-
Notifications
You must be signed in to change notification settings - Fork 4
/
orbitmap.py
782 lines (728 loc) · 32.3 KB
/
orbitmap.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
#!/usr/bin/env python
##############################################################
# Author: Yasuko Matsubara
# Email: yasuko@sanken.osaka-u.ac.jp
# URL: https://www.dm.sanken.osaka-u.ac.jp/~yasuko/
# Date: 2020-04-24
#------------------------------------------------------------#
# Copyright (C) 2020 Yasuko Matsubara & Yasushi Sakurai
# OrbitMap is freely available for non-commercial purposes
##############################################################
import numpy as np
import tool as tl
import nlds as nl
import om_trans as om_trans
import om_mdb as om_mdb
import om_viz as om_viz
import om_multiscale as om_multiscale
#------------------------------------------------#
# --- debug
COMMENT=tl.NO
DBG0=tl.NO
DBG1=tl.NO
# --- I/O
TSAVE=tl.NO #YES # save every time tick (default: NO)
SAVE_TRIALS=tl.NO #YES # plot trials
#------------------------------------------------#
# --- windows
LMIN=om_mdb.LMIN # minimum window length
LMAX=om_mdb.LMAX # maximum window length
LM_R=(1.0/1.0) # lm=ls*(1/1)--- minimum window length rate
LP_R=(1.0/5.0) # lp=ls*(1/5) --- sliding/reporting-windowrate
SMIN_R=LM_R #1.0 # minimum length of (multi-transitions)
# --- regimeID settings
REID_NULL=om_mdb.REID_NULL #regimeID (null: [-1])
REID_CPT=om_mdb.REID_CPT #regimeID (cpt: [-1])
# --- LMfit
W_RR=om_mdb.W_RR # weighted LMfit for RR
W_RE=om_mdb.W_RE # weighted LMfit for RE
W_FM=om_mdb.W_FM # weighted LMfit for FM
# seed, init rho
RHO_INIT=0.1 # if, want to estimate, RHO_INIT=-1
AVGR=1.0
AVG_TRIAL=10
# bottom-up
EARLYSTOP=tl.NO # default: NO
RHO_ITER_R_BU=1.2 #:1.5
MINTRIAL=5; MAXTRIAL=10 #6 #10
# check unfit patterns
WANT_AVOID_UNFIT=tl.YES
AVOIDUNFIT_R=1.5
# cut&paste
CUTNP="LAST" #LAST/MEAN (cut&paste: using last or mean value)
RHO_MIN_R=2 # minimum rho ratio
# print/output
IDT_E="\t"
IDT_F="\t\t"
# refinement
AP_MD=0; AP_MS=0 #1
#
OPTCNT=LMIN # _find_opt_cut, search for opt cut-points (OPTCNT times)
# multi-trans settings
MULTI_TRANS=tl.NO
# forecast
USE_Vb=tl.YES # hop-step-jump-forecast (use "Vb",Vp,Vc)
F_WANT_OPT=tl.NO # optimize multi-step-fit
FSTART_R=2 # start forecast at tc=ls*FSTART_R
# DPS
DPS=1 # dynamic point set ([1:lstep], default: 1)
#-------------------------------------------------#
# initialize env
def _set_env(lstep,mscale):
# set DPS
if(lstep>10):
global DPS; DPS=int(lstep/10)
#-------------------------------------------------#
###################################################
#-------------------------------------------------#
#-------------------------------------------------#
def run_est(Xorg, lstep, outdir, mscale_h=1):
tl.msg("run_est - start ...")
_set_env(lstep, mscale_h)
if(mscale_h==1): # single
run_est_single(Xorg, lstep, outdir)
else: # multi-scale
om_multiscale.run_est_mscale(Xorg, lstep, mscale_h, outdir)
tl.msg('run_est - end.')
#--------------------------------#
def run_scan(Xorg, lstep, modeldir, outdir, mscale_h=1):
tl.msg("run_scan - start ...")
_set_env(lstep, mscale_h)
if(mscale_h==1): # single
run_scan_single(Xorg, lstep, modeldir, outdir)
else: # multi-scale
om_multiscale.run_scan_mscale(Xorg, lstep, mscale_h, modeldir, outdir)
tl.msg('run_scan - end.')
#--------------------------------#
#-------------------------------------------------#
#-------------------------------------------------#
###################################################
#--------------------------------#
# run scan (streaming)
#--------------------------------#
def run_scan_single(Xorg, lstep, modeldir, outdir, wd_level=1):
tl.msg("(%d) run_scan_single: start ..."%wd_level)
mdb=tl.load_obj('%sMDB.obj'%(modeldir))
(mdb, Snaps) = _OrbitMap(Xorg, lstep, wd_level, mdb, outdir)
return (mdb, Snaps)
#--------------------------------#
#--------------------------------#
# run est (pre-processing)
#--------------------------------#
def run_est_single(Xorg, lstep, outdir, wd_level=1):
#--------------------------------#
want_refinement=tl.YES
#--------------------------------#
mdb_cb=om_mdb.MDB('%s'%outdir) # create modelDB
Snaps_cb=init_Snaps(Xorg, lstep, wd_level) # SnapShots
mdb_cb=mdb_cb.update_Xminmax(Xorg)
errF_cb=tl.INF #; Snaps_cb=[];
if(RHO_INIT==-1): rho_iter=_compute_rho_min(Xorg, lstep)
else: rho_iter=RHO_INIT
ERRs_full=[]; ERRs_cb_i=[]
outdir_tr="%s_trials/"%outdir
if(SAVE_TRIALS): tl.mkdir(outdir_tr) # create directory
tl.msg("(%d) run_est_single: start trial (find opt rho) ..."%wd_level)
for i_trial in range(0,MAXTRIAL):
#--------------------------------#
mdb_i=tl.dcopy(mdb_cb); Snaps_i=tl.dcopy(Snaps_cb)
#--------------------------------#
# set rho, cleansing (e.g., remove inappropriate regimes), etc.
mdb_i.update_rho(rho_iter)
(mdb_i, Snaps_i) = mdb_i.update_apoptosis(Snaps_i, AP_MD, AP_MS, want_refinement)
mdb_i = mdb_i.init_objects()
#--------------------------------#
# start OrbitMap
(mdb_i, Snaps_i) =_OrbitMap(Xorg, lstep, wd_level, mdb_i, "%sTrial_%d_"%(outdir_tr, i_trial), SAVE_TRIALS, SAVE_TRIALS) #tl.YES) #tl.NO)
#--------------------------------#
if(i_trial > 0 and Snaps_i['errF_half']<errF_cb): # update best-fit (if trial > 0)
errF_cb=Snaps_i['errF_half']; mdb_cb=tl.dcopy(mdb_i); Snaps_cb=tl.dcopy(Snaps_i)
#--------------------------------#
tl.msg("(mscale-wd: %d) Trial %d rho=%f, c=%d, errE: %f, errF(half):%f, errF(full):%f"%(wd_level, i_trial, rho_iter, mdb_i.get_c(), Snaps_i['errE'], Snaps_i['errF_half'], Snaps_i['errF']))
ERRs_full.append([i_trial, rho_iter, mdb_i.get_c(), Snaps_i['errE'], Snaps_i['errF_half'], Snaps_i['errF']])
tl.save_txt(ERRs_full, "%s_trials_results.txt"%(outdir))
#--------------------------------#
if(EARLYSTOP and Snaps_i['errF_half']>=errF_cb and i_trial>=MINTRIAL): break # if, no more better-fit, break
if(mdb_i.get_c()<=1 and i_trial > 0): break # if, # of regimes is less than one, then, stop
#--------------------------------#
#--------------------------------#
rho_iter*=RHO_ITER_R_BU # rho: try larger value
#--------------------------------#
# final-best-fit
(mdb_cb, Snaps_cb) = mdb_cb.update_apoptosis(Snaps_cb, AP_MD, AP_MS, want_refinement)
mdb_cb = mdb_cb.init_objects()
if(Snaps_cb['errF_half'] > Snaps_cb['errS']): mdb_cb.CC=tl.NO # if, not approp. (no causalchain), ignore model
(mdb_cb, Snaps_cb) =_OrbitMap(Xorg, lstep, wd_level, mdb_cb, "%s"%(outdir), tl.YES, tl.YES)
tl.msg("best est: c=%d, errE: %f, errF(half):%f, errF(full):%f"%( mdb_cb.get_c(), Snaps_cb['errE'], Snaps_cb['errF_half'], Snaps_cb['errF']))
return (mdb_cb, Snaps_cb)
#--------------------------------#
#--------------------------------#
# OrbitMap
#--------------------------------#
# (input)
# Xorg (original data stream)
# lstep (lstep-ahead-forecast)
# wd_level (window size (multi-scale))
# mdb (model parameter set)
#--------------------------------#
# (output)
# mdb (model parameter set)
# Snaps (snapshots, etc.,... )
#--------------------------------#
def _OrbitMap(Xorg, lstep, wd_level, mdb, outdir, save_fig=True, save_full=True):
#--------------------------------#
(n,d)=np.shape(Xorg)
Snaps=init_Snaps(Xorg, lstep, wd_level) # SnapShots
CGs=[] # for CGraph_snap_shots
n=Snaps['n']; d=Snaps['d']; pstep=Snaps['pstep']
PE_vc=_initP() # current best
PE_vp=_initP() # previous pattern
PE_vb=_initP() # previous-previous pattern
# init time-ticks
tc=pstep; PE_vc['tm_st']=0; PE_vc['tm_ed']=tc;
if(COMMENT): tl.msg("_OrbitMap X(%d,%d), lstep=%d, pstep=%d, wd_level=%d, c=%d, rho=%f, dps=%d"%(n,d, lstep, pstep, wd_level, mdb.get_c(), mdb.rho_RE, DPS))
#--------------------------------#
while(True):
#===================#
tic = tl.time.clock()
#===================#
if(tc>=n): break
if( np.isnan( Xorg[PE_vc['tm_st']:PE_vc['tm_ed'],:] ).sum()>0 ): PE_vc['tm_st']=tc; tc+=pstep; continue # if, nan-value, then ignore
if(DBG0): tl.comment("tc:%d"%(tc))
mdb=mdb.update_Xminmax(Xorg[(tc-pstep):tc,:]) # update Xmin, Xmax
#--------------------------------#
# 1. O-estimator
(PE_vb, PE_vp, PE_vc, mdb, Snaps) = _O_estimator(tc, Xorg, PE_vb, PE_vp, PE_vc, mdb, Snaps)
tic2 = tl.time.clock()
if(DBG1): tl.msg("time(estimate):%f"%(tl.time.clock()-tic))
#--------------------------------#
if(tc<=lstep*FSTART_R or tc+lstep>=n): tc+=pstep; continue
#--------------------------------#
# 2. O-generator
(Snaps) = _O_generator(tc, Xorg, PE_vb, PE_vp, PE_vc, mdb, Snaps)
#===================#
if(DBG1): tl.msg("time(forecast):%f"%(tl.time.clock()-tic2))
toc = tl.time.clock(); fittime= toc-tic;
if(DBG1): tl.msg("time:%f"%(fittime))
#if(save_full and TSAVE and outdir != ''): tl.save_mat(Snaps, "%sSnaps"%(outdir))
Snaps['T_full'][tc]=fittime
CGs.append(mdb.create_CGraph_rcds())
# continue
tc+=pstep
#===================#
#--------------------------------#
# final (SAVE)
#--------------------------------#
if(mdb.CC is tl.NO): Snaps['Ve_full']=Snaps['Xorg']
Snaps=compute_Snaps_Errs(Snaps)
want_refinement=tl.NO
(mdb, Snaps) = mdb.update_apoptosis(Snaps, 0, 0, want_refinement) # if there's any un-used regime, then delete (but, no-refinement)
# save results
if(save_full and outdir != ''):
tl.save_mat(Snaps, "%sSnaps"%(outdir))
tl.save_obj(mdb, "%sMDB"%(outdir))
tl.save_obj(CGs, "%sCGs"%(outdir))
if(save_fig and outdir!=''):
om_viz.saveResults_txt(Snaps, mdb, outdir)
om_viz.plotResultsE(Snaps, mdb, outdir)
om_viz.plotResultsF(Snaps, mdb, outdir)
om_viz.plotCG(mdb, outdir)
#--------------------------------#
# return mdb, Snaps
#--------------------------------#
return (mdb, Snaps)
def init_Snaps(Xorg, lstep, wd_level):
pstep=int(np.ceil(lstep*LP_R)) # set window (pstep)
(n,d)=np.shape(Xorg)
Snaps={'lstep':lstep, 'pstep':pstep, 'wd_level':wd_level,
'n':n, 'd':d, 'Xorg': Xorg, 'DPS': DPS,
'Xe':[], 'Ve':[], 'Se':[], 'Re':[], 'Te':[],
#'Se_full':np.nan*np.zeros((n,d+nl.KMAXPLS)), 'Sf_full':np.nan*np.zeros((n,d+nl.KMAXPLS)),
'Ve_full':np.nan*np.zeros((n,d)), 'Vf_full':np.nan*np.zeros((n,d)),
'Re_full':np.nan*np.zeros((n)), 'Rf_full':np.nan*np.zeros((n)),
'Es_full':[], 'Ex_full':[], 'Ee_full':[], 'Ef_full':[],
'errX':np.nan, 'errE':np.nan, 'errF':np.nan, 'errF_half':np.nan,
#'cpts':[],
'T_full': np.nan*np.zeros((n))}
return Snaps
def compute_Snaps_Errs(Snaps):
Xorg=Snaps['Xorg']; lstep=Snaps['lstep']; pstep=Snaps['pstep'];
n=Snaps['n']; d=Snaps['d']
Xsft=np.append(np.zeros((lstep+pstep,d)), Xorg[:-lstep-pstep,:],axis=0) # shifted-Xorg (ls+lp)
Snaps['Es_full']=tl.RMSE_each(Xsft, Xorg)
Snaps['Ex_full']=tl.RMSE_each(0*Xorg, Xorg)
Snaps['Ee_full']=tl.RMSE_each(Snaps['Ve_full'], Xorg)
Snaps['Ef_full']=tl.RMSE_each(Snaps['Vf_full'], Xorg)
Snaps['errS']=tl.mynanmean(Snaps['Es_full'][lstep:])
Snaps['errX']=tl.mynanmean(Snaps['Ex_full'][lstep:])
Snaps['errE']=tl.mynanmean(Snaps['Ee_full'][lstep:])
Snaps['errF']=tl.mynanmean(Snaps['Ef_full'][lstep:])
Snaps['errF_half']=tl.mynanmean(Snaps['Ef_full'][int(n*0.5):])
return Snaps
#--------------------------------#
# _initP()
#--------------------------------#
# PE_vb, PE_vp, PE_vc
#--------------------------------#
# md (model param set)
# cid (modelID)
# Vc (estimated events)
# siset (init set)
# tm_st (starting position)
# tm_ed (ending position)
# errV (error, ||Vc-Xc||)
#--------------------------------#
def _initP():
P={'md':[], 'cid':-1, 'Vc':[], 'siset':{}, 'tm_st':-1, 'tm_ed':-1, 'errV':tl.INF}
return P
def _isNullP(P):
if(P['md']==[]): return True
#if(P['cid']==-1): return True
else: return False
#--------------------------------#
# _RR, _RE, _RU
#--------------------------------#
# _RR: regime-reader
# _RE: regime-estimate
# _RU: regime-update
#--------------------------------#
# Xc (current window)
# mdb (full model set)
# PE_cr (current parameter set)
#--------------------------------#
# regime-reader
def _RR(Xc, mdb, PE_cr):
tm_st=PE_cr['tm_st']; tm_ed=PE_cr['tm_ed']
Si=PE_cr['siset']
(Vc, md_c, Si, cid)=mdb.search_md(Xc, Si, DPS)
if(md_c==[]):
if(DBG0): tl.msg("%s ...... _RR: mdb (null)"%(IDT_E))
else:
(Sc,Vc)=md_c.gen() # generate events
errV=tl.RMSE(Xc,Vc)
md_c.fn="seg:t%d-%d_e%.2f"%(tm_st, tm_ed, errV)
if(DBG0): tl.msg("%s ...... _RR: errV:%f (cid:%d)"%(IDT_E, errV, cid))
PE_cr={'md':md_c, 'cid':cid, 'Vc':Vc, 'Sc':Sc, 'siset':Si, 'tm_st':tm_st, 'tm_ed':tm_ed, 'errV':errV}
return PE_cr
# regime-estimate
def _RE(Xc, tm_st, tm_ed):
# estimate new model dynamics (md_c)
md_c=nl.NLDS(Xc, "t%d-%d"%(tm_st, tm_ed))
md_c=md_c.fit(W_RE) # lmfit
(Sc,Vc)=md_c.gen() # generate events
errV=tl.RMSE(Xc,Vc)
md_c.fn="seg:t%d-%d_e%.2f"%(tm_st, tm_ed, errV)
if(DBG0): tl.msg("%s ...... _RE: errV:%f (cid:-1)"%(IDT_E, errV))
PE_cr={'md':md_c, 'cid':-1, 'Vc':Vc, 'Sc':Sc, 'siset':{}, 'tm_st':tm_st, 'tm_ed':tm_ed, 'errV':errV}
return PE_cr
# regime-update
def _RU(Xc, PE_cr, fixORopt):
tm_st=PE_cr['tm_st']; tm_ed=PE_cr['tm_ed']
if(_isNullP(PE_cr)):
if(DBG0): tl.msg("%s ...... _RU: (null)"%(IDT_E))
else:
cid=PE_cr['cid'] # put cid
md_c=tl.dcopy(PE_cr['md']) # copy model
md_c.n=len(Xc); md_c.data=Xc # copy data
if(fixORopt=='OPT'):
md_c=md_c.fit_si(W_RR, DPS) # update init si (i.e., s(0)=si)
(Sc,Vc)=md_c.gen() # generate events
errV=tl.RMSE(Xc,Vc)
md_c.fn="seg:t%d-%d_e%.2f"%(tm_st, tm_ed, errV)
if(DBG0): tl.msg("%s ...... _RU: errV:%f (cid:%d)"%(IDT_E, errV, cid))
PE_cr={'md':md_c, 'cid':cid, 'Vc':Vc, 'Sc':Sc, 'siset':{}, 'tm_st':tm_st, 'tm_ed':tm_ed, 'errV':errV}
return PE_cr
# regime-trans-update
# PE_cr => {PE_vc : PE_vs}
def _RT(Xc, mdb, smin, PE_cr):
PE_vc=tl.dcopy(PE_cr); PE_vs=[]; lenc=len(Xc)
dictMS={'cid':PE_cr['cid'], 'si':tl.dcopy(PE_cr['md'].si), 'rho':mdb.rho_MS, 'smin':smin, 'errV':tl.INF}
(Ve, Se, Re)=om_trans.gen_forward(mdb, dictMS, lenc)
errV=tl.RMSE(Xc, Ve[0:lenc])
#tl.eprint("_RT: before:%f -> after:%f"%(PE_cr['errV'], errV))
#-------------------------#
#--- find first cpt ------#
cpt=-1
for t in range(0,lenc):
if(Re[t]==-1): cpt = t; break
if(cpt==-1 or PE_cr['errV'] <= errV): return (PE_vc, PE_vs) # if it cannot find better-trans, just ignore
#-------------------------#
# update params {PE_vc : PE_vs}
#tl.eprint("tm_st:%d, tm_ed:%d, cpt:%d"%(PE_vc["tm_st"], PE_vc["tm_ed"], cpt))
PE_vc['tm_ed']=PE_vc['tm_st']+cpt
PE_vs={'md':mdb.MD[Re[cpt+1]]['medoid'], 'cid':Re[cpt+1], 'Vc':Ve[cpt+1:,], 'Sc':Se[cpt+1:,], 'siset':{}, 'tm_st':PE_vc['tm_st']+(cpt+1), 'tm_ed':PE_cr['tm_ed'], 'errV':errV}
#tl.eprint("tm_st:tm_ed: %d %d %d %d"%(PE_vc['tm_st'], PE_vc['tm_ed'], PE_vs['tm_st'], PE_vs['tm_ed']))
# compute errorV
PE_vc['errV']=tl.RMSE(Xc[0:cpt], Ve[0:cpt])
PE_vs['errV']=tl.RMSE(Xc[(cpt+1):lenc], Ve[(cpt+1):lenc])
#tl.eprint("err_vc:%f - err_vs:%f (rho:%f)"%(PE_vc['errV'], PE_vs['errV'], mdb.rho_RE))
return (PE_vc, PE_vs)
#--------------------------------#
# O-estimator
#--------------------------------#
# tc (current time point)
# Xorg (original data stream)
# PE_vb (previous-previous pattern)
# PE_vp (previous pattern)
# PE_vc (current pattern)
# mdb (model parameter set)
# Snaps (snapshots, etc.,... )
#--------------------------------#
def _O_estimator(tc, Xorg, PE_vb, PE_vp, PE_vc, mdb, Snaps):
if(mdb.CC is tl.NO): return (PE_vb, PE_vp, PE_vc, mdb, Snaps) # if, no causal-chain, then ignore
lstep=Snaps['lstep']
pstep=Snaps['pstep']
wd_level=Snaps['wd_level']
lmin=min(LMAX, max(LMIN,int(np.ceil(lstep* LM_R))))
smin=min(LMAX, max(LMIN,int(np.ceil(lstep*SMIN_R))))
lmax=LMAX
PE_vc_prev=tl.dcopy(PE_vc) # current-best-previous
PE_vc['tm_ed']=tc # currrent time tick
if(PE_vc['tm_ed']-PE_vc['tm_st']<lmin): return (PE_vb, PE_vp, PE_vc, mdb, Snaps)
# create current window Xc
Xc=Xorg[PE_vc['tm_st']:PE_vc['tm_ed'],:]
if(COMMENT): tl.msg("%s |--- O_estimator: (wd=%d) t=%d:%d:%d:%d (rho:%f)-----------------|"%(IDT_E, wd_level, PE_vp['tm_st'], PE_vp['tm_ed'], PE_vc['tm_st'], PE_vc['tm_ed'],mdb.rho_RE))
#-----------------------------------------------#
# (I) find good-fit regime
#-----------------------------------------------#
#-----------------------------#
# (A) regime update
#-----------------------------#
PE_RU = _RU(Xc, PE_vc, 'FIX') # fixed, simply extend seg
PE_vc = PE_RU
if(PE_vc['errV'] > mdb.rho_RE): # if fixed is not good-enough
PE_RU = _RU(Xc, PE_vc, 'OPT') # update init value
if(PE_RU['errV'] <= PE_vc['errV']):
PE_vc=PE_RU
#-----------------------------#
# (B) regime-trans update (if, current is known but not good-fit)
#-----------------------------#
PE_vs=[]
if(PE_vc['cid'] !=-1 and PE_vc['errV'] > mdb.rho_RE):
(PE_RT_vc, PE_RT_vs) = _RT(Xc, mdb, smin, PE_vc)
if( PE_RT_vc['errV']< mdb.rho_RE and PE_RT_vs['errV']<mdb.rho_RE ):
PE_vc=PE_RT_vc; PE_vs=PE_RT_vs
#-----------------------------#
# (C) regime reader, if update is not good-enough
#-----------------------------#
if(PE_vc['cid'] ==-1 or (PE_vc['errV'] > mdb.rho_RE) ):
PE_RR = _RR(Xc, mdb, PE_vc)
if(PE_RR['errV'] <= PE_vc['errV']):
PE_vc=PE_RR
#-----------------------------#
# (D) regime creation (estimator), only if the previous-fit is unknown (i.e., splitted) and current regime is not in MDB
#-----------------------------#
if(PE_vc_prev['cid']==-1 and PE_vc['errV'] > mdb.rho_RE):
PE_RE = _RE(Xc, PE_vc['tm_st'], PE_vc['tm_ed'])
if(PE_RE['errV'] <= PE_vc['errV']):
PE_vc=PE_RE
#-----------------------------------------------#
# (II) split regime, if there is a new cut-point
#-----------------------------------------------#
#-----------------------------#
lenc=PE_vc['tm_ed']-PE_vc['tm_st']
#-----------------------------#
# split segment now, if required
#-----------------------------#
if( (PE_vc['errV'] > mdb.rho_RE and lenc>lmin) or # if current best is not-fit seg
(PE_vs != [] and lenc>lmin) or # if any transisions
(lenc > lmax) or # if too long segment
(PE_vc['tm_ed']+pstep >= Snaps['n']) ): # if it's close to the end point
#-----------------------------#
PE_vc=PE_vc_prev # use current-best-previous
#if(DBG0): tl.eprint("%s |-------- split segment: t=%d:%d, errV=%f (rho=%f)"%(IDT_E, PE_vc['tm_st'], PE_vc['tm_ed'], PE_vc['errV'], mdb.rho_RE))
#----------------------------------------------------#
# (d-a) if, good-enough-fit, then insert it into mdb
if(PE_vc['errV'] <= mdb.rho_RE*RHO_MIN_R):
# (d-a-1) insert Xc into mdb
(mdb, PE_vc) = mdb.update_md_insert(PE_vc)
# (d-a-2) insert MS(PE_vp, PE_vc) into mdb, if, we have PE_vp, PE_vc
if( (not _isNullP(PE_vp)) and (not _isNullP(PE_vc)) ):
# (d-a-2-i) find opt-cut-point (+-pstep, argmin ||Vp-Xp||+||Vc-Xc||)
(PE_vp, PE_vc) = _find_opt_cut(Xorg, PE_vp, PE_vc, lstep, lmin) #pstep)
# (d-a-2-ii) insert new regime-shift-dynamics into mdb
mdb = mdb.update_ms_insert(PE_vp, PE_vc) # add (from to)
# (d-a-2-iii) update estimated events
# previous set
Snaps['Ve_full'][PE_vp['tm_st']:PE_vp['tm_ed'],:]=PE_vp['Vc']
#Snaps['Se_full'][PE_vp['tm_st']:PE_vp['tm_ed'],:PE_vp['md'].k]=PE_vp['Sc']
Snaps['Re_full'][PE_vp['tm_st']:PE_vp['tm_ed']]=PE_vp['cid']
Snaps['Re_full'][PE_vp['tm_st']]=REID_CPT # cut-point (regime-switch)
# current set
Snaps['Ve_full'][PE_vc['tm_st']:PE_vc['tm_ed'],:]=PE_vc['Vc']
#Snaps['Se_full'][PE_vc['tm_st']:PE_vc['tm_ed'],:PE_vc['md'].k]=PE_vc['Sc']
Snaps['Re_full'][PE_vc['tm_st']:PE_vc['tm_ed']]=PE_vc['cid']
Snaps['Re_full'][PE_vc['tm_st']]=REID_CPT # cut-point (regime-switch)
#if(DBG0): tl.eprint("%s |-------- split segment(opt-cut): t=%d:%d, errV=%f (rho=%f)"%(IDT_E, PE_vc['tm_st'], PE_vc['tm_ed'], PE_vc['errV'], mdb.rho_RE))
#----------------------------------------------------#
# (d-b) reset params
PE_vb=PE_vp; PE_vp=PE_vc; PE_vc=_initP();
if(PE_vs!=[]): PE_vc=PE_vs
PE_vc['tm_st']=PE_vp['tm_ed'];
#-----------------------------#
# return current results
return (PE_vb, PE_vp, PE_vc, mdb, Snaps)
#--------------------------------#
# _find_opt_cut
#--------------------------------#
# (find optimum cut-point)
# Vp[tm_st:tm_ed] vs. Vc[tm_st:tm_ed]
#--------------------------------#
def _find_opt_cut(Xorg, PE_vp, PE_vc, lenw, lmin):
tp_st=PE_vp['tm_st']; tc_ed=PE_vc['tm_ed']; # start/end points
tcut=PE_vp['tm_ed'] # current cut-point
errs=[]; locs=[]
# find best-cut-position
t_st=max(tp_st+lmin, tcut-int(lenw/2)); t_ed=min(tc_ed-lmin, tcut+int(lenw/2))
PE_vps=[]; PE_vcs=[]
if(t_st==t_ed): return (PE_vp, PE_vc)
for loc in range(t_st, t_ed, int(max(1,(t_ed-t_st)/OPTCNT))):
PE_vpi=tl.dcopy(PE_vp); PE_vci=tl.dcopy(PE_vc)
PE_vpi['tm_ed']=loc; PE_vci['tm_st']=loc;
PE_vpi = _RU(Xorg[tp_st:loc,:], PE_vpi, 'OPT')
PE_vci = _RU(Xorg[loc:tc_ed,:], PE_vci, 'OPT')
err = PE_vpi['errV']+PE_vci['errV']
PE_vps.append(PE_vpi); PE_vcs.append(PE_vci)
locs.append(loc); errs.append(err)
if(DBG0): tl.msg("%s ...... _find_opt_cut l:%d (%f + %f = %f)"%(IDT_E, loc, PE_vpi['errV'], PE_vci['errV'], err))
ibest=np.argmin(errs); locbest=locs[ibest]
if(DBG0): tl.msg("%s ...... _find_opt_cut l:%d (%f + %f = %f)"%(IDT_E, locbest, PE_vps[ibest]['errV'], PE_vcs[ibest]['errV'], err))
PE_vp=PE_vps[ibest]; PE_vc=PE_vcs[ibest]
return (PE_vp, PE_vc)
#--------------------------------#
# _FS, _FM, _FP
#--------------------------------#
# _FS: forecast using single regime
# _FM: forecast using multi-step regimes
# _FP: forecast using current/last value Xc[tm_ed]
#--------------------------------#
# mdb (full model set)
# Xc (current window)
# lene (forecast length)
# lstep (lstep-ahead-forecast)
# PE (current parameter set)
# fixOropt (using fix/opt params)
#--------------------------------#
#
#------------------------------------------------#
# forecast single
def _FS(mdb, Xc, lene, PE, fixORopt):
md_c=PE['md'] #; si=md_c.si
(Se,Ve)=md_c.gen(lene) #(Se,Ve)=md_c.forward(si, lene)
Re=REID_NULL*np.ones(lene)
errV=tl.RMSE(Xc, Ve[0:len(Xc)])
if(_isUnfit(Ve, mdb)):
Ve[:]=np.nan; Se[:]=np.nan; Re[:]=REID_NULL; errV=tl.INF
if(DBG0): tl.msg("%s ++++++ _FS(%s): errV:%f (cid:%d) "%(IDT_F, fixORopt, errV, REID_NULL))
PF={'Ve':Ve, 'Se':Se, 'Re':Re, 'errV':errV}
return PF
#------------------------------------------------#
# forecast multi-step
def _FM(mdb, Xc, lene, smin, PE, fixORopt):
# dictMS (see, om_trans.py)
dictMS={'cid':PE['cid'], 'si':tl.dcopy(PE['md'].si), 'rho':mdb.rho_MS, 'smin':smin, 'errV':tl.INF}
if(fixORopt=='OPT'):
dictMS = om_trans.fit_forward(Xc, mdb, dictMS, W_FM) #, DPS)
(Ve, Se, Re)=om_trans.gen_forward_multi_trans(Xc, mdb, dictMS, lene, MULTI_TRANS)
#(Ve, Se, Re)=om_trans.gen_forward(mdb, dictMS, lene)
errV=tl.RMSE(Xc, Ve[0:len(Xc)])
if(_isUnfit(Ve, mdb)):
Ve[:]=np.nan; Se[:]=np.nan; Re[:]=REID_NULL; errV=tl.INF
if(DBG0): tl.msg("%s ++++++ _FM(%s): errV:%f (cid:%d) "%(IDT_F, fixORopt, errV, dictMS['cid']))
PF={'Ve':Ve, 'Se':Se, 'Re':Re, 'errV':errV}
return PF
#------------------------------------------------#
# forecast cut&paste i.e., use current/latest event Xc[tm_ed]
def _FP(Xc, lene):
(lenc,d)=np.shape(Xc)
Ve=np.zeros((lene,d))
if(CUTNP=="MEAN"): Ve[0:lene,:]=tl.mynanmean(Xc, axis=0)
if(CUTNP=="LAST"): Ve[0:lene,:]=Xc[-1,:]
Se=-1*np.ones((lene,1))
Re=REID_NULL*np.ones(lene)
errV=tl.RMSE(Xc, Ve[0:len(Xc)]) #=tl.INF #
if(DBG0): tl.msg("%s ++++++ _FP(FIX): errV:inf (cid:%d) "%(IDT_F, REID_NULL))
PF={'Ve':Ve, 'Se':Se, 'Re':Re, 'errV':errV, 'MSs':[]}
return PF
#------------------------------------------------#
#join a sequence of arrays along an existing axis.
def _con(A,B):
return np.concatenate((A,B), axis=0)
# insert null values of length lene
def _padding(PF, lene):
(lenc,d)=np.shape(PF['Ve'])
(lenc,k)=np.shape(PF['Se'])
PF['Ve']=_con( np.nan*np.zeros((lene,d)), PF['Ve'] )
PF['Se']=_con( np.nan*np.zeros((lene,k)), PF['Se'] )
PF['Re']=_con( np.nan*np.zeros((lene)), PF['Re'] )
return PF
#------------------------------------------------#
#------------------------------------------------#
# find unfit pattern (optional)
def _isUnfit(Ve, mdb):
if(not WANT_AVOID_UNFIT): return False
# check XminXmax (if, estimation (Ve) is too large or too small, ignore it
if( mdb.Xmin*AVOIDUNFIT_R > min(Ve.flatten())
or mdb.Xmax*AVOIDUNFIT_R < max(Ve.flatten()) ) : return True
return False
#------------------------------------------------#
#--------------------------------#
# O-generator
#--------------------------------#
# tc (current time point)
# Xorg (original data stream)
# PE_vb (previous-previous pattern)
# PE_vp (previous pattern)
# PE_vc (current pattern)
# mdb (model parameter set)
# Snaps (snapshots, etc.,... )
#--------------------------------#
#========================================================#
#
# ----------- PAST -------------|------ FUTURE ----------
# tc
# tb_st tb_ed |
# | tp_st tp_ed |
# | | tc_st tc_ed
# | | | | tf_st tf_ed
# | | | | | |
# |---Vb----|---Vp----|---Vc----|---Vs----|--Vf---|
#
#========================================================#
def _O_generator(tc, Xorg, PE_vb, PE_vp, PE_vc, mdb, Snaps):
#-----------------------------#
# setting
#-----------------------------#
# (1) set timeticks
lstep=Snaps['lstep']; pstep=Snaps['pstep']; wd_level=Snaps['wd_level']
smin=min(LMAX, max(LMIN,int(np.ceil(lstep*SMIN_R))))
tb_st=PE_vb['tm_st']; tb_ed=PE_vb['tm_ed']
tp_st=PE_vp['tm_st']; tp_ed=PE_vp['tm_ed']
tc_st=PE_vc['tm_st']; tc_ed=PE_vc['tm_ed']
# (2) if, previous P is null
if(_isNullP(PE_vp)):
tb_st=PE_vc['tm_st']; tb_ed=PE_vc['tm_st']
tp_st=PE_vc['tm_st']; tp_ed=PE_vc['tm_st']
elif(_isNullP(PE_vb)):
tb_st=PE_vp['tm_st']; tb_ed=PE_vp['tm_st']
if(_isNullP(PE_vc)):
tc_st=tp_ed; tc_ed=tp_ed
# (3) current time tick
tc_ed=tc
# (4) forecast window
tf_st=tc_ed+lstep; tf_ed=min(tf_st+pstep,Snaps['n'])
# (5) set original seq
Xb=Xorg[tb_st:tb_ed,:] # known
Xbp=Xorg[tb_st:tp_ed,:] # known
Xbpc=Xorg[tb_st:tc_ed,:] # known
Xpc=Xorg[tp_st:tc_ed,:] # known
Xp=Xorg[tp_st:tp_ed,:] # known
Xc=Xorg[tc_st:tc_ed,:] # known
Xbpcsf=Xorg[tb_st:tf_ed,:] # unknown
Xpcsf=Xorg[tp_st:tf_ed,:] # unknown
Xcsf=Xorg[tc_st:tf_ed,:] # unknown
Xe=Xbpcsf
# (6) seq-length
lenb=len(Xb);
lenbp=len(Xbp); lenpcsf=len(Xpcsf); lencsf=len(Xcsf)
lenpc=len(Xpc); lenp=len(Xp); lenc=len(Xc); lenbpcsf=len(Xbpcsf); #lenec=lene-lenp
lenbpc=len(Xbpc)
if(COMMENT): tl.msg("%s |>>> O_generator:(wd=%d) t=%d:%d|%d:%d|%d:%d|%d:%d >>>>>>>>>>>|"%(IDT_F, wd_level, tb_st, tb_ed, tp_st, tp_ed, tc_st, tc_ed, tf_st, tf_ed))
#-----------------------------#
# forecast
#-----------------------------#
# (A) cut-n-paste
PF_cb=_FP(Xpc, lenpcsf)
PF_cb=_padding(PF_cb, lenb)
if(mdb.CC is tl.YES):
#-------------------------------------------------#
# (B-Vb) if pre-previous (Vb) is given, and known
if(USE_Vb): # if you want hop-step-jump-fit
if(PE_vb['cid']!=-1):
# (B-Vb-1) use current params
PF_fix=_FM(mdb, Xbpc, lenbpcsf, smin, PE_vb, 'FIX')
if(PF_fix['errV']<PF_cb['errV']):
PF_cb=PF_fix
# (B-Vb-2) update params (if current params is not good enough)
if(F_WANT_OPT and PF_cb['errV']>mdb.rho_RF):
PF_upd=_FM(mdb, Xbpc, lenbpcsf, smin, PE_vb, 'OPT')
if(PF_upd['errV'] < PF_cb['errV']):
PF_cb=PF_upd
#-------------------------------------------------#
# (B-Vp) if previous (Vp) is given, and known
if(PE_vp['cid']!=-1):
# (B-Vp-1) use current params
PF_fix=_FM(mdb, Xpc, lenpcsf, smin, PE_vp, 'FIX')
if(PF_fix['errV']<PF_cb['errV']):
PF_cb=PF_fix
PF_cb=_padding(PF_cb, lenb)
# (B-Vp-2) update params (if current params is not good enough)
if(F_WANT_OPT and PF_cb['errV']>mdb.rho_RF):
PF_upd=_FM(mdb, Xpc, lenpcsf, smin, PE_vp, 'OPT')
if(PF_upd['errV'] < PF_cb['errV']):
PF_cb=PF_upd
PF_cb=_padding(PF_cb, lenb)
#-------------------------------------------------#
# (C) if current best is BAD-fit, try Vc
if(PF_cb['errV']>mdb.rho_RF):
#-------------------------------------------------#
# (C-1) if current is given, and known
if(PE_vc['cid']!=-1):
#-------------------------------------------------#
# (C-1-1) use current params
PF_fix=_FM(mdb, Xc, lencsf, smin, PE_vc, 'FIX')
if(PF_fix['errV']<PF_cb['errV']):
PF_cb=PF_fix
PF_cb=_padding(PF_cb, lenbp)
#-------------------------------------------------#
# (C-1-2) update params (if current params is not good enough)
if(F_WANT_OPT and PF_fix['errV']>mdb.rho_RF):
PF_upd=_FM(mdb, Xc, lencsf, smin, PE_vc, 'OPT')
if(PF_upd['errV'] < PF_cb['errV']):
PF_cb=PF_upd
PF_cb=_padding(PF_cb, lenbp)
#-------------------------------------------------#
# (C-2) if current is given, but unknown, try Vc but use only singlestep
if( not _isNullP(PE_vc) and PE_vc['cid']==-1 ):
# (C-2-1) use current param
PF_fix=_FS(mdb, Xc, lencsf, PE_vc, 'FIX')
if(PF_fix['errV']<PF_cb['errV']):
PF_cb=PF_fix
PF_cb=_padding(PF_cb, lenbp)
#-------------------------------------------------#
#Snaps['Sf_full'][tf_st:tf_ed,:PF_cb['md'].k]=PF_cb['Se'][lenbpc+lstep:,:]
Snaps['Vf_full'][tf_st:tf_ed,:]=PF_cb['Ve'][lenbpc+lstep:,:]
Snaps['Rf_full'][tf_st:tf_ed]=PF_cb['Re'][lenbpc+lstep:]
Snaps['Xe'].append(Xe)
Snaps['Ve'].append(PF_cb['Ve'])
Snaps['Se'].append(PF_cb['Se'])
Snaps['Re'].append(PF_cb['Re'])
Snaps['Te'].append([tb_st, tb_ed, tp_st, tp_ed, tc_st, tc_ed, tf_st, tf_ed])
return (Snaps)
#--------------------------------#
# hyper-parameter setting #
#--------------------------------#
def _compute_rho_min(Xorg, lstep):
(n,d)=np.shape(Xorg)
lmin=max(LMIN, int(lstep*AVGR))
errs=[]
for i in range(0,AVG_TRIAL):
tm_st=int((n-lmin)*np.random.rand()); tm_ed=tm_st+lmin
if(tm_st<0 or tm_ed>=n): continue
Xc=Xorg[tm_st:tm_ed,:]
md_c=nl.NLDS(Xc, "t-%d"%(i))
md_c=md_c.fit(W_RE) # lmfit
(Sc,Vc)=md_c.gen() # generate events
err=tl.RMSE(Xc,Vc)
errs.append(err)
if(DBG1): tl.msg("trial: %d, st:%d, ed:%d, err:%.2f"%(i, tm_st, tm_ed, err))
#md_c.plot("%s_s_%d"%(mdb.fn,i))
rho=om_mdb.compute_RHO_W_ERRs(errs)
tl.eprint("estimated initial rho: ", rho) #, errs)
return rho
#--------------------------------#
#---------------#
# main #
#---------------#
if __name__ == "__main__":
tl.msg("OrbitMap")