def run_sim(num_net, nets_inhm_attr, I9, nets_a, nets_g, vec_size_net,
vec_nmc_state, theta0, theta1, theta_setup, sampleinfo, sample_k,
mcJump, filename, jsim, jscid):
np.random.seed(jsim * num_net + jscid + 2026642028)
jattr = nets_inhm_attr[jscid]
jI9 = I9[jscid]
jA = nets_a[jscid]
jG = nets_g[jscid]
nn = int(vec_size_net[jscid])
jkk = sample_k[jscid]
nmc_state = vec_nmc_state[jscid]
[vv, ww, h, phi1, phi2, psi, qhat] = theta2param(theta1, jattr,
theta_setup, sampleinfo)
[jG1, jA1, p_th1_s1,
p_th1_s0] = gen_kCD_fixG(vv, ww, h, phi1, phi2, psi, qhat, jI9, nn, jkk,
mcJump, nmc_state, jG, jA)
## Could be more efficient.
[vv, ww, h, phi1, phi2, psi, qhat] = theta2param(theta0, jattr,
theta_setup, sampleinfo)
p_th0_s1 = potentialCwrap(vv, ww, h, phi1, phi2, psi, qhat, jI9, jG1, jA1,
nn)
p_th0_s0 = potentialCwrap(vv, ww, h, phi1, phi2, psi, qhat, jI9, jG, jA,
nn)
# if jsim>0 and jsim%5000==0:
# state2pickle(filename+'-kCD-state--scid-'+str(jscid+1).zfill(2)+'--sim-'+str(jsim+1).zfill(6),jG1,jA1)
# #print(f'Saving {posteriorfile}')
p_dTh_s1 = p_th1_s1 - p_th0_s1
p_dTh_s0 = p_th1_s0 - p_th0_s0
return (p_dTh_s0 - p_dTh_s1) #,mcstats
def gen_kCD_fixG(vv, ww, h, phi1, phi2, psi, qhat, I9, nn, kkin, mcJump, nmc,
G0, A0):
"""
Markov chain from the model via kCD
* mcJump prob large jumps inverting G and A
* large k acceptance is less likely
"""
## Random variables
#(a) meetings i
veci = (np.random.randint(0, nn, [nmc])).astype(np.int32)
#(b) meeting dimensions
nkk = kkin.shape[0] #kkin sample for dimensions of k
if nkk > 1:
veck = kkin[np.random.randint(0, nkk, [nmc])]
else:
veck = np.ones([nmc]) * kkin
veck = veck.astype(np.int32)
#(c) proposal mode
proposal = np.random.uniform(0, 1, [nmc])
#(d) accept
lnaccept = np.log(np.random.uniform(0, 1, [nmc]))
## Prep the data.
G = np.copy(G0)
A = np.copy(A0)
p = potentialCwrap(vv, ww, h, phi1, phi2, psi, qhat, I9, G, A, nn)
p0 = np.copy(p)
TempG1 = np.copy(G) #only for fixed G
G1 = np.copy(G) #only for fixed G
## Initializations & misc
dP = 0
p1 = 0
lnPbar = 0
index_set = np.arange(nn)
for ss in range(nmc):
## Draw a meeting i,jset.
ihat = veci[ss]
k = veck[ss]
feasible_jj = index_set[ihat != index_set]
jj = feasible_jj[np.random.permutation(nn - 1)[:k - 1]]
## Propose.
if proposal[ss] < mcJump:
#G1 = 1-G
#np.fill_diagonal(G1,0)
A1 = 1 - A
p1 = potentialCwrap(vv, ww, h, phi1, phi2, psi, qhat, I9, G1, A1,
nn)
else:
A1 = np.copy(A)
#G1 = np.copy(G)
#keep this draw to maintain parallel randomness with other estiamtion scenarios
TempG1[ihat, jj] = np.random.binomial(1, 0.5,
[k - 1]).astype(np.float64)
#G1[jj,ihat] = G1[ihat,jj].T
A1[ihat] = np.random.binomial(1, 0.5, 1).astype(np.float64)
dP = deltaPotentialCwrap(vv, ww, h, phi1, phi2, psi, qhat, I9, G,
G1, A, A1, ihat, nn)
p1 = p + dP
lnPbar = p1 - p
if lnaccept[ss] < lnPbar or lnPbar > 0:
A = A1
#G = G1
p = np.copy(p1)
return G, A, p, p0
def gen_kCD_checks(vv, ww, h, phi, psi, qhat, I9, nn, kkin, mcJump, nmc, G0,
A0):
"""
Markov chain from the model via kCD
* mcJump prob large jumps inverting G and A
* large k acceptance is less likely
"""
## Prep meetings
veci = (np.random.randint(0, nn, [nmc])).astype(np.int32)
nkk = kkin.shape[0] #kkin sample for dimensions of k
if nkk > 1:
veck = kkin[np.random.randint(0, nkk, [nmc])]
else:
veck = np.ones([nmc]) * kkin
veck = veck.astype(np.int32)
## Prep the data.
G = np.copy(G0)
A = np.copy(A0)
p = potentialCwrap(vv, ww, h, phi, psi, qhat, I9, G, A, nn)
p0 = np.copy(p)
## Draw acceptance
lnaccept = np.log(np.random.uniform(0, 1, [nmc]))
## Miscellanea & inits.
index_set = np.arange(nn)
dP = 0
p1 = 0
lnPbar = 0
for ss in range(nmc):
## Draw a meeting i,jset.
ihat = veci[ss]
k = veck[ss]
feasible_jj = index_set[ihat != index_set]
jj = feasible_jj[np.random.permutation(nn - 1)[:k - 1]]
## Propose.
if np.random.uniform() < mcJump:
G1 = 1 - G
np.fill_diagonal(G1, 0)
A1 = 1 - A
p1 = potentialCwrap(vv, ww, h, phi, psi, qhat, I9, G1, A1, nn)
else:
A1 = np.copy(A)
G1 = np.copy(G)
G1[ihat, jj] = (np.random.uniform(0, 1, [k - 1]) < 0.5).astype(
np.float64)
G1[jj, ihat] = G1[ihat, jj].T
A1[ihat] = float(np.random.uniform(0, 1) < 0.5)
# #CHECK SYMMETRIC
# if np.any(np.abs(G1-G1.T) > 0.5):
# print('Proposal not symmetric')
# #Next 4 lines change the parameters
# #to illustrate precision discrepancies
# vv=vv*0
#vv=np.ones(nn)*0.2
# if np.sum(np.iscomplex(vv))>0:
# print('Complex')
# if np.isnan(np.sum(vv))>0:
# print('NA')
#
# ww= np.ones([nn,nn])*1.0; np.fill_diagonal(ww,0)
# ww= np.ones([nn,nn])*0; np.fill_diagonal(ww,0)
# h=0.1
# phi=1 #phi*0 #?
# psi=1
# qhat=1 #qhat*0 #no problem
#G1=G
#A1=A
dP = deltaPotentialCwrap(vv, ww, h, phi, psi, qhat, I9, G, G1, A,
A1, ihat, nn)
p1 = p + dP
# p1_check = potentialCwrap(vv,ww,h,phi,psi,qhat,I9,G1,A1,nn)
# p0_check = potentialCwrap(vv,ww,h,phi,psi,qhat,I9,G,A,nn)
# if abs(dP-(p1_check-p0_check))>1e-12:
# print(f'dP={dP:10.4f} , p1check-p0check={p1_check-p0_check:10.4f}, error={dP-(p1_check-p0_check):10.8f}')
#
# if (abs(np.trace(G0)) + abs(np.trace(G1)))>1e-12:
# print(f'Trace 0/1 = {np.trace(G0)}, {np.trace(G1)}')
# #print(f'P1={p1:10.4f} , P1check={p1_check:10.4f}, dP={dP:10.4f}/dPP={p1_check-p0_check:10.4f}, error={abs(p1-p1_check):10.4f}')
lnPbar = p1 - p
if lnaccept[ss] < lnPbar or lnPbar > 0:
A = A1
G = G1
p = np.copy(p1)
return G, A, p, p0