-
Notifications
You must be signed in to change notification settings - Fork 0
/
model_learning_core_yumi.py
1174 lines (1076 loc) · 46.8 KB
/
model_learning_core_yumi.py
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
import numpy as np
import scipy as sp
import matplotlib.pyplot as plt
import matplotlib.colors as mat_col
from mpl_toolkits.mplot3d import Axes3D
from matplotlib import cm
from matplotlib.ticker import MaxNLocator
from model_leraning_utils import get_N_HexCol
from model_leraning_utils import train_trans_models
# from model_leraning_utils import SVMmodePrediction
from model_leraning_utils import SVMmodePredictionGlobal as SVMmodePrediction
from model_leraning_utils import DPGMMCluster
from collections import Counter
#from sklearn.gaussian_process import GaussianProcessRegressor
from sklearn.gaussian_process.kernels import RBF, ConstantKernel as C, WhiteKernel as W
# from multidim_gp import MultidimGP
from multidim_gp import MdGpyGPwithNoiseEst as MultidimGP
# from multidim_gp import MdGpyGP as MultidimGP
# from multidim_gp import MdGpySparseGP as MultidimGP
from model_leraning_utils import UGP, SimplePolicy
from model_leraning_utils import dummySVM, traj_with_globalgp
# from YumiKinematics import YumiKinematics
from model_leraning_utils import logsum
from sklearn.preprocessing import StandardScaler
# from sklearn.model_selection import GridSearchCV
# from sklearn.svm import SVC
from copy import deepcopy
import operator
#import datetime
import time
from itertools import compress
import pickle
from blocks_sim import MassSlideWorld
from mjc_exp_policy import Policy, exp_params_rob, kin_params
# from mjc_exp_policy import SimplePolicy
import copy
# np.random.seed(1) # good value for clustering new yumi exp
# np.random.seed(5)
# np.random.seed(4) # good for big data wom10 without normalizing for clustering
# logfile = "./Results/yumi_peg_exp_new_preprocessed_data_train_big_data_wom10.p"
logfile = "./Results/Final/yumi_peg_exp_new_preprocessed_data_train_big_data_wom10_fixed.p"
# logfile = "./Results/Final/yumi_peg_exp_new_preprocessed_data_train_small_data_wom10_fixed.p"
# gp_result_file = "./Results/Final/results_yumi_gp_d40.p"
# moe_result_file = "./Results/Final/results_yumi_moe_d40.p"
# moe_result_file = "./Results/Final/results_wo_init_yumi_moe_d40.p"
# gp_result_file = "./Results/Final/results_yumi_gp_d15.p"
# moe_result_file = "./Results/Final/results_yumi_moe_d15.p"
gp_results = {}
gp_results['rmse'] = []
gp_results['nll'] = []
# gp_results = pickle.load(open(gp_result_file, "rb")) # global gp training done with 15 base policy trials
moe_results = {}
moe_results['rmse'] = []
moe_results['nll'] = []
# moe_results = pickle.load( open(moe_result_file, "rb" ) )
vbgmm_refine = False
global_gp = False
delta_model = True
fit_moe = True
load_gp = False
load_dpgmm = True
load_transition_gp = True
load_experts = True
load_svms = True
load_global_lt_pred = True
upgate_results = True
min_prob_grid = 0.001 # 1%
grid_size = 0.005
prob_min = 1e-3
mc_factor = 10
min_mc_particles = 3
# both pos and vel var was set to 6.25e-4 initially
p_noise_var = np.full(7, 1e-6)
v_noise_var = np.full(7, 2.5e-3)
# pol_per_facor = -0.02
jitter_var_tl = 1e-6
# jitter_var_tl = 0
num_tarj_samples = 50
exp_data = pickle.load( open(logfile, "rb" ) )
exp_params = exp_data['exp_params']
dP = exp_params['dP']
dV = exp_params['dV']
dU = exp_params['dU']
dX = dP+dV
dEP = 6
dEV = 6
dEX = 12
dF = 6
T = exp_params['T'] - 1
dt = exp_params['dt']
# n_train = exp_data['n_train']
# n_test = exp_data['n_test']-1 # TODO: remove -1 this is done to fix a bug in the logfile but already fixed in the code.
# n_test = exp_data['n_test']
# data set for joint space
# D15 dataset
# XUs_t_train = exp_data['XUs_t_train'][:15]
# Xs_t_train = exp_data['Xs_t_train'][:15]
# Xs_t1_train = exp_data['Xs_t1_train'][:15]
# EXs_t_train = exp_data['EXs_t_train'][:15]
# EXs_ee_t_train = exp_data['EXs_ee_t_train'][:15]
# D40 dataset
XUs_t_train = exp_data['XUs_t_train']
Xs_t_train = exp_data['Xs_t_train']
Xs_t1_train = exp_data['Xs_t1_train']
EXs_t_train = exp_data['EXs_t_train']
EXs_ee_t_train = exp_data['EXs_ee_t_train']
Us_t_test = exp_data['Us_t_test']
Xs_t_test = exp_data['Xs_t_test']
XU_t_train = XUs_t_train.reshape(-1, XUs_t_train.shape[-1])
n_train, _, _ = XUs_t_train.shape
Xrs_t_test = exp_data['Xrs_t_test']
X_t_train = Xs_t_train.reshape(-1, Xs_t_train.shape[-1])
X_t1_train = Xs_t1_train.reshape(-1, Xs_t1_train.shape[-1])
EX_t_train = EXs_t_train.reshape(-1, EXs_t_train.shape[-1])
dX_t_train = X_t1_train - X_t_train
EX_ee_t_train = EXs_ee_t_train.reshape(-1, EXs_ee_t_train.shape[-1])
# # unperturbed simple policy
# # Xrs_data = Xrs_t_train
# # Us_data = Us_t_train
# Xrs_data = Xrs_t_test
# Us_data = Us_t_test
# XUs_t_test = exp_data['XUs_t_test']
# XUs_test_data = XUs_t_test
# exp_params_data = deepcopy(exp_params_rob)
# m10 simple policy
# Xrs_data = Xrs_t_train
# Us_data = Us_t_train
Xrs_data = Xrs_t_test
Us_data = Us_t_test
XUs_t_test = exp_data['XUs_t_test']
n_test, _, _, = XUs_t_test.shape
XUs_test_data = XUs_t_test
Kp = exp_params_rob['Kp']
pol_per_facor = -0.1
exp_params_data = deepcopy(exp_params_rob)
exp_params_data['Kp'] = Kp + Kp * pol_per_facor
# filter vel signal for estimating vel noise variance
# plt.figure()
# tm = range(T)
# for i in range(n_train):
# x = Xs_t_train[i, :, dP + 6]
# x_fil = sp.ndimage.gaussian_filter1d(x, 4)
# plt.plot(tm, x, alpha=0.2)
# plt.plot(tm, x_fil)
# plt.show()
# filter vel signal for estimating pos noise variance
# plt.figure()
# tm = range(T)
# for i in range(n_train):
# x = Xs_t_train[i, :, 6]
# x_fil = sp.ndimage.gaussian_filter1d(x, 1.)
# plt.plot(tm, x, alpha=1.)
# plt.plot(tm, x_fil)
# plt.show()
# estimate vel noise variance for each joint
# v_res_s = np.zeros((n_train, T, dV))
# v_var_s = np.zeros((n_train, 7))
# for i in range(n_train):
# for j in range(7):
# v = Xs_t_train[i, :, dP + j]
# v_f = sp.ndimage.gaussian_filter1d(v, 4)
# v_res_s[i, :, j] = v - v_f
# v_var_s[i, j] = np.var(v_res_s[i, :, j])
# v_res = v_res_s.reshape(-1, v_res_s.shape[-1])
# v_var = np.var(v_res, axis=0)
# v_std = np.sqrt(v_var)
# v_var_mean = np.mean(v_var)
# # estimate pos noise variance for each joint
# p_res_s = np.zeros((n_train, T, dP))
# p_var_s = np.zeros((n_train, 7))
# for i in range(n_train):
# for j in range(7):
# p = Xs_t_train[i, :, j]
# p_f = sp.ndimage.gaussian_filter1d(p, 1)
# p_res_s[i, :, j] = p - p_f
# p_var_s[i, j] = np.var(p_res_s[i, :, j])
# p_res = p_res_s.reshape(-1, p_res_s.shape[-1])
# p_var = np.var(p_res, axis=0)
# p_std = np.sqrt(p_var)
# p_var_mean = np.mean(p_var)
ugp_params = {
'alpha': 1.,
'kappa': 2.,
'beta': 0.,
}
agent_hyperparams = {
'dt': 0.05,
'T': 50,
'smooth_noise': False,
'smooth_noise_var': 1.,
'smooth_noise_renormalize': False
}
gpr_params_global = {
'normalize': True,
'constrain_ls': True,
'ls_b_mul': (0.1, 10.),
'constrain_sig_var': True,
'sig_var_b_mul': (0.1, 10.),
'noise_var': np.concatenate([p_noise_var, v_noise_var]),
# 'noise_var': None,
'constrain_noise_var': True,
'noise_var_b_mul': (1e-2, 10.),
'fix_noise_var': False,
'restarts': 1,
}
# expl_noise = 3.
H = T # prediction horizon
if global_gp:
# global gp fit
if not load_gp:
mdgp_glob = MultidimGP(gpr_params_global, dX)
start_time = time.time()
if not delta_model:
mdgp_glob.fit(XU_t_train, X_t1_train)
else:
mdgp_glob.fit(XU_t_train, dX_t_train)
gp_training_time = time.time() - start_time
print 'Global GP fit time', gp_training_time
gp_results['gp_training_time'] = gp_training_time
exp_data['mdgp_glob'] = deepcopy(mdgp_glob)
# pickle.dump(exp_data, open(logfile, "wb"))
else:
if 'mdgp_glob' not in exp_data:
assert(False)
else:
mdgp_glob = exp_data['mdgp_glob']
global_gp_noise_level = []
global_gp_ls = []
global_gp_variance = []
for gp in mdgp_glob.gp_list:
global_gp_noise_level.append(gp.parameters[1].parameters[0])
global_gp_ls.append(gp.parameters[0].parameters[1])
global_gp_variance.append(gp.parameters[0].parameters[0])
print('Pos noise', global_gp_noise_level[:dP])
print('Vel noise', global_gp_noise_level[dP:])
print('Pos variance', global_gp_variance[dP:])
print('Vel variance', global_gp_variance[:dP])
print('Pos ls', global_gp_ls[dP:])
print('Vel ls', global_gp_ls[:dP])
if not load_global_lt_pred:
# global gp long-term prediction
# long-term prediction for MoE method
# original simple policy
pol = Policy(agent_hyperparams, exp_params_rob)
# pol1 = Policy(agent_hyperparams, exp_params_rob)
# sim_pol = SimplePolicy(Xrs_t_train, Us_t_train, exp_params_rob)
sim_pol = SimplePolicy(Xrs_data, Us_data, exp_params_data)
ugp_global_dyn = UGP(dX + dU, **ugp_params)
ugp_global_pol = UGP(dX, **ugp_params)
x_mu_t = exp_data['X0_mu']
# x_mu_t = exp_data['X0_mu'] + 0.5
x_var_t = np.diag(exp_data['X0_var'])
# x_var_t[0, 0] = 1e-6
# x_var_t[1,1] = 1e-6 # TODO: cholesky failing for zero v0 variance
Y_mu = np.zeros((2*(dX + dU) + 1, dX))
X_mu_pred = []
X_var_pred = []
U_mu_pred = []
U_mu_pred_x_avg = []
U2_mu_pred = []
U_mu_pred_sp = []
U_var_pred = []
U_mu_pred_avg = []
X_particles = []
start_time = time.time()
for t in range(H):
# standard case
# x_t = np.random.multivariate_normal(x_mu_t, x_var_t)
# u_mu_t, u_var_t, _, _, xu_cov = ugp_global_pol.get_posterior(pol, x_mu_t, x_var_t, t)
u_mu_t, u_var_t, _, _, xu_cov = ugp_global_pol.get_posterior(sim_pol, x_mu_t, x_var_t, t)
U_mu_pred.append(u_mu_t)
U_var_pred.append(u_var_t)
X_mu_pred.append(x_mu_t)
X_var_pred.append(x_var_t)
# X_particles.append(Y_mu)
xu_mu_t = np.append(x_mu_t, u_mu_t)
xu_var_t = np.block([[x_var_t, xu_cov],
[xu_cov.T, u_var_t]])
# simple policy evaluation
_, u_mu_t_sp = sim_pol.act(x_mu_t, t)
U_mu_pred_sp.append(u_mu_t_sp)
# fix u with mean u data
# u_mu_t_avg = XU_t_train_avg[t, dX:dX + dU]
# U_mu_pred_avg.append(u_mu_t_avg)
# xu_mu_t = np.append(x_mu_t, u_mu_t_avg)
# to test the policy with mean state data, the action should correspond to mean action data
# x_mu_t_avg = XU_t_train_avg[t, :dX]
############ TODO: remove after debugging
# x_mu_t_avg = np.array([-1.3048, -1.35466, 0.947929, 0.317889, 2.06793, 1.49044, -2.14021, 0.000531959, 0.00055548, -0.000337065, -7.55786e-05, 0.00385989, -0.000255539, -0.00792514])
############
# u_mu_t_x_avg, _ = pol1.predict(x_mu_t_avg.reshape(1, -1), t)
# u_mu_t_x_avg = pol1.act(x_mu_t_avg, None, t, noise=None)
# u_mu_t_x_avg = u_mu_t_x_avg.reshape(-1)
# U_mu_pred_x_avg.append(u_mu_t_x_avg)
if not delta_model:
x_mu_t, x_var_t, Y_mu, _, _ = ugp_global_dyn.get_posterior(mdgp_glob, xu_mu_t, xu_var_t)
else:
dx_mu_t, dx_var_t, dY_mu, _, xudx_covar = ugp_global_dyn.get_posterior(mdgp_glob, xu_mu_t, xu_var_t)
xdx_covar = xudx_covar[:dX, :]
x_mu_t = X_mu_pred[t] + dx_mu_t
x_var_t = X_var_pred[t] + dx_var_t + xdx_covar + xdx_covar.T
# Y_mu = X_particles[t] + dY_mu
x_var_t = x_var_t + np.eye(dX, dX) * jitter_var_tl # to prevent collapse of the Gaussian
gp_pred_time = time.time() - start_time
print 'Global GP prediction time for horizon', H, ':', gp_pred_time
gp_results['gp_pred_time'] = gp_pred_time
exp_data['global_lt_pred'] = {'X_mu_pred': X_mu_pred, 'X_var_pred': X_var_pred, 'U_mu_pred': U_mu_pred,'X_particles': X_particles}
# pickle.dump(exp_data, open(logfile, "wb"))
else:
if 'global_lt_pred' not in exp_data:
assert (False)
else:
X_mu_pred = exp_data['global_lt_pred']['X_mu_pred']
X_var_pred = exp_data['global_lt_pred']['X_var_pred']
U_mu_pred = exp_data['global_lt_pred']['U_mu_pred']
# X_particles = exp_data['global_lt_pred']['X_particles']
# compute long-term prediction score
XUs_t_test = exp_data['XUs_t_test']
assert(XUs_t_test.shape[0]==n_test)
X_test_log_ll = np.zeros((H, n_test))
for t in range(H): # one data point less than in XU_test
for i in range(n_test):
XU_test = XUs_t_test[i]
x_t = XU_test[t, :dX]
x_mu_t = X_mu_pred[t]
x_var_t = X_var_pred[t]
X_test_log_ll[t, i] = sp.stats.multivariate_normal.logpdf(x_t, x_mu_t, x_var_t)
tm = np.array(range(H)) * dt
plt.figure()
plt.title('Average NLL of test trajectories w.r.t time ')
plt.xlabel('Time, t')
plt.ylabel('NLL')
plt.plot(tm.reshape(H,1), np.mean(X_test_log_ll, axis=1).reshape(H, 1))
nll_mean = np.mean(X_test_log_ll.reshape(-1))
nll_std = np.std(X_test_log_ll.reshape(-1))
print 'UT NLL mean: ', nll_mean, 'UT NLL std: ', nll_std
# plot long-term prediction
X_mu_pred = np.array(X_mu_pred)
U_mu_pred = np.array(U_mu_pred)
# U_mu_pred_x_avg = np.array(U_mu_pred_x_avg)
# U2_mu_pred = np.array(U2_mu_pred)
# U_mu_pred_avg = np.array(U_mu_pred_avg)
# U_mu_pred_sp = np.array(U_mu_pred_sp)
P_mu_pred = X_mu_pred[:, :dP]
V_mu_pred = X_mu_pred[:, dP:]
P_sig_pred = np.zeros((H,dP))
V_sig_pred = np.zeros((H,dV))
U_sig_pred = np.zeros((H,dU))
P_sigma_points = np.zeros((H, 2 * (dX + dU) + 1, dP))
V_sigma_points = np.zeros((H, 2 * (dX + dU) + 1, dV))
for t in range(H):
P_sig_pred[t] = np.sqrt(np.diag(X_var_pred[t])[:dP])
V_sig_pred[t] = np.sqrt(np.diag(X_var_pred[t])[dP:])
U_sig_pred[t] = np.sqrt(np.diag(U_var_pred[t]))
# P_sigma_points[t, :, :] = X_particles[t][:, :dP]
# V_sigma_points[t, :, :] = X_particles[t][:, dP:]
# tm = np.array(range(H)) * dt
tm = np.array(range(H))
plt.figure()
plt.title('Long-term prediction with GP')
# jPos
for j in range(dP):
plt.subplot(3, 7, 1 + j)
# plt.xlabel('Time (s)')
# plt.ylabel('Joint Pos (rad)')
plt.title('j%dPos' % (j + 1))
plt.plot(tm, Xs_t_test[:, :H, j].T, alpha=0.2)
plt.plot(tm, P_mu_pred[:H, j], color='g', marker='s', markersize=2,)
plt.fill_between(tm, P_mu_pred[:H, j] - P_sig_pred[:H, j] * 1.96, P_mu_pred[:H, j] + P_sig_pred[:H, j] * 1.96, alpha=0.2, color='g')
# jVel
for j in range(dV):
plt.subplot(3, 7, 8 + j)
# plt.xlabel('Time (s)')
# plt.ylabel('Joint Vel (rad/s)')
plt.title('j%dVel' % (j + 1))
plt.plot(tm, Xs_t_test[:, :H, dP+j].T, alpha=0.2)
plt.plot(tm, V_mu_pred[:H, j], color='b', marker='s', markersize=2,)
plt.fill_between(tm, V_mu_pred[:H, j] - V_sig_pred[:H, j] * 1.96, V_mu_pred[:H, j] + V_sig_pred[:H, j] * 1.96,
alpha=0.2, color='b')
for j in range(dV):
plt.subplot(3, 7, 15 + j)
# plt.xlabel('Time (s)')
# plt.ylabel('Joint Trq (Nm)')
plt.title('j%dTrq' % (j + 1))
plt.plot(tm, XUs_t_test[:, :H, dX+j].T, alpha=0.2)
plt.plot(tm, U_mu_pred[:H, j], color='r', marker='s', markersize=2, label='mean pred')
plt.fill_between(tm, U_mu_pred[:H, j] - U_sig_pred[:H, j] * 1.96, U_mu_pred[:H, j] + U_sig_pred[:H, j] * 1.96,
alpha=0.2, color='r')
# plt.plot(tm, U_mu_pred_avg[:H, j], color='r', linestyle='--', label='mean data')
# plt.plot(tm, U_mu_pred_x_avg[:H, j], color='r', linestyle='-.', label='avg state based')
# plt.plot(tm, U_mu_pred_sp[:H, j], color='r', linestyle='-.', label='simple pol')
plt.legend()
plt.show(block=False)
# trajectory sampling approach (for consistency )
x_mu_0 = exp_data['X0_mu']
x_var_0 = np.diag(exp_data['X0_var'])
traj_with_globalgp_ = traj_with_globalgp(x_mu_0, x_var_0, mdgp_glob, sim_pol, dlt_mdl=delta_model)
traj_samples = traj_with_globalgp_.sample(num_tarj_samples, H)
gp_results['traj_samples'] = traj_samples
traj_mean = np.mean(traj_samples, axis=0)
traj_std = np.sqrt(np.var(traj_samples, axis=0))
traj_covar = np.zeros((H, dX, dX))
for t in range(H):
traj_covar[t] = np.cov(traj_samples[:, t, :], rowvar=False)
# tm = np.array(range(H)) * dt
tm = np.array(range(H))
plt.figure()
plt.title('Long-term prediction with GP (traj sampling)')
# jPos
for j in range(dP):
plt.subplot(3, 7, 1 + j)
# plt.xlabel('Time (s)')
# plt.ylabel('Joint Pos (rad)')
plt.title('j%dPos' % (j + 1))
plt.autoscale(True)
plt.plot(tm, Xs_t_train[:, :H, j].T, alpha=0.2, color='k')
# plt.autoscale(False)
plt.plot(tm, traj_samples[:, :H, j].T, alpha=0.2, color='g')
plt.plot(tm, traj_mean[:H, j], color='g')
plt.fill_between(tm, traj_mean[:H, j] - traj_std[:H, j] * 1.96, traj_mean[:H, j] + traj_std[:H, j] * 1.96,
alpha=0.2, color='g')
# jVel
for j in range(dV):
plt.subplot(3, 7, 8 + j)
# plt.xlabel('Time (s)')
# plt.ylabel('Joint Vel (rad/s)')
plt.title('j%dVel' % (j + 1))
plt.autoscale(True)
plt.plot(tm, Xs_t_train[:, :H, dP + j].T, alpha=0.2, color='k')
# plt.autoscale(False)
plt.plot(tm, traj_samples[:, :H, dP+j].T, alpha=0.2, color='b')
plt.plot(tm, traj_mean[:H, dP + j], color='b')
plt.fill_between(tm, traj_mean[:H, dP + j] - traj_std[:H, dP + j] * 1.96,
traj_mean[:H, dP + j] + traj_std[:H, dP + j] * 1.96,
alpha=0.2, color='b')
for j in range(dV):
plt.subplot(3, 7, 15 + j)
# plt.xlabel('Time (s)')
# plt.ylabel('Joint Trq (Nm)')
plt.title('j%dTrq' % (j + 1))
plt.autoscale(True)
plt.plot(tm, XUs_t_train[:, :H, dX + j].T, alpha=0.2, color='k')
plt.legend()
plt.show(block=False)
# loglikelihood score
XUs_t_test = exp_data['XUs_t_test']
assert (XUs_t_test.shape[0] == n_test)
X_test_log_ll = np.zeros((H, n_test))
X_test_SE = np.zeros((H, n_test))
for t in range(H): # one data point less than in XU_test
for i in range(n_test):
XU_test = XUs_t_test[i]
x_t = XU_test[t, :dX].reshape(-1)
x_mu_t = traj_mean[t].reshape(-1)
x_var_t = traj_covar[t]
# x_var_t = traj_covar[t] + np.eye(dX) * jitter_var_tl
x_var_t = np.diag(np.diag(x_var_t))
X_test_log_ll[t, i] = sp.stats.multivariate_normal.logpdf(x_t, x_mu_t, x_var_t)
X_test_SE[t, i] = np.dot((x_t - x_mu_t), (x_t - x_mu_t))
tm = np.array(range(H)) * dt
plt.figure()
plt.title('Average NLL of test trajectories w.r.t time ')
plt.xlabel('Time, t')
plt.ylabel('NLL')
plt.plot(tm.reshape(H,1), np.mean(X_test_log_ll, axis=1).reshape(H, 1))
tm = np.array(range(H)) * dt
plt.figure()
plt.title('Average RMSE of test trajectories w.r.t time ')
plt.xlabel('Time, t')
plt.ylabel('NLL')
plt.plot(tm.reshape(H, 1), np.mean(X_test_SE, axis=1).reshape(H, 1))
nll_mean = np.mean(X_test_log_ll.reshape(-1))
nll_std = np.std(X_test_log_ll.reshape(-1))
rmse = np.sqrt(np.mean(X_test_SE.reshape(-1)))
print('Yumi exp GP', 'NLL mean: ', nll_mean, 'NLL std: ', nll_std, 'RMSE:', rmse)
gp_results['rmse'].append(rmse)
gp_results['nll'].append((nll_mean, nll_std))
if upgate_results:
pickle.dump(gp_results, open(gp_result_file, "wb"))
if fit_moe:
if not load_dpgmm:
# clust_data = X_t_train
# clust_data = EX_t_train
clust_data = EX_ee_t_train
dof = clust_data.shape[1] + 2
dpgmm_params = {
'n_components': 20, # cluster size
'covariance_type': 'full',
'tol': 1e-6,
'n_init': 10,
'max_iter': 300,
'weight_concentration_prior_type': 'dirichlet_process',
'weight_concentration_prior':1e-1,
'mean_precision_prior':None,
'mean_prior': None,
'degrees_of_freedom_prior': dof,
'covariance_prior': None,
'warm_start': False,
'init_params': 'random',
'verbose': 1,
}
dpgmm_params_extra = {
'min_clust_size': 50,
# 'min_clust_size': 20, # for new yumi exp
'standardize': False,
'vbgmm_refine': False,
'min_size_filter': True,
'seg_filter': True,
'n_train': n_train,
}
##########Clustering notes for yumi exp###########
##################################################
dpgmm = DPGMMCluster(dpgmm_params, dpgmm_params_extra, clust_data)
start_time = time.time()
clustered_labels, labels, counts = dpgmm.cluster()
cluster_time = time.time() - start_time
moe_results['cluster_time'] = cluster_time
print('Clustering time:', cluster_time)
exp_data['dpgmm'] = deepcopy(dpgmm)
exp_data['clust_result'] = {'assign': clustered_labels, 'labels': labels, 'counts': counts}
# pickle.dump(exp_data, open(logfile, "wb"))
else:
if 'dpgmm' not in exp_data:
assert (False)
else:
dpgmm = exp_data['dpgmm']
clustered_labels = exp_data['clust_result']['assign']
labels = exp_data['clust_result']['labels']
counts = exp_data['clust_result']['counts']
clustered_labels_t = clustered_labels
clustered_labels_t_s = clustered_labels_t.reshape(n_train, -1)
clustered_labels_t1 = np.append(clustered_labels[1:], clustered_labels[-1])
clustered_labels_t1_s = clustered_labels_t1.reshape(n_train, -1)
K = len(labels)
colors = get_N_HexCol(K)
colors = np.asarray(colors) / 255.
fig = plt.figure()
ax = fig.gca()
ax.xaxis.set_major_locator(MaxNLocator(integer=True))
plt.bar(labels, counts, color=colors)
plt.title('DPGMM clustering')
plt.ylabel('Cluster sizes')
plt.xlabel('Cluster labels')
plt.show(block=False)
plt.savefig('dpgmm_yumi_cluster counts.pdf')
# plt.savefig('dpgmm_1d_dyn_cluster counts.png', format='png', dpi=1000)
# pi = dpgmm.dpgmm.weights_
# ax = plt.figure().gca()
# ax.xaxis.set_major_locator(MaxNLocator(integer=True))
# plt.bar(labels, pi[list(labels)], color=colors)
# plt.title('DPGMM clustering')
# plt.ylabel('Cluster weights')
# plt.xlabel('Cluster labels')
# plot clustered trajectory
col = np.zeros([EX_t_train.shape[0], 3])
i = 0
for label in labels:
col[(clustered_labels == label)] = colors[i]
i += 1
cols = col.reshape(n_train, T, -1)
label_col_dict = dict(zip(labels, colors))
fig = plt.figure()
ax = fig.add_subplot(1, 1, 1, projection='3d')
ax.scatter3D(EX_t_train[:,0], EX_t_train[:,1], EX_t_train[:,2], c=col)
ax.set_xlabel('X')
ax.set_ylabel('Y')
ax.set_zlabel('Z')
ax.set_title('DPGMM clustering')
plt.show(block=False)
if not load_transition_gp:
# transition GP
# gpr_params_trans = {
# 'noise_var': np.concatenate((p_noise_var, v_noise_var)),
# # 'noise_var': None,
# 'normalize': True,
# }
gpr_params_trans = {
'normalize': True,
'constrain_ls': False,
'ls_b_mul': (0.01, 10.),
'constrain_sig_var': False,
'sig_var_b_mul': (0.01, 10.),
# 'noise_var': np.concatenate([p_noise_var, v_noise_var]),
'noise_var': None,
'constrain_noise_var': False,
'noise_var_b_mul': (0.1, 10.),
'fix_noise_var': False,
'restarts': 1,
}
start_time = time.time()
trans_dicts = train_trans_models(gpr_params_trans, XUs_t_train, clustered_labels_t_s, dX, dU)
trans_gp_time = time.time() - start_time
moe_results['trans_gp_time'] = trans_gp_time
print ('Transition GP training time:', trans_gp_time)
exp_data['transition_gp'] = deepcopy(trans_dicts)
# pickle.dump(exp_data, open(logfile, "wb"))
else:
if 'transition_gp' not in exp_data:
assert(False)
else:
trans_dicts = exp_data['transition_gp']
if not load_experts:
# expert training
# gpr_params_experts = {
# 'noise_var': np.concatenate((p_noise_var, v_noise_var)),
# 'normalize': True,
# }
# gpr_params_experts = {
# 'normalize': False,
# 'constrain_ls': True,
# 'ls_b_mul': (0.01, 10.),
# 'constrain_sig_var': True,
# 'sig_var_b_mul': (0.01, 10.),
# # 'noise_var': np.concatenate([p_noise_var, v_noise_var]),
# 'noise_var': None,
# 'constrain_noise_var': True,
# 'noise_var_b_mul': (0.01, 10.),
# 'fix_noise_var': False,
# 'restarts': 1,
# }
gpr_params_experts = {
'normalize': True,
'constrain_ls': True,
'ls_b_mul': (0.1, 10.),
'constrain_sig_var': True,
'sig_var_b_mul': (0.1, 10.),
'noise_var': np.concatenate([p_noise_var, v_noise_var]),
# 'noise_var': None,
'constrain_noise_var': True,
'noise_var_b_mul': (1e-2, 10.),
'fix_noise_var': False,
'restarts': 1,
}
experts = {}
start_time = time.time()
for label in labels:
expert_idx = np.logical_and((clustered_labels_t == label), (clustered_labels_t1 == label))
x_train = XU_t_train[expert_idx]
y_train = X_t1_train[expert_idx]
if delta_model:
y_train = y_train - x_train[:, :dX]
mdgp = MultidimGP(gpr_params_experts, y_train.shape[1])
mdgp.fit(x_train, y_train)
experts[label] = deepcopy(mdgp)
del mdgp
expert_train_time = time.time() - start_time
moe_results['expert_train_time'] = expert_train_time
print 'Experts training time:', expert_train_time
exp_data['experts'] = deepcopy(experts)
# pickle.dump(exp_data, open(logfile, "wb"))
else:
if 'experts' not in exp_data:
assert(False)
else:
experts = exp_data['experts']
# plot experts data
# plt.figure()
# tm = np.tile(np.array(range(H)), n_train)
# for label in labels:
# expert_idx = np.logical_and((clustered_labels_t == label), (clustered_labels_t1 == label))
# x_train = XU_t_train[expert_idx]
# y_train = X_t1_train[expert_idx]
# tm_exp = tm[expert_idx]
# for j in range(7):
# plt.subplot(3, 7, 1+j)
# plt.scatter(tm_exp, x_train[:, j], s=2)
# plt.subplot(3, 7, 8 + j)
# plt.scatter(tm_exp, x_train[:, 7+j], s=2)
# plt.subplot(3, 7, 15 + j)
# plt.scatter(tm_exp, x_train[:, 14 + j], s=2)
# plt.show(block=False)
if not load_svms:
# gating network training
svm_grid_params = {
'param_grid': {"C": np.logspace(-12, 12, endpoint=True, num=11, base=2.),
"gamma": np.logspace(-12, 12, endpoint=True, num=11, base=2.)},
'scoring': 'accuracy',
# 'cv': 5,
'n_jobs':-1,
'iid': False,
'cv':3,
}
svm_params = {
'kernel': 'rbf',
'decision_function_shape': 'ovr',
'tol': 1e-06,
}
# svm for each mode
mode_prediction_data_t = XUs_t_train
mode_predictor = SVMmodePrediction(svm_grid_params, svm_params)
start_time = time.time()
mode_predictor.train(mode_prediction_data_t, clustered_labels_t_s, labels)
svm_train_time = time.time() - start_time
moe_results['svm_train_time'] = svm_train_time
print 'SVM training time:', svm_train_time
exp_data['mode_predictor'] = deepcopy(mode_predictor)
# pickle.dump(exp_data, open(logfile, "wb"))
else:
if 'mode_predictor' not in exp_data:
assert (False)
else:
mode_predictor = exp_data['mode_predictor']
# yumiKin = YumiKinematics(kin_params)
# long-term prediction for MoE method
# pol = Policy(agent_hyperparams, exp_params_rob)
pol = SimplePolicy(Xrs_data, Us_data, exp_params_data)
ugp_experts_dyn = UGP(dX + dU, **ugp_params)
ugp_experts_pol = UGP(dX, **ugp_params)
x_mu_t = exp_data['X0_mu']
x_var_t = np.diag(exp_data['X0_var'])
# x_var_t[0, 0] = 1e-6
# x_var_t[1, 1] = 1e-6 # TODO: cholesky failing for zero v0 variance
# ex_mu_t = exp_data['EX0_mu']
# mode0 = dpgmm.predict(x_mu_t.reshape(1, -1)) # TODO: vel multiplier?
mode0 = clustered_labels[0] # TODO: vel multiplier?
mode0 = np.asscalar(mode0)
mc_sample_size = (dX + dU) * mc_factor # TODO: put this param in some proper place
num_modes = len(labels)
modes = labels
X_mu_pred = []
X_var_pred = []
X_particles = []
sim_data_tree = [[[mode0, -1, x_mu_t, x_var_t, None, None, 1., pol]]]
start_time = time.time()
for t in range(H):
tracks = sim_data_tree[t]
for track in tracks:
md = track[0]
md_prev = track[1]
x_mu_t = track[2]
x_var_t = track[3]
p = track[6]
pi = track[7]
assert(pi is not None)
u_mu_t, u_var_t, _, _, xu_cov = ugp_experts_pol.get_posterior(pi, x_mu_t, x_var_t, t)
# u_mu_t = XU_t_train_avg[t, dX:]
# u_var_t = np.zeros((dU,dU))
xu_mu_t = np.append(x_mu_t, u_mu_t)
# xu_var_t = np.block([[x_var_t, np.zeros((dX,dU))],
# [np.zeros((dU,dX)), u_var_t]])
xu_var_t = np.block([[x_var_t, xu_cov],
[xu_cov.T, u_var_t]])
track[4] = u_mu_t
track[5] = u_var_t
xtut_s = np.random.multivariate_normal(xu_mu_t, xu_var_t, mc_sample_size)
assert (xtut_s.shape == (mc_sample_size, dX + dU))
mode_dst = mode_predictor.predict(xtut_s, md)
mode_counts = Counter(mode_dst).items()
# mode_counts_ = copy.deepcopy(mode_counts)
# for i in range(len(mode_counts_)):
# if mode_counts_[i][1] < min_mc_particles:
# del(mode_counts[i])
total_samples = 0
mode_prob = dict(zip(labels, [0] * len(labels)))
# mode_p = {}
for mod in mode_counts:
if (md == mod[0]) or ((md, mod[0]) in trans_dicts):
total_samples = total_samples + mod[1]
# for mod in mode_counts:
# if (md == mod[0]) or ((md, mod[0]) in trans_dicts):
# prob = float(mod[1]) / float(total_samples)
# mode_p[mod[0]] = prob
# mode_prob.update(mode_p)
# alternate mode_prob with state values also
mode_pred_dict = {}
for label in labels:
mode_pred_dict[label] = {'p': 0., 'mu': None, 'var': None}
for mod in mode_counts:
if (md == mod[0]) or ((md, mod[0]) in trans_dicts):
prob = float(mod[1]) / float(total_samples)
mode_pred_dict[mod[0]]['p'] = prob
XU_mode = np.array(list(compress(xtut_s, (mode_dst==mod[0]))))
mode_pred_dict[mod[0]]['mu'] = np.mean(XU_mode, axis=0)
if XU_mode.shape[0]==1:
# mode_pred_dict[mod[0]]['var'] = np.diag(np.full(dX+dU, 1e-6))
mode_pred_dict[mod[0]]['var'] = np.diag(np.concatenate((p_noise_var, v_noise_var, np.full(dU, 1e-6)))) # TODO: check this again and update in blocks
else:
mode_pred_dict[mod[0]]['var'] = np.cov(XU_mode, rowvar=False)
mode_pred_dict[mod[0]]['XU'] = XU_mode
if len(sim_data_tree) == t + 1:
sim_data_tree.append([]) # create the next (empty) time step
# for md_next, p_next in mode_prob.iteritems():
for mode_pred_key in mode_pred_dict:
mode_pred = mode_pred_dict[mode_pred_key]
md_next = mode_pred_key
p_next = mode_pred['p']
xu_mu_s_ = mode_pred['mu']
xu_var_s_ = mode_pred['var']
if p_next > prob_min:
# get the next state
if md_next == md:
md_ = md_prev
pi_next = pi
gp = experts[md]
if not delta_model:
x_mu_t_next_new, x_var_t_next_new, _, _, _ = ugp_experts_dyn.get_posterior(gp, xu_mu_t, xu_var_t)
else:
dx_mu_t_next_new, dx_var_t_next_new, _, _, xudx_covar = ugp_experts_dyn.get_posterior(gp,
xu_mu_t,
xu_var_t)
dx_var_t_next_new = dx_var_t_next_new + np.eye(dX,dX)*jitter_var_tl
xdx_covar = xudx_covar[:dX, :]
x_mu_t_next_new = x_mu_t + dx_mu_t_next_new
x_var_t_next_new = x_var_t + dx_var_t_next_new + xdx_covar + xdx_covar.T
# x_var_t_next_new = x_var_t + dx_var_t
# x_var_t_next_new = x_var_t
else:
md_ = md
gp_trans = trans_dicts[(md, md_next)]['mdgp']
xu_var_s_ = xu_var_s_ + np.eye(dX + dU) * jitter_var_tl
# x_mu_t_next_new, x_var_t_next_new, _, _, _ = ugp_experts_dyn.get_posterior(gp_trans, xu_mu_t,
# xu_var_t)
x_mu_t_next_new, x_var_t_next_new, _, _, _ = ugp_experts_dyn.get_posterior(gp_trans, xu_mu_s_, xu_var_s_)
# exp_params_ = deepcopy(exp_params_rob)
# exp_params_['x0'] = x_mu_t_next_new
# pi_next = Policy(agent_hyperparams, exp_params_)
pi_next = pi
assert (len(sim_data_tree) == t + 2)
tracks_next = sim_data_tree[t + 1]
if len(tracks_next)==0:
if p*p_next > prob_min:
sim_data_tree[t+1].append([md_next, md_, x_mu_t_next_new, x_var_t_next_new, 0., 0., p*p_next, pi_next])
else:
md_next_curr_list = [track_next[0] for track_next in tracks_next]
if md_next not in md_next_curr_list:
# md_next not already in the t+1 time step
if p * p_next > prob_min:
sim_data_tree[t + 1].append(
[md_next, md_, x_mu_t_next_new, x_var_t_next_new, 0., 0., p * p_next, pi_next])
else:
# md_next already in the t+1 time step
# if md == md_next:
# md_ = md_prev
# pi_next = pi
# else:
# md_ = md
# pi_next = None
md_next_curr_trans_list = [(track_next[1], track_next[0]) for track_next in tracks_next]
if (md_, md_next) not in md_next_curr_trans_list:
# the same transition track is not present
if p * p_next > prob_min:
sim_data_tree[t + 1].append(
[md_next, md_, x_mu_t_next_new, x_var_t_next_new, 0., 0., p * p_next, pi_next])
else:
it = 0
for track_next in tracks_next:
md_next_curr = track_next[0]
md_prev_curr = track_next[1]
x_mu_t_next_curr = track_next[2]
x_var_t_next_curr = track_next[3]
p_next_curr = track_next[6]
pi_next = track_next[7]
if md_next == md_next_curr:
next_trans = (md_, md_next)
curr_trans = (md_prev_curr, md_next_curr)
if curr_trans == next_trans:
p_next_new = p*p_next
tot_new_p = p_next_curr + p_next_new
w1 = p_next_curr / tot_new_p
w2 = p_next_new / tot_new_p
mu_next_comb = w1 * x_mu_t_next_curr + w2 * x_mu_t_next_new
var_next_comb = w1 * x_var_t_next_curr + w2 * x_var_t_next_new + \
w1 * np.outer(x_mu_t_next_curr,x_mu_t_next_curr) + \
w2 * np.outer(x_mu_t_next_new, x_mu_t_next_new) -\
np.outer(mu_next_comb,mu_next_comb)
p_next_comb = p_next_curr + p_next_new
if p_next_comb > prob_min:
sim_data_tree[t + 1][it] = \
[md_next, md_, mu_next_comb, var_next_comb, 0., 0., p_next_comb, pi_next]
it+=1
# probability check
prob_mode_tot = 0.
for track_ in sim_data_tree[t]:
prob_mode_tot += track_[6]
if (prob_mode_tot - 1.0) > 1e-4:
assert (False)
moe_pred_time = time.time() - start_time
moe_results['moe_pred_time'] = moe_pred_time
print 'Prediction time for MoE UGP with horizon', H, ':', moe_pred_time
moe_results['track_data'] = sim_data_tree
# plot each path (in mode) separately
# path is assumed to be a path arising out from a unique transtions
# different paths arising out of the same transition at different time is allowed in our model not here
tm = np.array(range(H)) * dt
path_dict = {}
col_mode = np.zeros((H, 3))
for i in range(H):
t = tm[i]
tracks = sim_data_tree[i]
prob = 0.
label_mode=np.zeros(H, dtype=int)
for track in tracks:
path = (track[0], track[1])
if path not in path_dict:
path_dict[path] = {'time':[] ,'X':[], 'X_var':[], 'X_std':[], 'U':[], 'U_var':[], 'U_std':[], 'prob':[], 'col':label_col_dict[path[0]]}
path_dict[path]['time'].append(t)
path_dict[path]['X'].append(track[2])
path_dict[path]['X_var'].append(track[3])
path_dict[path]['U'].append(track[4])
path_dict[path]['U_var'].append(track[5])
path_dict[path]['X_std'].append(np.sqrt(np.diag(track[3])))
path_dict[path]['U_std'].append(np.sqrt(np.diag(track[5])))
path_dict[path]['prob'].append(track[6])
p = track[6]
if p>prob:
prob = p
col_mode[i] = label_col_dict[path[0]]
label_mode[i] = path[0]
moe_results['path_data'] = path_dict
# plot for tree structure
# plot long term prediction results of UGP
plt.figure()
for path_key in path_dict:
# path_key = (10,-1)
path = path_dict[path_key]
time = np.array(path['time'])
pos = np.array(path['X'])[:,:dP]