-
Notifications
You must be signed in to change notification settings - Fork 64
/
spm_DEM.m
770 lines (647 loc) · 28.6 KB
/
spm_DEM.m
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
function [DEM] = spm_DEM(DEM)
% Dynamic expectation maxmisation (Variational Laplacian filtering)
% FORMAT DEM = spm_DEM(DEM)
%
% DEM.M - hierarchical model
% DEM.Y - response variable, output or data
% DEM.U - explanatory variables, inputs or prior expectation of causes
% DEM.X - confounds
%__________________________________________________________________________
%
% generative model
%--------------------------------------------------------------------------
% M(i).g = y(t) = g(x,v,P) {inline function, string or m-file}
% M(i).f = dx/dt = f(x,v,P) {inline function, string or m-file}
%
% M(i).pE = prior expectation of p model-parameters
% M(i).pC = prior covariances of p model-parameters
% M(i).hE = prior expectation of h log-precision (cause noise)
% M(i).hC = prior covariances of h log-precision (cause noise)
% M(i).gE = prior expectation of g log-precision (state noise)
% M(i).gC = prior covariances of g log-precision (state noise)
% M(i).Q = precision components (input noise)
% M(i).R = precision components (state noise)
% M(i).V = fixed precision (input noise)
% M(i).W = fixed precision (state noise)
% M(i).xP = precision (states)
%
% M(i).m = number of inputs v(i + 1);
% M(i).n = number of states x(i);
% M(i).l = number of output v(i);
%
% conditional moments of model-states - q(u)
%--------------------------------------------------------------------------
% qU.x = Conditional expectation of hidden states
% qU.v = Conditional expectation of causal states
% qU.w = Conditional prediction error (states)
% qU.z = Conditional prediction error (causes)
% qU.C = Conditional covariance: cov(v)
% qU.S = Conditional covariance: cov(x)
%
% conditional moments of model-parameters - q(p)
%--------------------------------------------------------------------------
% qP.P = Conditional expectation
% qP.C = Conditional covariance
%
% conditional moments of hyper-parameters (log-transformed) - q(h)
%--------------------------------------------------------------------------
% qH.h = Conditional expectation (cause noise)
% qH.g = Conditional expectation (state noise)
% qH.C = Conditional covariance
%
% F = log evidence = log marginal likelihood = negative free energy
%__________________________________________________________________________
%
% spm_DEM implements a variational Bayes (VB) scheme under the Laplace
% approximation to the conditional densities of states (u), parameters (p)
% and hyperparameters (h) of any analytic nonlinear hierarchical dynamic
% model, with additive Gaussian innovations. It comprises three
% variational steps (D,E and M) that update the conditional moments of u, p
% and h respectively
%
% D: qu.u = max <L>q(p,h)
% E: qp.p = max <L>q(u,h)
% M: qh.h = max <L>q(u,p)
%
% where qu.u corresponds to the conditional expectation of hidden states x
% and causal states v and so on. L is the ln p(y,u,p,h|M) under the model
% M. The conditional covariances obtain analytically from the curvature of
% L with respect to u, p and h.
%
% The D-step is embedded in the E-step because q(u) changes with each
% sequential observation. The dynamical model is transformed into a static
% model using temporal derivatives at each time point. Continuity of the
% conditional trajectories q(u,t) is assured by a continuous ascent of F(t)
% in generalised co-ordinates. This means DEM can deconvolve online and
% can represents an alternative to Kalman filtering or alternative Bayesian
% update procedures.
%
%
% To accelerate computations one can specify the nature of the model using
% the field:
%
% M(1).E.linear = 0: full - evaluates 1st and 2nd derivatives
% M(1).E.linear = 1: linear - equations are linear in x and v
% M(1).E.linear = 2: bilinear - equations are linear in x, v and x.v
% M(1).E.linear = 3: nonlinear - equations are linear in x, v, x.v, and x.x
% M(1).E.linear = 4: full linear - evaluates 1st derivatives (for generalised
% filtering, where parameters change)
%__________________________________________________________________________
% Copyright (C) 2008 Wellcome Trust Centre for Neuroimaging
% Karl Friston
% $Id: spm_DEM.m 5892 2014-02-23 11:00:16Z karl $
% check model, data, priors and confounds and unpack
%--------------------------------------------------------------------------
[M,Y,U,X] = spm_DEM_set(DEM);
% find or create a DEM figure
%--------------------------------------------------------------------------
Fdem = spm_figure('GetWin','DEM');
% tolerance for changes in norm
%--------------------------------------------------------------------------
TOL = exp(-4);
% order parameters (d = n = 1 for static models) and checks
%==========================================================================
d = M(1).E.d + 1; % embedding order of q(v)
n = M(1).E.n + 1; % embedding order of q(x) (n >= d)
% number of states and parameters
%--------------------------------------------------------------------------
nY = size(Y,2); % number of samples
nl = size(M,2); % number of levels
nv = sum(spm_vec(M.m)); % number of v (casual states)
nx = sum(spm_vec(M.n)); % number of x (hidden states)
ny = M(1).l; % number of y (inputs)
nc = M(end).l; % number of c (prior causes)
nu = nv*d + nx*n; % number of generalised states
% number of iterations
%--------------------------------------------------------------------------
try, nD = M(1).E.nD; catch, nD = 1; end
try, nE = M(1).E.nE; catch, nE = 8; end
try, nM = M(1).E.nM; catch, nM = 8; end
try, K = M(1).E.K; catch, K = 1; end
% initialise regularisation parameters
%--------------------------------------------------------------------------
if nx
td = 1/nD; % integration time for D-Step
else
td = {2};
end
if M(1).E.linear == 1
te = 4; % integration time for E-Step
else
te = 0;
end
tm = 4; % integration time for M-Step
% precision components Q{} requiring [Re]ML estimators (M-Step)
%==========================================================================
Q = {};
for i = 1:nl
v0{i,i} = sparse(M(i).l,M(i).l);
w0{i,i} = sparse(M(i).n,M(i).n);
end
V0 = kron(sparse(n,n),spm_cat(v0));
W0 = kron(sparse(n,n),spm_cat(w0));
Qp = blkdiag(V0,W0);
for i = 1:nl
% precision (R) and covariance of generalised errors
%----------------------------------------------------------------------
iVv = spm_DEM_R(n,M(i).sv);
iVw = spm_DEM_R(n,M(i).sw);
% noise on causal states (Q)
%----------------------------------------------------------------------
for j = 1:length(M(i).Q)
q = v0;
q{i,i} = M(i).Q{j};
Q{end + 1} = blkdiag(kron(iVv,spm_cat(q)),W0);
end
% and fixed components (V)
%----------------------------------------------------------------------
q = v0;
q{i,i} = M(i).V;
Qp = Qp + blkdiag(kron(iVv,spm_cat(q)),W0);
% noise on hidden states (R)
%----------------------------------------------------------------------
for j = 1:length(M(i).R)
q = w0;
q{i,i} = M(i).R{j};
Q{end + 1} = blkdiag(V0,kron(iVw,spm_cat(q)));
end
% and fixed components (W)
%----------------------------------------------------------------------
q = w0;
q{i,i} = M(i).W;
Qp = Qp + blkdiag(V0,kron(iVw,spm_cat(q)));
end
% number of hyperparameters
%--------------------------------------------------------------------------
nh = length(Q);
% fixed priors on states (u)
%--------------------------------------------------------------------------
xP = spm_cat(spm_diag({M.xP}));
Px = kron(spm_DEM_R(n,0),xP);
Pv = kron(spm_DEM_R(d,0),sparse(nv,nv));
Pu = spm_cat(spm_diag({Px Pv}));
Pu = Pu + speye(nu,nu)*nu*eps;
% hyperpriors
%--------------------------------------------------------------------------
ph.h = spm_vec({M.hE M.gE}); % prior expectation of h
ph.c = spm_cat(spm_diag({M.hC M.gC})); % prior covariances of h
qh.h = ph.h; % conditional expectation
qh.c = ph.c; % conditional covariance
ph.ic = spm_pinv(ph.c); % prior precision
% priors on parameters (in reduced parameter space)
%==========================================================================
pp.c = cell(nl,nl);
qp.p = cell(nl,1);
for i = 1:(nl - 1)
% eigenvector reduction: p <- pE + qp.u*qp.p
%----------------------------------------------------------------------
qp.u{i} = spm_svd(M(i).pC,0); % basis for parameters
M(i).p = size(qp.u{i},2); % number of qp.p
qp.p{i} = sparse(M(i).p,1); % initial qp.p
pp.c{i,i} = qp.u{i}'*M(i).pC*qp.u{i}; % prior covariance
end
Up = spm_cat(spm_diag(qp.u));
% initialise and augment with confound parameters B; with flat priors
%--------------------------------------------------------------------------
np = sum(spm_vec(M.p)); % number of model parameters
nb = size(X,1); % number of confounds
nn = nb*ny; % number of nuisance parameters
nf = np + nn; % number of free parameters
ip = (1:np);
ib = (1:nn) + np;
pp.c = spm_cat(pp.c);
pp.ic = spm_inv(pp.c);
pp.p = spm_vec(qp.p);
% initialise conditional density q(p) := qp.e (for D-Step)
%--------------------------------------------------------------------------
for i = 1:(nl - 1)
try
qp.e{i} = qp.p{i} + qp.u{i}'*(spm_vec(M(i).P) - spm_vec(M(i).pE));
catch
qp.e{i} = qp.p{i};
end
end
qp.e = spm_vec(qp.e);
qp.c = sparse(nf,nf);
qp.b = sparse(ny,nb);
% initialise dedb
%--------------------------------------------------------------------------
for i = 1:nl
dedbi{i,1} = sparse(M(i).l,nn);
end
for i = 1:nl - 1
dndbi{i,1} = sparse(M(i).n,nn);
end
for i = 1:n
dEdb{i,1} = spm_cat(dedbi);
end
for i = 1:n
dNdb{i,1} = spm_cat(dndbi);
end
dEdb = [dEdb; dNdb];
% initialise cell arrays for D-Step; e{i + 1} = (d/dt)^i[e] = e[i]
%==========================================================================
qu.x = cell(n,1);
qu.v = cell(n,1);
qu.y = cell(n,1);
qu.u = cell(n,1);
[qu.x{:}] = deal(sparse(nx,1));
[qu.v{:}] = deal(sparse(nv,1));
[qu.y{:}] = deal(sparse(ny,1));
[qu.u{:}] = deal(sparse(nc,1));
% initialise cell arrays for hierarchical structure of x[0] and v[0]
%--------------------------------------------------------------------------
x = {M(1:end - 1).x};
v = {M(1 + 1:end).v};
qu.x{1} = spm_vec(x);
qu.v{1} = spm_vec(v);
% derivatives for Jacobian of D-step
%--------------------------------------------------------------------------
Dx = kron(spm_speye(n,n,1),spm_speye(nx,nx,0));
Dv = kron(spm_speye(d,d,1),spm_speye(nv,nv,0));
Dy = kron(spm_speye(n,n,1),spm_speye(ny,ny,0));
Dc = kron(spm_speye(d,d,1),spm_speye(nc,nc,0));
D = spm_cat(spm_diag({Dx,Dv,Dy,Dc}));
% and null blocks
%--------------------------------------------------------------------------
dVdy = sparse(n*ny,1);
dVdc = sparse(d*nc,1);
dVdyy = sparse(n*ny,n*ny);
dVdcc = sparse(d*nc,d*nc);
% gradients and curvatures for conditional uncertainty
%--------------------------------------------------------------------------
dWdu = sparse(nu,1);
dWdp = sparse(nf,1);
dWduu = sparse(nu,nu);
dWdpp = sparse(nf,nf);
% preclude unnecessary iterations
%--------------------------------------------------------------------------
if ~nh, nM = 1; end
if ~nf && ~nh, nE = 1; end
% preclude very precise states from entering free-energy/action
%--------------------------------------------------------------------------
ix = (1:(nx*n)) + ny*n + nv*n;
iv = (1:(nv*d)) + ny*n;
je = diag(Qp) < exp(16);
ju = [je(ix); je(iv)];
% E-Step: (with embedded D and M-Steps)
%==========================================================================
Fi = -Inf;
for iE = 1:nE
% get time and celar persistent variables in evaluation routines
%----------------------------------------------------------------------
tic; clear spm_DEM_eval
% [re-]set accumulators for E-Step
%----------------------------------------------------------------------
dFdh = sparse(nh,1); % gradient (hyperparamteres)
dFdhh = sparse(nh,nh); % curvatiure (hyperparamteres)
dFdp = sparse(nf,1); % gradient (paramteres)
dFdpp = sparse(nf,nf); % curvatiure (paramteres)
qp.ic = sparse(0); % conditional precision (p)
iqu.c = sparse(0); % conditional information (p)
EE = sparse(0);
ECE = sparse(0);
% [re-]set precisions using ReML hyperparameter estimates
%----------------------------------------------------------------------
iS = Qp;
for i = 1:nh
iS = iS + Q{i}*exp(qh.h(i));
end
% [re-]adjust for confounds
%----------------------------------------------------------------------
Y = Y - qp.b*X;
% [re-]set states & their derivatives
%----------------------------------------------------------------------
try, qu = qU(1); end
% D-Step: (nD D-Steps for each sample)
%======================================================================
for iY = 1:nY
% [re-]set states for static systems
%------------------------------------------------------------------
if ~nx, try, qu = qU(iY); end, end
% D-Step: until convergence for static systems
%==================================================================
Fd = -exp(64);
for iD = 1:nD
% sampling time
%--------------------------------------------------------------
ts = iY + (iD - 1)/nD;
% derivatives of responses and inputs
%--------------------------------------------------------------
try
qu.y(1:n) = spm_DEM_embed(Y,n,ts,1,M(1).delays);
qu.u(1:d) = spm_DEM_embed(U,d,ts);
catch
qu.y(1:n) = spm_DEM_embed(Y,n,ts);
qu.u(1:d) = spm_DEM_embed(U,d,ts);
end
% compute dEdb (derivatives of confounds)
%--------------------------------------------------------------
b = spm_DEM_embed(X,n,ts);
for i = 1:n
dedbi{1} = -kron(b{i}',speye(ny,ny));
dEdb{i,1} = spm_cat(dedbi);
end
% evaluate functions:
% E = v - g(x,v) and derivatives dE.dx, ...
%==============================================================
[E,dE] = spm_DEM_eval(M,qu,qp);
% conditional covariance [of states {u}]
%--------------------------------------------------------------
qu.p = real(dE.du'*iS*dE.du) + Pu;
qu.c = diag(ju)*spm_inv(qu.p)*diag(ju);
iqu.c = iqu.c + spm_logdet(qu.c);
% and conditional covariance [of parameters {P}]
%--------------------------------------------------------------
dE.dP = spm_cat({dE.dp dEdb});
ECEu = dE.du*qu.c*dE.du';
ECEp = dE.dP*qp.c*dE.dP';
if ~nx
% Evaluate objective function L(t) (for static models)
%----------------------------------------------------------
L = - trace(real(E'*iS*E))/2 ... % states (u)
- trace(real(iS*ECEp))/2; % expectation q(p)
% if F is increasing, save expansion point
%----------------------------------------------------------
if L > Fd
td = {min(td{1} + 1, 4)};
Fd = L;
B.qu = qu;
B.E = E;
B.dE = dE;
B.ECEp = ECEp;
else
% otherwise, return to previous expansion point
%------------------------------------------------------
qu = B.qu;
E = B.E;
dE = B.dE;
ECEp = B.ECEp;
td = {min(td{1} - 2,-4)};
end
end
% save states at qu(t)
%--------------------------------------------------------------
if iD == 1
qE{iY} = E;
qU(iY) = qu;
end
% uncertainty about parameters dWdv, ... ; W = ln(|qp.c|)
%==============================================================
if np
for i = 1:nu
CJp(:,i) = spm_vec(qp.c(ip,ip)*dE.dpu{i}'*iS);
dEdpu(:,i) = spm_vec(dE.dpu{i}');
end
dWdu = real(CJp'*spm_vec(dE.dp'));
dWduu = real(CJp'*dEdpu);
end
% D-step update: of causes v{i}, and hidden states x(i)
%==============================================================
% conditional modes
%--------------------------------------------------------------
q = {qu.x{1:n} qu.v{1:d} qu.y{1:n} qu.u{1:d}};
u = spm_vec(q);
% first-order derivatives
%--------------------------------------------------------------
dVdu = -real(dE.du'*iS*E) - dWdu/2 - Pu*u(1:nu);
% and second-order derivatives
%--------------------------------------------------------------
dVduu = -real(dE.du'*iS*dE.du) - dWduu/2 - Pu;
dVduy = -real(dE.du'*iS*dE.dy);
dVduc = -real(dE.du'*iS*dE.dc);
% gradient
%--------------------------------------------------------------
dFdu = spm_vec({dVdu; dVdy; dVdc });
% Jacobian (variational flow)
%--------------------------------------------------------------
dFduu = spm_cat({dVduu dVduy dVduc ;
[] dVdyy [] ;
[] [] dVdcc});
% update conditional modes of states
%==============================================================
f = K*dFdu + D*u;
dfdu = K*dFduu + D;
if iD == 1 && isfield(DEM,'E')
DEM.E(:,iY) = eig(full(dfdu));
end
du = spm_dx(dfdu,f,td);
q = spm_unvec(u + du,q);
% and save them
%--------------------------------------------------------------
qu.x(1:n) = q((1:n));
qu.v(1:d) = q((1:d) + n);
% D-Step: break if convergence (for static models)
%--------------------------------------------------------------
if ~nx
qU(iY) = qu;
end
if ~nx && ((dFdu'*du < TOL) || (norm(du,1) < TOL))
break
end
end % D-Step
% Gradients and curvatures for E-Step: W = tr(C*J'*iS*J)
%==================================================================
if np
for i = ip
CJu(:,i) = spm_vec(qu.c*dE.dup{i}'*iS);
dEdup(:,i) = spm_vec(dE.dup{i}');
end
dWdp(ip) = CJu'*spm_vec(dE.du');
dWdpp(ip,ip) = CJu'*dEdup;
end
% Accumulate; dF/dP = <dL/dp>, dF/dpp = ...
%------------------------------------------------------------------
dFdp = dFdp - dWdp/2 - real(dE.dP'*iS*E);
dFdpp = dFdpp - dWdpp/2 - real(dE.dP'*iS*dE.dP);
qp.ic = qp.ic + real(dE.dP'*iS*dE.dP);
% and quantities for M-Step
%------------------------------------------------------------------
EE = real(E*E') + EE;
ECE = ECE + ECEu + ECEp;
end % sequence (nY)
% M-step - optimise hyperparameters (mh = total update)
%======================================================================
mh = 0;
for iM = 1:nM
% [re-]set precisions using ReML hyperparameter estimates
%------------------------------------------------------------------
iS = Qp;
for i = 1:nh
iS = iS + Q{i}*exp(qh.h(i));
end
S = spm_inv(iS);
dS = ECE + EE - S*nY;
% 1st-order derivatives: dFdh = dF/dh
%------------------------------------------------------------------
for i = 1:nh
dPdh{i} = Q{i}*exp(qh.h(i));
dFdh(i,1) = -trace(dPdh{i}*dS)/2;
end
% 2nd-order derivatives: dFdhh
%------------------------------------------------------------------
for i = 1:nh
for j = 1:nh
dFdhh(i,j) = -trace(dPdh{i}*S*dPdh{j}*S*nY)/2;
end
end
% hyperpriors
%------------------------------------------------------------------
qh.e = qh.h - ph.h;
dFdh = dFdh - ph.ic*qh.e;
dFdhh = dFdhh - ph.ic;
% update ReML estimate of parameters
%------------------------------------------------------------------
dh = spm_dx(dFdhh,dFdh,{tm});
dh = max(min(dh,2),-2);
qh.h = qh.h + dh;
mh = mh + dh;
% conditional covariance of hyperparameters
%------------------------------------------------------------------
qh.c = -spm_inv(dFdhh);
% convergence (M-Step)
%------------------------------------------------------------------
if (dFdh'*dh < TOL) || (norm(dh,1) < TOL), break, end
end % M-Step
% conditional precision of parameters
%------------------------------------------------------------------
qp.ic(ip,ip) = qp.ic(ip,ip) + pp.ic;
qp.c = spm_inv(qp.ic);
% evaluate objective function (F)
%======================================================================
% free-energy and action
%----------------------------------------------------------------------
Lu = - trace(iS(je,je)*EE(je,je))/2 ... % states (u)
- n*ny*log(2*pi)*nY/2 ... % constant
+ spm_logdet(iS(je,je))*nY/2 ... % entropy - error
+ iqu.c/(2*nD); % entropy q(u)
Lp = - trace(qp.e'*pp.ic*qp.e)/2 ... % parameters (p)
- trace(qh.e'*ph.ic*qh.e)/2 ... % hyperparameters (h)
+ spm_logdet(qp.c(ip,ip)*pp.ic)/2 ... % entropy q(p)
+ spm_logdet(qh.c*ph.ic)/2; % entropy q(h)
La = - trace(qp.e'*pp.ic*qp.e)*nY/2 ... % parameters (p)
- trace(qh.e'*ph.ic*qh.e)*nY/2 ... % hyperparameters (h)
+ spm_logdet(qp.c(ip,ip)*pp.ic*nY)*nY/2 ... % entropy q(p)
+ spm_logdet(qh.c*ph.ic*nY)*nY/2; % entropy q(h)
Li = Lu + Lp; % free-energy
Ai = Lu + La; % free-action
% if F is increasing, save expansion point and derivatives
%------------------------------------------------------------------
if Li > Fi || iE < 2
% Accept free-energy and save current parameter estimates
%------------------------------------------------------------------
Fi = Li;
te = min(te + 1/2,4);
tm = min(tm + 1/2,4);
B.qp = qp;
B.qh = qh;
B.pp = pp;
% E-step: update expectation (p)
%==================================================================
% gradients and curvatures
%------------------------------------------------------------------
dFdp(ip) = dFdp(ip) - pp.ic*(qp.e - pp.p);
dFdpp(ip,ip) = dFdpp(ip,ip) - pp.ic;
% update conditional expectation
%------------------------------------------------------------------
dp = spm_dx(dFdpp,dFdp,{te});
qp.e = qp.e + dp(ip);
qp.p = spm_unvec(qp.e,qp.p);
qp.b = spm_unvec(dp(ib),qp.b);
else
% otherwise, return to previous expansion point
%------------------------------------------------------------------
nM = 1;
qp = B.qp;
pp = B.pp;
qh = B.qh;
te = min(te - 2, -2);
tm = min(tm - 2, -2);
end
F(iE) = Fi;
A(iE) = Ai;
% save model-states (for each time point)
%==================================================================
for t = 1:length(qU)
v = spm_unvec(qU(t).v{1},v);
x = spm_unvec(qU(t).x{1},x);
z = spm_unvec(qE{t}(1:(ny + nv)),{M.v});
w = spm_unvec(qE{t}([1:nx] + (ny + nv)*n),{M.x});
for i = 1:(nl - 1)
if M(i).m, QU.v{i + 1}(:,t) = spm_vec(v{i}); end
if M(i).n, QU.x{i}(:,t) = spm_vec(x{i}); end
if M(i).n, QU.w{i}(:,t) = spm_vec(w{i}); end
if M(i).l, QU.z{i}(:,t) = spm_vec(z{i}); end
end
QU.v{1}(:,t) = spm_vec(qU(t).y{1}) - spm_vec(z{1});
if M(nl).l, QU.z{nl}(:,t) = spm_vec(z{nl}); end
% and conditional covariances
%--------------------------------------------------------------
i = (1:nx);
QU.S{t} = qU(t).c(i,i);
i = (1:nv) + nx*n;
QU.C{t} = qU(t).c(i,i);
end
% report and break if convergence
%------------------------------------------------------------------
spm_figure('Select', Fdem)
spm_DEM_qU(QU)
if np
subplot(2*nl,2,4*nl)
bar(full(Up*qp.e))
xlabel({'parameters {minus prior}'})
end
if nh
subplot(2*nl,4,8*nl - 4)
bar(full(qh.h))
title({'log-precision'})
end
if length(F) > 2
subplot(2*nl,4,8*nl - 5)
plot(F - F(1))
xlabel('updates')
title('free-energy')
end
drawnow
% report (EM-Steps)
%------------------------------------------------------------------
str{1} = sprintf('DEM: %i (%i:%i)',iE,iD,iM);
str{2} = sprintf('F:%.4e',full(F(iE) - F(1)));
str{3} = sprintf('p:%.2e',full(norm(dp,1)));
str{4} = sprintf('h:%.2e',full(norm(mh,1)));
str{5} = sprintf('(%.2e sec)',full(toc));
fprintf('%-16s%-16s%-14s%-14s%-16s\n',str{:})
% Convergence
%------------------------------------------------------------------
if (norm(dp,1) < TOL*norm(spm_vec(qp.p),1)) && (norm(mh,1) < TOL), break, end
if te < -8, break, end
end
spm_figure('Focus', Fdem)
% Assemble output arguments
%==========================================================================
% conditional moments of model-parameters (rotated into original space)
%--------------------------------------------------------------------------
qP.P = spm_unvec(Up*qp.e + spm_vec(M.pE),M.pE);
qP.C = Up*qp.c(ip,ip)*Up';
qP.V = spm_unvec(diag(qP.C),M.pE);
qP.dFdp = Up*dFdp(ip);
qP.dFdpp = Up*dFdpp(ip,ip)*Up';
% conditional moments of hyper-parameters (log-transformed)
%--------------------------------------------------------------------------
qH.h = spm_unvec(qh.h,{{M.hE} {M.gE}});
qH.g = qH.h{2};
qH.h = qH.h{1};
qH.C = qh.c;
qH.V = spm_unvec(diag(qH.C),{{M.hE} {M.gE}});
qH.W = qH.V{2};
qH.V = qH.V{1};
% assign output variables
%--------------------------------------------------------------------------
DEM.M = M;
DEM.U = U; % causes
DEM.X = X; % confounds
DEM.qU = QU; % conditional moments of model-states
DEM.qP = qP; % conditional moments of model-parameters
DEM.qH = qH; % conditional moments of hyper-parameters
DEM.F = F; % [-ve] Free energy
DEM.S = A; % [-ve] Free action