def computeMI_cond(model, node, minDist, maxDist, neighbours_G, snapshots,
nTrials, nSamples, modelSettings):
MIs = []
HXs = []
subgraph_nodes = [node]
for d in range(1, maxDist + 1):
# get subgraph and outer neighbourhood at distance d
if len(neighbours_G[d]) > 0:
subgraph = nx.ego_graph(model.graph, node, d)
if d >= minDist:
print(
f'------------------- distance d={d}, num neighbours = {len(neighbours_G[d])}, subgraph size = {len(subgraph)}, num states = {len(snapshots[d-1])} -----------------------'
)
model_subgraph = fastIsing.Ising(subgraph, **modelSettings)
threads = nthreads if len(subgraph_nodes) > 20 else 1
initState = 1 #if args.fixMag else -1
_, _, MI, HX, HXgiveny, keys, probs = simulation.neighbourhoodMI(model_subgraph, node, \
d, neighbours_G, snapshots[d-1], nTrials, \
args.burninSteps, nSamples, \
args.distSamples, threads=threads, \
initStateIdx=initState, uniformPDF=1, out='MI')
print(HXgiveny)
MIs.append(MI)
HXs.append(HX)
print(MIs)
return MIs, HXs
def computeMI_cond(model, node, minDist, maxDist, neighbours_G, snapshots,
nTrials, nSamples, modelSettings, corrTimeSettings):
all_MI = np.full(maxDist - minDist + 1, np.nan)
all_HX = np.full(maxDist - minDist + 1, np.nan)
stop = False
subgraph_nodes = [node]
for idx, d in enumerate(minDist, maxDist + 1):
# get subgraph and outer neighbourhood at distance d
if len(neighbours_G[d]) > 0:
subgraph = nx.ego_graph(model.graph, node, d)
print(f'------------------- distance d={d}, num neighbours = \
{len(neighbours_G[d])}, subgraph size = {len(subgraph)}, \
num states = {len(snapshots[d-1])} -----------------------')
model_subgraph = fastIsing.Ising(subgraph, **modelSettings)
# determine correlation time for subgraph Ising model
if args.maxCorrTime == args.minCorrTime:
distSamples_subgraph = args.maxCorrTime
mixingTime_subgraph = corrTimeSettings['burninSteps']
else:
mixingTime_subgraph, meanMag, distSamples_subgraph, _ = \
simulation.determineCorrTime(model_subgraph, nodeG=node, **corrTimeSettings)
if args.maxCorrTime > 0:
distSamples_subgraph = min(distSamples_subgraph,
args.maxCorrTime)
distSamples_subgraph = max(distSamples_subgraph,
args.minCorrTime)
print(f'correlation time = {distSamples_subgraph}')
print(f'mixing time = {mixingTime_subgraph}')
threads = nthreads if len(subgraph) > 20 else 1
_, _, MI, HX, HXgiveny, keys, probs = \
simulation.neighbourhoodMI(model_subgraph, node, \
d, neighbours_G, snapshots[d-1], nTrials=nTrials, \
burninSteps=mixingTime_subgraph, nSamples=nSamples, \
distSamples=distSamples_subgraph, threads=threads, \
initStateIdx=args.initState, uniformPDF=args.uniformPDF, out='MI')
all_MI[idx] = MI
all_HX[idx] = HX
if MI < args.threshold: break
return all_MI, all_HX
graph = nx.read_gpickle(args.graph)
N = len(graph)
maxDist = args.maxDist if args.maxDist > 0 else nx.diameter(graph)
networkSettings = dict( \
path = args.graph, \
size = N, \
node = node
)
# setup Ising model with N=networkSize spin flip attempts per simulation step
modelSettings = dict( \
temperature = T, \
updateType = 'async' ,\
magSide = args.magSide if args.magSide in ['pos', 'neg'] else ''
)
model = fastIsing.Ising(graph, **modelSettings)
try:
mixingResults = IO.loadResults(targetDirectory, 'mixingResults')
corrTimeSettings = IO.loadResults(targetDirectory, 'corrTimeSettings')
burninSteps = mixingResults['burninSteps']
distSamples = mixingResults['distSamples']
except:
subprocess.call(['python3', 'run_mixing.py', f'{T}', f'{args.dir}', f'{args.graph}', \
'--maxcorrtime', '10000', \
'--maxmixing', '10000'])
mixingResults = IO.loadResults(targetDirectory, 'mixingResults')
corrTimeSettings = IO.loadResults(targetDirectory, 'corrTimeSettings')
burninSteps = mixingResults['burninSteps']
distSamples = mixingResults['distSamples']
burninSteps = burninSteps, \
updateMethod = updateType
)
IO.saveSettings(targetDirectory, settings)
for i, g in enumerate(ensemble):
graph = nx.read_gpickle(g)
filename = os.path.split(g)[-1].strip('.gpickle')
modelSettings = dict(\
graph = graph,\
updateType = updateType,\
magSide = magSide
)
model = fastIsing.Ising(**modelSettings)
Tc = Tc_idx = -1
while Tc < 0:
mags, sus, binder, abs_mags = simulation.magnetizationParallel(model, \
temps = temps, \
n = nSamples, \
burninSteps = burninSteps)
Tc, Tc_idx = find_Tc_gaussian(sus, binder, temps)
if Tc < 0:
print(f'failed to estimate T_c. Another estimation will be performed, using a larger temperature range')
temps = np.linspace(args.minT, temps[-1] + args.maxT, args.numT)
#a, b = optimize.curve_fit(sig, temps, abs_mags)
#mag_Tc = sig(Tc, *a)
from scipy.ndimage import gaussian_filter1d
nudges = np.logspace(-1, 1, 10)
out = np.zeros((N, nudges.size, deltas))
mis = np.zeros((N, graph.number_of_nodes(), deltas))
fig, ax = plt.subplots()
tax = ax.twinx()
modelsettings = dict(\
graph = graph, \
updateType = 'single', \
nudgeType = 'constant',\
magSide = 'neg')
m = fastIsing.Ising(\
**modelsettings)
temps = np.logspace(-6, np.log10(20), 30)
samps = [m.matchMagnetization(temps, 100) for i in range(1)]
samps = np.array(samps)
samps = gaussian_filter1d(samps, 3, axis=-1)
samps = samps.mean(0)
mag, sus = samps
ax.plot(temps, mag)
tax.plot(temps, sus)
fig.show()
IDX = np.argmin(abs(mag - .8 * mag.max()))
# IDX = np.argmax(sus)
m.t = temps[IDX]