def likelihood_wrapper(theta, timeBins, N, freqs, z_bins, z_logcdf, z_logsf,
ancGLs, ancHapGLs, gens, noCoals, currFreq, h, sMax,
changePts):
S = theta
Sprime = np.concatenate((S, [0.0]))
if np.any(np.abs(Sprime) > sMax):
return np.inf
sel = Sprime[np.digitize(epochs, timeBins, right=False) - 1]
tShape = times.shape
if tShape[2] == 0:
t = np.zeros((2, 0))
importanceSampling = False
elif tShape[2] == 1:
t = times[:, :, 0]
importanceSampling = False
else:
importanceSampling = True
if importanceSampling:
M = tShape[2]
loglrs = np.zeros(M)
for i in range(M):
betaMat = backward_algorithm(sel,
times[:, :, i],
epochs,
N,
freqs,
z_bins,
z_logcdf,
z_logsf,
ancGLs,
ancHapGLs,
changePts,
noCoals=noCoals,
currFreq=currFreq,
h=h)
logl = logsumexp(betaMat[-2, :])
logl0 = proposal_density(times[:, :, i], epochs, N)
loglrs[i] = logl - logl0
logl = -1 * (-np.log(M) + logsumexp(loglrs))
else:
betaMat = backward_algorithm(sel,
t,
epochs,
N,
freqs,
z_bins,
z_logcdf,
z_logsf,
ancGLs,
ancHapGLs,
changePts,
noCoals=noCoals,
currFreq=currFreq,
h=h)
logl = -logsumexp(betaMat[-2, :])
#print(logl,S)
return logl
def traj_wrapper(theta,timeBins,N,freqs,z_bins,z_logcdf,z_logsf,ancGLs,ancHapGLs,gens,noCoals,currFreq,h,sMax,changePts):
S = theta
Sprime = np.concatenate((S,[0.0]))
if np.any(np.abs(Sprime) > sMax):
print('WARNING: selection coefficient exceeds bounds. Maybe change --sMax?')
return np.inf
sel = Sprime[np.digitize(epochs,timeBins,right=False)-1]
T = len(epochs)
F = len(freqs)
tShape = times.shape
if tShape[2] == 0:
t = np.zeros((2,0))
importanceSampling = False
elif tShape[2] == 1:
t = times[:,:,0]
importanceSampling = False
else:
importanceSampling = True
if importanceSampling:
M = tShape[2]
loglrs = np.zeros(M)
postBySamples = np.zeros((F,T-1,M))
for i in range(M):
betaMat = backward_algorithm(sel,times[:,:,i],epochs,N,freqs,z_bins,z_logcdf,z_logsf,ancGLs,ancHapGLs,changePts,noCoals=noCoals,currFreq=currFreq,h=h)
alphaMat = forward_algorithm(sel,times[:,:,i],epochs,N,freqs,z_bins,z_logcdf,z_logsf,ancGLs,ancHapGLs,changePts,noCoals=noCoals,h=h)
logl = logsumexp(betaMat[-2,:])
logl0 = proposal_density(times[:,:,i],epochs,N)
loglrs[i] = logl-logl0
postBySamples[:,:,i] = (alphaMat[1:,:] + betaMat[:-1,:]).transpose()
post = logsumexp(loglrs + postBySamples,axis=2)
post -= logsumexp(post,axis=0)
else:
post = np.zeros((F,T))
betaMat = backward_algorithm(sel,t,epochs,N,freqs,z_bins,z_logcdf,z_logsf,ancGLs,ancHapGLs,changePts,noCoals=noCoals,currFreq=currFreq,h=h)
alphaMat = forward_algorithm(sel,t,epochs,N,freqs,z_bins,z_logcdf,z_logsf,ancGLs,ancHapGLs,changePts,noCoals=noCoals,h=h)
post = (alphaMat[1:,:] + betaMat[:-1,:]).transpose()
post -= logsumexp(post,axis=0)
return post