import os
from math import *
import numpy as np
import numpy.fft as npft
[docs]def makeFitStruct_GLM(dtsim, kbasprs, nkt, flag_exp):
gg = {}
gg['k'] = []
gg['dc'] = 0
gg['kt'] = np.zeros((nkt,1))
gg['ktbas'] = []
gg['kbasprs'] = kbasprs
gg['dt'] = dtsim
nkt = nkt
if flag_exp==0:
ktbas = makeBasis_StimKernel(kbasprs,nkt)
else:
ktbas = makeBasis_StimKernel_exp(kbasprs,nkt)
gg['ktbas'] = ktbas
gg['k'] = gg['ktbas']*gg['kt']
return gg
[docs]def makeBasis_StimKernel(kbasprs, nkt):
neye = kbasprs['neye']
ncos = kbasprs['ncos']
kpeaks = kbasprs['kpeaks']
kdt = 1
b = kbasprs['b']
delays_raw = kbasprs['delays']
delays = delays_raw[0].astype(int)
ylim = np.array([100.,200.]) # !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!HARD-CODED FOR NOW
yrnge = nlin(ylim + b*np.ones(np.shape(kpeaks)))
db = (yrnge[-1]-yrnge[0])/(ncos-1)
ctrs = nlin(np.array(kpeaks)) # yrnge
mxt = invnl(yrnge[ncos-1]+2*db)-b
kt0 = np.arange(0, mxt, kdt) # -delay
nt = len(kt0)
e1 = np.tile(nlin(kt0 + b*np.ones(np.shape(kt0))), (ncos, 1))
e2 = np.transpose(e1)
e3 = np.tile(ctrs, (nt, 1))
kbasis0 = []
for kk in range(ncos):
kbasis0.append(ff(e2[:,kk],e3[:,kk],db))
#Concatenate identity vectors
nkt0 = np.size(kt0, 0)
a1 = np.concatenate((np.eye(neye), np.zeros((nkt0,neye))),axis=0)
a2 = np.concatenate((np.zeros((neye,ncos)),np.array(kbasis0).T),axis=0)
kbasis = np.concatenate((a1, a2),axis=1)
kbasis = np.flipud(kbasis)
nkt0 = np.size(kbasis,0)
if nkt0 < nkt:
kbasis = np.concatenate((np.zeros((nkt - nkt0, ncos + neye)), kbasis), axis=0)
elif nkt0 > nkt:
kbasis = kbasis[-1-nkt:-1, :]
kbasis = normalizecols(kbasis)
# Add delays for both functions. tack on delays (array of 0s) to the end of the function, then readjusts the second
# function so both are the same size.
kbasis2_0 = np.concatenate((kbasis[:, 0], np.zeros((delays[0], ))), axis=0)
kbasis2_1 = np.concatenate((kbasis[:, 1], np.zeros((delays[1], ))), axis=0)
len_diff = delays[1] - delays[0]
kbasis2_1 = kbasis2_1[len_diff:]
# combine and renormalize
kbasis2 = np.zeros((len(kbasis2_0), 2))
kbasis2[:, 0] = kbasis2_0
kbasis2[:, 1] = kbasis2_1
kbasis2 = normalizecols(kbasis2)
return kbasis2
[docs]def makeBasis_StimKernel_exp(kbasprs,nkt):
ks = kbasprs['ks']
b = kbasprs['b']
x0 = np.arange(0,nkt)
kbasis = np.zeros((nkt,len(ks)))
for ii in range(len(ks)):
kbasis[:,ii] = invnl(-ks[ii]*x0) # (1.0/ks[ii])*
kbasis = np.flipud(kbasis)
# kbasis = normalizecols(kbasis)
return kbasis
[docs]def nlin(x):
eps = 1e-20
# x.clip(0.)
return np.log(x+eps)
[docs]def invnl(x):
eps = 1e-20
return np.exp(x)-eps
[docs]def ff(x, c, dc):
rowsize = np.size(x,0)
m = []
for i in range(rowsize):
xi = x[i]
ci = c[i]
val=(np.cos(np.max([-pi, np.min([pi, (xi-ci)*pi/dc/2])])) + 1)/2
m.append(val)
return np.array(m)
[docs]def normalizecols(A):
B = A/np.tile(np.sqrt(sum(A**2,0)),(np.size(A,0),1))
return B
[docs]def sameconv(A,B):
am = np.size(A)
bm = np.size(B)
nn = am+bm-1
q = npft.fft(A,nn)*npft.fft(np.flipud(B),nn)
p = q
G = npft.ifft(p)
G = G[range(am)]
return G
