-
Notifications
You must be signed in to change notification settings - Fork 0
/
d5mDecodePCgm2.py
438 lines (356 loc) · 15.1 KB
/
d5mDecodePCgm2.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
import cPickle as pickle
import numpy as _N
import matplotlib.pyplot as _plt
import kde as _kde
import scipy.stats as _ss
#import fitMvNormE as fMN
import fitMvNorm2 as fMN
import time as _tm
import kdeutil as _ku
import scipy.integrate as _si
# Decode unsorted spike train
# In our case with no history dependence
# p(X_t | dN_t, m_t)
class simDecode():
nTets = 1
bWmaze= False
pX_Nm = None
Lklhd = None
kde = None
lmd = None
lmd0 = None
mltMk = 1 # multiply mark values to
AR = 0.5
AR0 = 1
# xp position grid. need this in decode
xp = None
xpr = None # reshaped xp
dxp = None
# current posterior model parameters
po_us = None
po_covs = None
po_ws = None
# initting fitMvNorm
sepHash = False
sepHashMthd=0
pctH = None
MS = None
kde = False
Bx = None; bx = None; Bm = None
tetfile = "marks.pkl"
usetets = None
utets_str= ""
tt0 = None
tt1 = None
dbgMvt = False
minds = None #
encN = 0 # how many time points used in encode
snpsht_us = []
snpsht_covs= []
snpsht_ms = []
curr_encpos = None
def init(self, kde=False, bx=None, Bx=None, Bm=None):
oo = self
oo.kde = kde
with open(oo.tetfile, "rb") as f:
lm = pickle.load(f)
f.close()
if oo.usetets is None:
oo.tetlist = lm.tetlist
oo.marks = lm.marks
oo.utets_str = "all"
else:
stetlist= set(oo.usetets)
mis = [i for i, item in enumerate(lm.tetlist) if item in stetlist]
oo.marks = lm.marks[:, mis]
print mis
for l in xrange(len(mis)):
sc = "" if (l == len(mis)-1) else ","
oo.utets_str += "%(l)s%(c)s" % {"l" : lm.tetlist[mis[l]], "c" : sc}
if oo.mltMk != 1:
nons = _N.equal(oo.marks, None)
for nt in xrange(oo.nTets):
spks = _N.where(nons[:, nt] == False)[0]
for l in xrange(len(spks)):
mklst = oo.marks[spks[l], nt][0]
for ll in xrange(len(mklst)):
mklst[ll] *= oo.mltMk
oo.nTets = oo.marks.shape[1]
oo.Nx = lm.Nx; oo.Nm = lm.Nm
oo.xA = lm.xA; oo.mA = lm.mA
oo.mdim = lm.k #kde.mdim
try:
if lm.minds is not None:
oo.minds = lm.minds
except AttributeError:
pass
#### spatial grid for evaluating firing rates
oo.xp = _N.linspace(-oo.xA, oo.xA, oo.Nx) # space points
oo.xpr = oo.xp.reshape((oo.Nx, 1))
# bin space for occupation histogram. same # intvs as space points
oo.dxp = oo.xp[1] - oo.xp[0]
oo.xb = _N.empty(oo.Nx+1)
oo.xb[0:oo.Nx] = oo.xp - 0.5*oo.dxp
oo.xb[oo.Nx] = oo.xp[-1]+ 0.5*oo.dxp
####
shp = [oo.Nx] # shape of kde
shp.extend([oo.Nm]*oo.mdim)
#oo.lmd= kde.kde.reshape(shp)
#oo.lmdFLaT = oo.lmd.reshape(oo.Nx, oo.Nm**oo.mdim)
oo.dt = lm.dt
oo.pos = lm.pos
oo.Fp, oo.q2p = 1, 0.003
oo.pX_Nm = _N.zeros((oo.pos.shape[0], oo.Nx))
oo.Lklhd = _N.zeros((oo.nTets, oo.pos.shape[0], oo.Nx))
oo.decmth = "kde"
if not oo.kde:
oo.decmth = "mxn"
oo.mvNrm= []
for nt in xrange(oo.nTets):
oo.mvNrm.append(fMN.fitMvNorm(oo.ITERS, oo.M, oo.mdim + 1))
oo.mvNrm[nt].ITERS = oo.ITERS
oo.mvNrm[nt].AR = oo.AR
oo.mvNrm[nt].AR0 = oo.AR0
def encode(self, t0=None, t1=None, encIntvs=None, initPriors=False, doTouchUp=False, MF=None, kmeansinit=True, telapse=0):
"""
eIntvs: array of times whose spikes we use to create encoding model
"""
oo = self
if encIntvs is None:
encIntvs = _N.array([[t0, t1]])
oo.N = t1-t0
else:
oo.N = encIntvs[-1, 1] - encIntvs[0, 0]
oo.tt0 = encIntvs[0, 0]
oo.tt1 = encIntvs[-1, 1]
tt1 = _tm.time()
oo.encN = _N.sum(encIntvs[:, 1] - encIntvs[:, 0])
oo.curr_encpos = _N.empty(oo.encN)
if initPriors:
oo.all_pos = _N.empty(oo.encN)
ii = 0
for ein in xrange(encIntvs.shape[0]):
oo.all_pos[ii:ii+(encIntvs[ein,1]-encIntvs[ein,0])] = oo.pos[encIntvs[ein,0]:encIntvs[ein,1]]
oo.curr_encpos[ii:ii+(encIntvs[ein,1]-encIntvs[ein,0])] = oo.pos[encIntvs[ein,0]:encIntvs[ein,1]]
ii += encIntvs[ein,1]-encIntvs[ein,0]
else: # just make this longer
ii = 0
for ein in xrange(encIntvs.shape[0]):
oo.curr_encpos[ii:ii+(encIntvs[ein,1]-encIntvs[ein,0])] = oo.pos[encIntvs[ein,0]:encIntvs[ein,1]]
ii += encIntvs[ein,1]-encIntvs[ein,0]
oo.all_pos = _N.array(oo.all_pos.tolist() + oo.curr_encpos.tolist())
dat = _N.empty(oo.N, dtype=list) # include times when no mvt.
stpos = [] # pos @ time of spikes
marks = [] # mark @ time of spikes
nspks = _N.zeros(oo.nTets, dtype=_N.int)
for nt in xrange(oo.nTets):
marks.append([])
stpos.append([])
nspks[nt] = 0
for ein in xrange(encIntvs.shape[0]):
t0, t1 = encIntvs[ein]
for n in xrange(t0, t1): # oo.marks "list of arrays"
if oo.marks[n, nt] is not None: # [arr(k-dim mark1), arr(k-dim mark2)]
themarks = oo.marks[n, nt]
for l in xrange(len(themarks)):
stpos[nt].append(oo.pos[n])
marks[nt].append(themarks[l])
nspks[nt] += 1
print "nspikes tet %(tet)d %(s)d from %(t0)d %(t1)d" % {"t0" : t0, "t1" : t1, "s" : nspks[nt], "tet" : nt}
oo.tr_pos = []
oo.tr_marks = []
for nt in xrange(oo.nTets):
oo.tr_pos.append(_N.array(stpos[nt]))
oo.tr_marks.append(_N.array(marks[nt]))
if not oo.kde:
oo.snpsht_us.append([])
oo.snpsht_covs.append([])
oo.snpsht_ms.append([])
tt2 = _tm.time()
if initPriors:
for nt in xrange(oo.nTets):
oo.mvNrm[nt].init0(stpos[nt], marks[nt], 0, nspks[nt], sepHash=oo.sepHash, pctH=oo.pctH, MS=oo.MS, sepHashMthd=oo.sepHashMthd, doTouchUp=doTouchUp, MF=MF, kmeansinit=kmeansinit)
tt3 = _tm.time()
for nt in xrange(oo.nTets):
print "encode Doing fit tetrode %d" % nt
oo.mvNrm[nt].fit(oo.mvNrm[nt].M, stpos[nt], marks[nt], 0, nspks[nt], oo.curr_encpos, init=initPriors)
oo.mvNrm[nt].set_priors_and_initial_values(telapse=telapse)
oo.snpsht_us[-1].append(_N.array(oo.mvNrm[nt].us))
oo.snpsht_covs[-1].append(_N.array(oo.mvNrm[nt].covs))
oo.snpsht_ms[-1].append(_N.array(oo.mvNrm[nt].ms))
tt4 = _tm.time()
print (tt2-tt1)
print (tt3-tt2)
print (tt4-tt3)
oo.setLmd0(nspks)
def decode(self, t0, t1):
oo = self
## each
oo.pX_Nm[t0] = 1. / oo.Nx
if oo.dbgMvt:
oo.pX_Nm[t0, 20:30] = 151/5.
A = _N.trapz(oo.pX_Nm[t0], dx=oo.dxp)
oo.pX_Nm[t0] /= A
oo.intgrd= _N.empty(oo.Nx)
oo.intgrd2d= _N.empty((oo.Nx, oo.Nx))
oo.intgrl = _N.empty(oo.Nx)
oo.xTrs = _N.zeros((oo.Nx, oo.Nx)) # Gaussian
x = _N.linspace(-oo.xA, oo.xA, oo.Nx)
## (xk - a xk1)
i = 0
if not oo.bWmaze:
for x1 in x:
j = 0
for x0 in x: # j is current position
oo.xTrs[i, j] = _N.exp(-((x1-oo.Fp*x0)**2)/(2*oo.q2p))
j += 1
i += 1
else:
grdsz = (12./oo.Nx)
spdGrdUnts = _N.diff(oo.pos) / grdsz # unit speed ( per ms ) in grid units
# avg. time it takes to move 1 grid is 1 / _N.mean(_N.abs(spdGrdUnts))
# p(i+1, i) = 1/<avg spdGrdUnts>
p1 = _N.mean(_N.abs(spdGrdUnts))*0.5
# assume Nx is even
k2 = 0.02
for i in xrange(0, oo.Nx/2):
oo.xTrs[i, i] = 1-p1
if i > 0:
oo.xTrs[i-1, i] = p1
if i > 1: ## next nearest neighbor
oo.xTrs[i-2, i] = p1*k2
oo.xTrs[i+1, i] = p1*k2*0.5
elif i == 1:
oo.xTrs[oo.Nx/2-1, 1] = p1*k2/2
oo.xTrs[oo.Nx/2, 1] = p1*k2/2
oo.xTrs[i+1, i] = p1*k2*0.5
oo.xTrs[oo.Nx/2-1, 0] = p1/2
oo.xTrs[oo.Nx/2, 0] = p1/2
for i in xrange(oo.Nx/2, oo.Nx):
oo.xTrs[i, i] = 1-p1
if i < oo.Nx - 1:
oo.xTrs[i+1, i] = p1
if i < oo.Nx - 2:
oo.xTrs[i-1, i] = p1*k2*0.5
oo.xTrs[i+2, i] = p1*k2
elif i == oo.Nx-2:
oo.xTrs[i-1, i] = p1*k2*0.5
oo.xTrs[oo.Nx/2-1, oo.Nx-2] = p1*k2/2
oo.xTrs[oo.Nx/2, oo.Nx-2] = p1*k2/2
oo.xTrs[oo.Nx/2-1, oo.Nx-1] = p1/2
oo.xTrs[oo.Nx/2, oo.Nx-1] = p1/2
#oo.xTrs[:, j] += _N.mean(oo.xTrs[:, j])*0.01
for i in xrange(oo.Nx):
A = _N.trapz(oo.xTrs[:, i])*((2.*oo.xA)/oo.Nx)
oo.xTrs[:, i] /= A
# keep in mind that k_{k-1} is not treated as a value with a correct answer.
# integrate over all possible values of x_{k-1}
# Need value of integrand for all x_{k-1}
# I will perform integral L times for each time step
# multiply integral with p(\Delta N_k, m_k | x_k)
pNkmk0 = _N.exp(-oo.dt * oo.Lam_xk) # one for each tetrode
pNkmk = _N.ones(oo.Nx)
fxdMks = _N.empty((oo.Nx, oo.mdim+1)) # for each pos, a fixed mark
fxdMks[:, 0] = oo.xp
pNkmk = _N.empty((oo.Nx, oo.nTets))
tStart = _tm.time()
if oo.kde:
occ = 1./oo.iocc
iBx2 = 1. / (oo.Bx * oo.Bx)
sptl = []
for nt in xrange(oo.nTets):
sptl.append(-0.5*iBx2*(oo.xpr - oo.tr_pos[nt])**2) # this piece doesn't need to be evalu
if not oo.kde:
for nt in xrange(oo.nTets):
oo.mvNrm[nt].iSgs = _N.linalg.inv(oo.mvNrm[nt].covs)
oo.mvNrm[nt].i2pidcovs = 1/_N.sqrt(2*_N.pi*_N.linalg.det(oo.mvNrm[nt].covs))
oo.mvNrm[nt].i2pidcovsr= oo.mvNrm[nt].i2pidcovs.reshape((oo.mvNrm[nt].M, 1))
for t in xrange(t0+1,t1): # start at 1 because initial condition
#tt1 = _tm.time()
for nt in xrange(oo.nTets):
oo.Lklhd[nt, t] = pNkmk0[:, nt]
# build likelihood
if (oo.marks[t, nt] is not None) and (not oo.dbgMvt):
nSpks = len(oo.marks[t, nt])
for ns in xrange(nSpks):
fxdMks[:, 1:] = oo.marks[t, nt][ns]
if oo.kde:
#(atMark, fld_x, tr_pos, tr_mks, all_pos, mdim, Bx, cBm, bx)
#_ku.kerFr(fxdMks[0, 1:], fxdMks[:, 0], oo.tr_pos, oo.tr_mks, oo.mvpos, oo.mdim, oo.Bx, oo.cBm, oo.bx)
oo.Lklhd[nt, t] *= _ku.kerFr(fxdMks[0, 1:], sptl[nt], oo.tr_marks[nt], oo.mdim, oo.Bx, oo.Bm, oo.bx)* oo.iocc*oo.dt
else:
oo.Lklhd[nt, t] *= oo.mvNrm[nt].evalAtFxdMks_new(fxdMks)*oo.lmd0[nt] * oo.dt
ttt1 =0
ttt2 =0
ttt3 =0
#tt2 = _tm.time()
# transition convolved with previous posterior
_N.multiply(oo.xTrs, oo.pX_Nm[t-1], out=oo.intgrd2d)
oo.intgrl = _N.trapz(oo.intgrd2d, dx=oo.dxp, axis=1)
#for ixk in xrange(oo.Nx): # above trapz over 2D array
# oo.intgrl[ixk] = _N.trapz(oo.intgrd2d[ixk], dx=oo.dxp)
#tt3 = _tm.time()
oo.pX_Nm[t] = oo.intgrl * _N.product(oo.Lklhd[:, t], axis=0)
A = _N.trapz(oo.pX_Nm[t], dx=oo.dxp)
oo.pX_Nm[t] /= A
#tt4 = _tm.time()
#print "%(1).3e %(2).3e %(3).3e" % {"1" : (tt2-tt1), "2" : (tt3-tt2), "3" : (tt4-tt3)}
#print "%(1).3e %(2).3e" % {"1" : ttt1, "2" : ttt2}
tEnd = _tm.time()
print "decode %(1).3e" % {"1" : (tEnd-tStart)}
def setLmd0(self, nspks):
"""
Lmd0.
"""
oo = self
#####
if oo.dbgMvt:
occ = _N.ones(oo.Nx) / oo.Nx
oo.iocc = 1./occ
oo.Lam_xk = _N.zeros((oo.Nx, oo.nTets))
return
oo.lmd0 = _N.empty(oo.nTets)
oo.Lam_xk = _N.ones((oo.Nx, oo.nTets))
ibx2 = 1./ (oo.bx*oo.bx)
#occ = _N.sum(_N.exp(-0.5*ibx2*(oo.xpr - oo.all_pos[t0:t1])**2), axis=1) # this piece doesn't need to be evaluated for every new spike
occ = _N.sum(_N.exp(-0.5*ibx2*(oo.xpr - oo.all_pos)**2), axis=1) # this piece doesn't need to be evaluated for every new spike
Tot_occ = _N.sum(occ)
#oo.iocc = 1./(occ + Tot_occ*0.01)
oo.iocc = 1./occ
if oo.kde:
for nt in xrange(oo.nTets):
oo.Lam_xk[:, nt] = _ku.Lambda(oo.xpr, oo.tr_pos[nt], oo.all_pos, oo.Bx, oo.bx)
else: ##### fit mix gaussian
for nt in xrange(oo.nTets):
cmps = _N.zeros((oo.mvNrm[nt].M, oo.Nx))
for m in xrange(oo.mvNrm[nt].M):
var = oo.mvNrm[nt].covs[m, 0, 0]
ivar = 1./var
cmps[m] = (1/_N.sqrt(2*_N.pi*var)) * _N.exp(-0.5*ivar*(oo.xp - oo.mvNrm[nt].us[m, 0])**2)
y = _N.sum(oo.mvNrm[nt].ms*cmps, axis=0)
MargLam = y# * oo.iocc
oo.lmd0[nt] = (nspks[nt] / (oo.encN*0.001)) / _N.trapz(MargLam, dx=oo.dxp)
oo.Lam_xk[:, nt] = oo.lmd0[nt] * MargLam
def scoreDecode(self, dt0, dt1):
oo = self
maxPos = _N.max(oo.pX_Nm[dt0:dt1], axis=1)
inds = _N.empty(dt1-dt0, dtype=_N.int)
for t in xrange(dt0, dt1):
inds[t-dt0] = _N.where(oo.pX_Nm[t] == maxPos[t-dt0])[0][0]
diffPos = oo.xp[inds] - oo.pos[dt0:dt1]
return _N.sum(_N.abs(diffPos)) / (dt1-dt0)
def dump(self, dir):
oo = self
pcklme = {}
pcklme["posteriors"] = oo.pX_Nm
pcklme["Lklhds"] = oo.Lklhd
if not oo.kde:
pcklme["snpsht_us"] = oo.snpsht_us
pcklme["snpsht_covs"] = oo.snpsht_covs
pcklme["snpsht_ms"] = oo.snpsht_ms
dmp = open("%s/dec.dump" % dir, "wb")
pickle.dump(pcklme, dmp, -1)
dmp.close()
# import pickle
# with open("mARp.dump", "rb") as f:
# lm = pickle.load(f)