_covs[action, :, :] = old_cov - CovCon * CovCon' ./ (1 + dot(context, old_cov, context) / obs_sd^2)
bandit_posterior_covs[action, :, :] = ((bandit_posterior_covs[action,:,:]) + bandit_posterior_covs[action,:,:]')/2
bandit_posterior_means[action, :] = (bandit_posterior_covs[action, :, :]) * (old_cov \ (bandit_posterior_means[action,:]) + context * obs / obs_sd^2)
println("Ep: ", ep, " - ", t, " of ", T, " for ", String(Symbol(action_function)))
flush(stdout)
#GC.gc()
end
return EPREWARDS, EPOPTREWARDS
end
function contextual_bandit_simulator(action_function, T, rollout_length, n_episodes, n_rollouts, n_opt_rollouts, context_dim, context_mean,
context_sd, obs_sd, bandit_count, bandit_prior_mean, bandit_prior_sd, discount, epsilon)
REWARDS = zeros(T, n_episodes)
OPTREWARDS = zeros(T, n_episodes)
ep_count = 1
## USING BAYESIAN FORMULATION FOR FAIR COMPARISON
#global_bandit_param = [1 0; 0 1; 2 -1]
#threadreps = zeros(Threads.nthreads())
for ep in 1:n_episodes
#global_bandit_param = rand(Normal(bandit_prior_mean, bandit_prior_sd), bandit_count, context_dim)
global_bandit_param = ground_truth_param[:, :, ep]
if use_context_prior
EPREWARDS, EPOPTREWARDS = ep_contextual_bandit_simulator_context_mvn(ep,action_function, T, rollout_length, n_episodes, n_rollouts, n_opt_rollouts, context_dim, context_mean,
context_sd, obs_sd, bandit_count, bandit_prior_mean, bandit_prior_sd, discount, epsilon, global_bandit_param)
else
EPREWARDS, EPOPTREWARDS = ep_contextual_bandit_simulator(ep,action_function, T, rollout_length, n_episodes, n_rollouts, n_opt_rollouts, context_dim, context_mean,
context_sd, obs_sd, bandit_count, bandit_prior_mean, bandit_prior_sd, discount, epsilon, global_bandit_param)
end
ep_count += 1
REWARDS[:, ep] = EPREWARDS
OPTREWARDS[:, ep] = EPOPTREWARDS
end
#print(threadreps)
return REWARDS', OPTREWARDS'
end
function multi_ep_contextual_bandit_simulator(ep,action_function, T, rollout_length, n_episodes, n_rollouts, n_opt_rollouts, context_dim, context_mean,
context_sd, obs_sd, bandit_count, bandit_prior_mean, bandit_prior_sd, discount, epsilon, global_bandit_param)
bandit_posterior_means = zeros(bandit_count, context_dim)
bandit_posterior_covs = zeros(bandit_count, context_dim, context_dim)
bandit_param = copy(global_bandit_param)
true_bandit_param = copy(global_bandit_param)
EPREWARDS = zeros(T)
EPOPTREWARDS = zeros(T)
actions = zeros(multi_count)
contexts = zeros(multi_count, context_dim)
obss = zeros(multi_count)
for i in 1:bandit_count
bandit_posterior_means[i, :] = repeat([bandit_prior_mean], context_dim)
bandit_posterior_covs[i, :, :] = Diagonal(repeat([bandit_prior_sd^2], context_dim))
end
for t in 1:T
for m in 1:multi_count
context = generate_context(context_dim, context_mean, context_sd, context_constant)
contexts[m, :] = context
true_expected_rewards = true_bandit_param * context
#true_expected_rewards = bandit_posterior_means * context
action = action_function(t, T, bandit_count, context, bandit_posterior_means, bandit_posterior_covs, discount, epsilon, rollout_length, n_rollouts, n_opt_rollouts, context_dim)
actions[m] = action
true_expected_reward = true_expected_rewards[action]
EPREWARDS[t] += true_expected_reward
EPOPTREWARDS[t] += maximum(true_expected_rewards)
#obs = randn() * sqrt(obs_sd^2 + dot(context, bandit_posterior_covs[action,:,:],context)) + true_expected_reward
obss[m] = randn() * obs_sd + true_expected_reward
end
for act in 1:bandit_count
action_obss = obss[actions .== act]
action_contexts = contexts[actions .== act, :]
old_cov = bandit_posterior_covs[act, :, :]
#CovCon = old_cov * context ./ obs_sd
#bandit_posterior_covs[act, :, :] = old_cov - CovCon * CovCon' ./ (1 + dot(context, old_cov, context))
#bandit_posterior_covs[act, :, :] = ((bandit_posterior_covs[act,:,:]) + bandit_posterior_covs[act,:,:]')/2
#bandit_posterior_means[act, :] = (bandit_posterior_covs[act, :, :]) * (old_cov \ (bandit_posterior_means[act,:]) + context * obs / obs_sd^2)
#
bandit_posterior_covs[act, :, :] = inv(inv(old_cov) + action_contexts' * action_contexts ./ obs_sd^2)
bandit_posterior_covs[act, :, :] = ((bandit_posterior_covs[act,:,:]) + bandit_posterior_covs[act,:,:]')/2
bandit_posterior_means[act, :, :] = bandit_posterior_covs[act, :, :] * (old_cov \ bandit_posterior_means[act, :] + action_contexts' * action_obss ./ obs_sd^2)
end
#println("Ep: ", ep, " - ", t, " of ", T, " for ", String(Symbol(action_function)))
#flush(stdout)
end
return EPREWARDS, EPOPTREWARDS
end
function coord_greedy_ep_contextual_bandit_simulator(ep, T, rollout_length, n_episodes, n_rollouts, n_opt_rollouts, context_dim, context_mean,
context_sd, obs_sd, bandit_count, bandit_prior_mean, bandit_prior_sd, discount, epsilon, global_bandit_param)
bandit_posterior_means = zeros(bandit_count, context_dim)
bandit_posterior_covs = zeros(bandit_count, context_dim, context_dim)
bandit_param = copy(global_bandit_param)
true_bandit_param = copy(global_bandit_param)
EPREWARDS = zeros(T)
EPOPTREWARDS = zeros(T)
actions = convert(Vector{Int64}, zeros(multi_count))
contexts = zeros(multi_count, context_dim)
obss = zeros(multi_count)
for i in 1:bandit_count
bandit_posterior_means[i, :] = repeat([bandit_prior_mean], context_dim)
bandit_posterior_covs[i, :, :] = Diagonal(repeat([bandit_prior_sd^2], context_dim))
end
agree_count = 0
base_count = 0
for t in 1:T
remains = []
if decreasing
thresh = epsilon / t
else
thresh = epsilon
end
for m in 1:multi_count
context = generate_context(context_dim, context_mean, context_sd, context_constant)
contexts[m, :] = context
#true_expected_rewards = true_bandit_param * context
posterior_expected_rewards = bandit_posterior_means * context
if rand() > thresh
actions[m] = findmax(posterior_expected_rewards)[2]
else
push!(remains, m)
end
#true_expected_reward = true_expected_rewards[action]
#EPREWARDS[t] += true_expected_reward
#EPOPTREWARDS[t] += maximum(true_expected_rewards)
#obs = randn() * sqrt(obs_sd^2 + dot(context, bandit_posterior_covs[action,:,:],context)) + true_expected_reward
#obss[m] = randn() * obs_sd + true_expected_reward
end
if length(remains) > 0
post_covs_copy = copy(bandit_posterior_covs)
remain_contexts = contexts[remains, :]
for m in 1:multi_count
if ! (m in remains)
m_context = contexts[m, :]
post_covs_copy[actions[m], :, :] = inv(inv(post_covs_copy[actions[m], :, :]) + m_context * m_context' / obs_sd^2)
end
end
if length(remains) == 1
actions[remains[1]] = findmax([logdet(I + post_covs_copy[i, :, :] * contexts[remains[1], :] * contexts[remains[1], :]' / obs_sd^2) for i = 1:bandit_count])[2]
else
print("\n")
print(post_covs_copy)
print("\n")
print(remain_contexts ./ obs_sd)
print("\n")
flush(stdout)
alloc = KL_optimal_allocations(post_covs_copy, remain_contexts ./ obs_sd)
remain_actions = [findmax(alloc[i, :])[2] for i = 1:length(remains)]
for i in 1:length(remains)
actions[remains[i]] = remain_actions[i]
end
end
end
# if length(remains) > 0
# max_val = 0
# free_actions = copy(actions)
# max_tup = zeros(length(remains))
# for tup in CartesianIndices(ntuple(d->1:bandit_count, length(remains)))
# for i in 1:length(remains)
# free_actions[remains[i]] = tup[i]
# end
# eeg = expected_entropy_gain(bandit_posterior_covs, contexts, free_actions)
# if eeg > max_val
# max_val = eeg
# max_tup .= free_actions[remains]
# end
# end
# sss_actions = sequential_max_entropy(bandit_posterior_covs, contexts, actions)
# sss_actions = sss(bandit_posterior_covs, contexts, sss_actions, remains, 1000, 0)
#
# actions[remains] .= max_tup
# agree_count += (actions == sss_actions)
# base_count += 1
# actions = convert(Vector{Int64}, sss_actions)
#
# end
#
for m in 1:multi_count
#print(actions)
#print("\n")
action = actions[m]
context = contexts[m, :]
true_expected_rewards = true_bandit_param * context
EPREWARDS[t] += true_expected_rewards[action]
EPOPTREWARDS[t] += maximum(true_expected_rewards)
#obs = randn() * sqrt(obs_sd^2 + dot(context, bandit_posterior_covs[action,:,:],context)) + true_expected_reward
#obs = randn() * obs_sd + true_expected_rewards[action]
obss[m] = randn() * obs_sd + true_expected_rewards[action]
end
for act in 1:bandit_count
action_obss = obss[actions .== act]
action_contexts = contexts[actions .== act, :]
old_cov = bandit_posterior_covs[act, :, :]
#CovCon = old_cov * context ./ obs_sd
#bandit_posterior_covs[act, :, :] = old_cov - CovCon * CovCon' ./ (1 + dot(context, old_cov, context))
#bandit_posterior_covs[act, :, :] = ((bandit_posterior_covs[act,:,:]) + bandit_posterior_covs[act,:,:]')/2
#bandit_posterior_means[act, :] = (bandit_posterior_covs[act, :, :]) * (old_cov \ (bandit_posterior_means[act,:]) + context * obs / obs_sd^2)
#
bandit_posterior_covs[act, :, :] = inv(inv(old_cov) + action_contexts' * action_contexts ./ obs_sd^2)
bandit_posterior_covs[act, :, :] = ((bandit_posterior_covs[act,:,:]) + bandit_posterior_covs[act,:,:]')/2
bandit_posterior_means[act, :, :] = bandit_posterior_covs[act, :, :] * (old_cov \ bandit_posterior_means[act, :] + action_contexts' * action_obss ./ obs_sd^2)
end
#println("Ep: ", ep, " - ", t, " of ", T, " for coord eps greedy")
#flush(stdout)
end
return EPREWARDS, EPOPTREWARDS, agree_count, base_count
end
function approx_coord_greedy_ep_contextual_bandit_simulator(ep, T, rollout_length, n_episodes, n_rollouts, n_opt_rollouts, context_dim, context_mean,
context_sd, obs_sd, bandit_count, bandit_prior_mean, bandit_prior_sd, discount, epsilon, global_bandit_param)
bandit_posterior_means = zeros(bandit_count, context_dim)
bandit_posterior_covs = zeros(bandit_count, context_dim, context_dim)
bandit_param = copy(global_bandit_param)
true_bandit_param = copy(global_bandit_param)
EPREWARDS = zeros(T)
EPOPTREWARDS = zeros(T)
actions = convert(Vector{Int64}, zeros(multi_count))
contexts = zeros(multi_count, context_dim)
obss = zeros(multi_count)
for i in 1:bandit_count
bandit_posterior_means[i, :] = repeat([bandit_prior_mean], context_dim)
bandit_posterior_covs[i, :, :] = Diagonal(repeat([bandit_prior_sd^2], context_dim))
end
agree_count = 0
base_count = 0
for t in 1:T
remains = []
if decreasing
thresh = epsilon / t
else
thresh = epsilon
end
for m in 1:multi_count
context = generate_context(context_dim, context_mean, context_sd, context_constant)
contexts[m, :] = context
#true_expected_rewards = true_bandit_param * context
posterior_expected_rewards = bandit_posterior_means * context
if rand() > thresh
actions[m] = findmax(posterior_expected_rewards)[2]
else
push!(remains, m)
end
#true_expected_reward = true_expected_rewards[action]
#EPREWARDS[t] += true_expected_reward
#EPOPTREWARDS[t] += maximum(true_expected_rewards)
#obs = randn() * sqrt(obs_sd^2 + dot(context, bandit_posterior_covs[action,:,:],context)) + true_expected_reward
#obss[m] = randn() * obs_sd + true_expected_reward
end
#if length(remains) > 0
# post_covs_copy = copy(bandit_posterior_covs)
#
# remain_contexts = contexts[remains, :]
# for m in 1:multi_count
# if ! (m in remains)
# m_context = contexts[m, :]
# post_covs_copy[actions[m], :, :] = inv(inv(post_covs_copy[actions[m], :, :]) + m_context * m_context' / obs_sd^2)
# end
# end
#
# if length(remains) == 1
# actions[remains[1]] = findmax([logdet(I + post_covs_copy[i, :, :] * contexts[remains[1], :] * contexts[remains[1], :]' / obs_sd^2) for i = 1:bandit_count])[2]
# else
#
# print("\n")
# print(post_covs_copy)
# print("\n")
# print(remain_contexts ./ obs_sd)
# print("\n")
# flush(stdout)
# alloc = KL_optimal_allocations(post_covs_copy, remain_contexts ./ obs_sd)
# remain_actions = [findmax(alloc[i, :])[2] for i = 1:length(remains)]
# for i in 1:length(remains)
# actions[remains[i]] = remain_actions[i]
# end
# end
#end
if length(remains) > 0
#max_val = 0
#free_actions = copy(actions)
#max_tup = zeros(length(remains))
#for tup in CartesianIndices(ntuple(d->1:bandit_count, length(remains)))
# for i in 1:length(remains)
# free_actions[remains[i]] = tup[i]
# end
# eeg = expected_entropy_gain(bandit_posterior_covs, contexts, free_actions)
# if eeg > max_val
# max_val = eeg
# max_tup .= free_actions[remains]
# end
#end
sa_actions = sequential_max_entropy(bandit_posterior_covs, contexts, actions)
sa_actions = sa(bandit_posterior_covs, contexts, sa_actions, remains, 10000, 10, .99)
#actions = se_actions
#agree_count += (actions == sss_actions)
#base_count += 1
actions = convert(Vector{Int64}, sa_actions)
end
for m in 1:multi_count
#print(actions)
#print("\n")
action = actions[m]
context = contexts[m, :]
true_expected_rewards = true_bandit_param * context
EPREWARDS[t] += true_expected_rewards[action]
EPOPTREWARDS[t] += maximum(true_expected_rewards)
#obs = randn() * sqrt(obs_sd^2 + dot(context, bandit_posterior_covs[action,:,:],context)) + true_expected_reward
#obs = randn() * obs_sd + true_expected_rewards[action]
obss[m] = randn() * obs_sd + true_expected_rewards[action]
end
for act in 1:bandit_count
action_obss = obss[actions .== act]
action_contexts = contexts[actions .== act, :]
old_cov = bandit_posterior_covs[act, :, :]
#CovCon = old_cov * context ./ obs_sd
#bandit_posterior_covs[act, :, :] = old_cov - CovCon * CovCon' ./ (1 + dot(context, old_cov, context))
#bandit_posterior_covs[act, :, :] = ((bandit_posterior_covs[act,:,:]) + bandit_posterior_covs[act,:,:]')/2
#bandit_posterior_means[act, :] = (bandit_posterior_covs[act, :, :]) * (old_cov \ (bandit_posterior_means[act,:]) + context * obs / obs_sd^2)
#
bandit_posterior_covs[act, :, :] = inv(inv(old_cov) + action_contexts' * action_contexts ./ obs_sd^2)
bandit_posterior_covs[act, :, :] = ((bandit_posterior_covs[act,:,:]) + bandit_posterior_covs[act,:,:]')/2
bandit_posterior_means[act, :, :] = bandit_posterior_covs[act, :, :] * (old_cov \ bandit_posterior_means[act, :] + action_contexts' * action_obss ./ obs_sd^2)
end
#println("Ep: ", ep, " - ", t, " of ", T, " for coord eps greedy")
#flush(stdout)
end
return EPREWARDS, EPOPTREWARDS, agree_count, base_count
end
function coord_thompson_ep_contextual_bandit_simulator(ep, T, rollout_length, n_episodes, n_rollouts, n_opt_rollouts, context_dim, context_mean,
context_sd, obs_sd, bandit_count, bandit_prior_mean, bandit_prior_sd, discount, epsilon, global_bandit_param)
bandit_posterior_means = zeros(bandit_count, context_dim)
bandit_posterior_covs = zeros(bandit_count, context_dim, context_dim)
bandit_param = copy(global_bandit_param)
true_bandit_param = copy(global_bandit_param)
EPREWARDS = zeros(T)
EPOPTREWARDS = zeros(T)
actions = convert(Vector{Int64}, zeros(multi_count))
contexts = zeros(multi_count, context_dim)
obss = zeros(multi_count)
for i in 1:bandit_count
bandit_posterior_means[i, :] = repeat([bandit_prior_mean], context_dim)
bandit_posterior_covs[i, :, :] = Diagonal(repeat([bandit_prior_sd^2], context_dim))
end
for t in 1:T
for m in 1:multi_count
context = generate_context(context_dim, context_mean, context_sd, context_constant)
contexts[m, :] = context
end
# coordinated action selection
curr_actions = convert(Vector{Int64}, zeros(multi_count))
max_actions = convert(Vector{Int64}, zeros(multi_count))
max_gain = 0
for n in 1:thompson_count
thompson_samples = zeros(bandit_count, context_dim)
for bandit in 1:bandit_count
thompson_samples[bandit, :] = rand(MvNormal((bandit_posterior_means[bandit,:]), (bandit_posterior_covs[bandit,:,:])))
end
for m in 1:multi_count
curr_actions[m] = findmax(thompson_samples * contexts[m, :])[2]
end
curr_gain = expected_entropy_gain(bandit_posterior_covs, contexts, curr_actions)
print("\n")
print(curr_gain)
print("\n")
flush(stdout)
if curr_gain > max_gain
max_gain = curr_gain
max_actions = curr_actions
end
end
actions .= max_actions
for m in 1:multi_count
#print(actions)
#print("\n")
action = actions[m]
context = contexts[m, :]
true_expected_rewards = true_bandit_param * context
EPREWARDS[t] += true_expected_rewards[action]
EPOPTREWARDS[t] += maximum(true_expected_rewards)
#obs = randn() * sqrt(obs_sd^2 + dot(context, bandit_posterior_covs[action,:,:],context)) + true_expected_reward
#obs = randn() * obs_sd + true_expected_rewards[action]
obss[m] = randn() * obs_sd + true_expected_rewards[action]
end
for act in 1:bandit_count
action_obss = obss[actions .== act]
action_contexts = contexts[actions .== act, :]
old_cov = bandit_posterior_covs[act, :, :]
#CovCon = old_cov * context ./ obs_sd
#bandit_posterior_covs[act, :, :] = old_cov - CovCon * CovCon' ./ (1 + dot(context, old_cov, context))
#bandit_posterior_covs[act, :, :] = ((bandit_posterior_covs[act,:,:]) + bandit_posterior_covs[act,:,:]')/2
#bandit_posterior_means[act, :] = (bandit_posterior_covs[act, :, :]) * (old_cov \ (bandit_posterior_means[act,:]) + context * obs / obs_sd^2)
#
bandit_posterior_covs[act, :, :] = inv(inv(old_cov) + action_contexts' * action_contexts ./ obs_sd^2)
bandit_posterior_covs[act, :, :] = ((bandit_posterior_covs[act,:,:]) + bandit_posterior_covs[act,:,:]')/2
bandit_posterior_means[act, :, :] = bandit_posterior_covs[act, :, :] * (old_cov \ bandit_posterior_means[act, :] + action_contexts' * action_obss ./ obs_sd^2)
end
#println("Ep: ", ep, " - ", t, " of ", T, " for coord eps greedy")
#flush(stdout)
end
return EPREWARDS, EPOPTREWARDS
end
function coord_thompson_2_ep_contextual_bandit_simulator(ep, T, rollout_length, n_episodes, n_rollouts, n_opt_rollouts, context_dim, context_mean,
context_sd, obs_sd, bandit_count, bandit_prior_mean, bandit_prior_sd, discount, epsilon, global_bandit_param)
bandit_posterior_means = zeros(bandit_count, context_dim)
bandit_posterior_covs = zeros(bandit_count, context_dim, context_dim)
bandit_param = copy(global_bandit_param)
true_bandit_param = copy(global_bandit_param)
EPREWARDS = zeros(T)
EPOPTREWARDS = zeros(T)
actions = convert(Vector{Int64}, zeros(multi_count))
contexts = zeros(multi_count, context_dim)
obss = zeros(multi_count)
for i in 1:bandit_count
bandit_posterior_means[i, :] = repeat([bandit_prior_mean], context_dim)
bandit_posterior_covs[i, :, :] = Diagonal(repeat([bandit_prior_sd^2], context_dim))
end
for t in 1:T
for m in 1:multi_count
context = generate_context(context_dim, context_mean, context_sd, context_constant)
contexts[m, :] = context
end
# coordinated action selection
thompson_samples = zeros(bandit_count, context_dim)
for bandit in 1:bandit_count
thompson_samples[bandit, :] = rand(MvNormal((bandit_posterior_means[bandit,:]), (bandit_posterior_covs[bandit,:,:])))
end
actions = constrained_thompson_actions(100000, thompson_samples, bandit_posterior_means, bandit_posterior_covs, contexts)
actions = convert(Vector{Int64}, actions)
for m in 1:multi_count
#print(actions)
#print("\n")
action = actions[m]
context = contexts[m, :]
true_expected_rewards = true_bandit_param * context
EPREWARDS[t] += true_expected_rewards[action]
EPOPTREWARDS[t] += maximum(true_expected_rewards)
#obs = randn() * sqrt(obs_sd^2 + dot(context, bandit_posterior_covs[action,:,:],context)) + true_expected_reward
#obs = randn() * obs_sd + true_expected_rewards[action]
obss[m] = randn() * obs_sd + true_expected_rewards[action]
end
for act in 1:bandit_count
action_obss = obss[actions .== act]
action_contexts = contexts[actions .== act, :]
old_cov = bandit_posterior_covs[act, :, :]
#CovCon = old_cov * context ./ obs_sd
#bandit_posterior_covs[act, :, :] = old_cov - CovCon * CovCon' ./ (1 + dot(context, old_cov, context))
#bandit_posterior_covs[act, :, :] = ((bandit_posterior_covs[act,:,:]) + bandit_posterior_covs[act,:,:]')/2
#bandit_posterior_means[act, :] = (bandit_posterior_covs[act, :, :]) * (old_cov \ (bandit_posterior_means[act,:]) + context * obs / obs_sd^2)
#
bandit_posterior_covs[act, :, :] = inv(inv(old_cov) + action_contexts' * action_contexts ./ obs_sd^2)
bandit_posterior_covs[act, :, :] = ((bandit_posterior_covs[act,:,:]) + bandit_posterior_covs[act,:,:]')/2
bandit_posterior_means[act, :, :] = bandit_posterior_covs[act, :, :] * (old_cov \ bandit_posterior_means[act, :] + action_contexts' * action_obss ./ obs_sd^2)
end
#println("Ep: ", ep, " - ", t, " of ", T, " for coord eps greedy")
#flush(stdout)
end
return EPREWARDS, EPOPTREWARDS
end
function coord_thompson_3_ep_contextual_bandit_simulator(ep, T, rollout_length, n_episodes, n_rollouts, n_opt_rollouts, context_dim, context_mean,
context_sd, obs_sd, bandit_count, bandit_prior_mean, bandit_prior_sd, discount, epsilon, global_bandit_param)
bandit_posterior_means = zeros(bandit_count, context_dim)
bandit_posterior_covs = zeros(bandit_count, context_dim, context_dim)
bandit_param = copy(global_bandit_param)
true_bandit_param = copy(global_bandit_param)
EPREWARDS = zeros(T)
EPOPTREWARDS = zeros(T)
actions = convert(Vector{Int64}, zeros(multi_count))
contexts = zeros(multi_count, context_dim)
obss = zeros(multi_count)
for i in 1:bandit_count
bandit_posterior_means[i, :] = repeat([bandit_prior_mean], context_dim)
bandit_posterior_covs[i, :, :] = Diagonal(repeat([bandit_prior_sd^2], context_dim))
end
for t in 1:T
for m in 1:multi_count
context = generate_context(context_dim, context_mean, context_sd, context_constant)
contexts[m, :] = context
end
# coordinated action selection
thompson_samples = zeros(multi_count, bandit_count, context_dim)
for m in 1:multi_count
for bandit in 1:bandit_count
thompson_samples[m, bandit, :] = rand(MvNormal((bandit_posterior_means[bandit,:]), (bandit_posterior_covs[bandit,:,:])))
end
end
#actions = constrained_thompson_actions(100000, thompson_samples, bandit_posterior_means, bandit_posterior_covs, contexts)
#actions = convert(Vector{Int64}, actions)
for m in 1:multi_count
actions[m] = findmax(thompson_samples[m, :, :] * contexts[m, :])[2]
end
thompson_expected_rewards = [dot(bandit_posterior_means[actions[m], :], contexts[m, :]) for m = 1:multi_count]
restrictions = convert(Matrix{Int64}, zeros(multi_count, bandit_count))
for m in 1:multi_count
for b in 1: bandit_count
if dot(bandit_posterior_means[b, :], contexts[m, :]) >= thompson_expected_rewards[m]
restrictions[m, b] = 1
end
end
end
if sum(restrictions) > multi_count
#alloc = KL_optimal_allocations_with_restrictions(bandit_posterior_covs, contexts, restrictions)
#actions = [findmax(alloc[i, :])[2] for i = 1:multi_count]
actions = sa_restricted(bandit_posterior_covs, contexts, actions, restrictions, 50000, 10, .99)
actions = convert(Vector{Int64}, actions)
end
for m in 1:multi_count
#print(actions)
#print("\n")
action = actions[m]
context = contexts[m, :]
true_expected_rewards = true_bandit_param * context
EPREWARDS[t] += true_expected_rewards[action]
EPOPTREWARDS[t] += maximum(true_expected_rewards)
#obs = randn() * sqrt(obs_sd^2 + dot(context, bandit_posterior_covs[action,:,:],context)) + true_expected_reward
#obs = randn() * obs_sd + true_expected_rewards[action]
obss[m] = randn() * obs_sd + true_expected_rewards[action]
end
for act in 1:bandit_count
action_obss = obss[actions .== act]
action_contexts = contexts[actions .== act, :]
old_cov = bandit_posterior_covs[act, :, :]
#CovCon = old_cov * context ./ obs_sd
#bandit_posterior_covs[act, :, :] = old_cov - CovCon * CovCon' ./ (1 + dot(context, old_cov, context))
#bandit_posterior_covs[act, :, :] = ((bandit_posterior_covs[act,:,:]) + bandit_posterior_covs[act,:,:]')/2
#bandit_posterior_means[act, :] = (bandit_posterior_covs[act, :, :]) * (old_cov \ (bandit_posterior_means[act,:]) + context * obs / obs_sd^2)
#
bandit_posterior_covs[act, :, :] = inv(inv(old_cov) + action_contexts' * action_contexts ./ obs_sd^2)
bandit_posterior_covs[act, :, :] = ((bandit_posterior_covs[act,:,:]) + bandit_posterior_covs[act,:,:]')/2
bandit_posterior_means[act, :, :] = bandit_posterior_covs[act, :, :] * (old_cov \ bandit_posterior_means[act, :] + action_contexts' * action_obss ./ obs_sd^2)
end
#println("Ep: ", ep, " - ", t, " of ", T, " for coord eps greedy")
#flush(stdout)
end
return EPREWARDS, EPOPTREWARDS
end
function multi_contextual_bandit_simulator(action_function, T, rollout_length, n_episodes, n_rollouts, n_opt_rollouts, context_dim, context_mean,
context_sd, obs_sd, bandit_count, bandit_prior_mean, bandit_prior_sd, discount, epsilon)
REWARDS = zeros(T, n_episodes)
OPTREWARDS = zeros(T, n_episodes)
ep_count = 1
## USING BAYESIAN FORMULATION FOR FAIR COMPARISON
#global_bandit_param = [1 0; 0 1; 2 -1]