template = sampleTemplate(groundgraph, numtemplate)
if 'WORKHASH' in os.environ:
jobhash = os.environ['WORKHASH']
if not r.hexists('jobs:grounds', jobhash):
r.hset('jobs:grounds', jobhash,
zlib.compress(cPickle.dumps(groundgraph)))
random.seed()
np.random.seed()
datasizes = [4, 16, 32, 64, 128, 256]
temps = [1.0, 1.0, 2.0, 2.0, 5.0, 5.0]
for temperature, numdata in zip(temps, datasizes):
data = generateData(groundgraph, joint, numdata)
groundbnet = BayesNetCPD(states, data, limparent=3)
groundbnet.set_cpds(joint)
obj = BayesNetCPD(states, data, limparent=3)
b = BayesNetSampler(obj, template, groundbnet, priorweight)
s = SAMCRun(b, burn, stepscale, refden, thin)
s.sample(iters, temperature)
s.compute_means()
if 'WORKHASH' in os.environ:
r.lpush('jobs:done:' + jobhash, s.read_db())
r.lpush('custom:%s:samplesize=%d' % (jobhash, numdata),
s.db.root.computed.means._v_attrs['kld'])
s.db.close()
#time()
#s1.sample(iters, temperature)
#time()
#s1.compute_means()
#if 'WORKHASH' in os.environ:
#r.lpush('jobs:done:' + jobhash, s1.read_db())
#s1.db.close()
############# bayesnetcpd ############
#import pstats, cProfile
joint = utils.graph_to_joint(groundgraph)
states = np.ones(len(joint.dists),dtype=np.int32)*2
groundbnet = BayesNetCPD(states, data)
groundbnet.set_cpds(joint)
obj = BayesNetCPD(states, data)
b2 = BayesNetSampler(obj, template, groundbnet, priorweight)
s2 = SAMCRun(b2,burn,stepscale,refden,thin)
time()
#cProfile.runctx("s2.sample(iters, temperature)", globals(), locals(), "prof.prof")
s2.sample(iters, temperature)
time()
s2.compute_means()
#s2.compute_means(cummeans=False)
if 'WORKHASH' in os.environ:
r.lpush('jobs:done:' + jobhash, s2.read_db())
s2.db.close()
#######################################
random.seed()
np.random.seed()
############### TreeNet ##############
groundtree = TreeNet(N, data=data, graph=groundgraph)
b1 = TreeNet(N, data, template, priorweight, groundtree)
#s1 = SAMCRun(b1,burn,stepscale,refden,thin)
#s1.sample(iters, temperature)
############## bayesnetcpd ############
joint = utils.graph_to_joint(groundgraph)
states = np.ones(len(joint.dists),dtype=np.int32)*2
groundbnet = BayesNetCPD(states, data)
groundbnet.set_cpds(joint)
obj = BayesNetCPD(states, data)
b2 = BayesNetSampler(obj, template, groundbnet, priorweight)
#s2 = SAMCRun(b2,burn,stepscale,refden,thin)
#s2.sample(iters, temperature)
#######################################
def test():
def close_enough(n1,n2):
return (n1-n2) < 1e-5
return (n1-n2) < np.finfo(float).eps
assert close_enough(groundtree.kld(b1), groundbnet.kld(b2.bayesnet))
random.seed()
np.random.seed()
############### TreeNet ##############
groundtree = TreeNet(N, data=data, graph=groundgraph)
b1 = TreeNet(N, data, template, priorweight, groundtree)
#s1 = SAMCRun(b1,burn,stepscale,refden,thin)
#s1.sample(iters, temperature)
############## bayesnetcpd ############
joint = utils.graph_to_joint(groundgraph)
states = np.ones(len(joint.dists), dtype=np.int32) * 2
groundbnet = BayesNetCPD(states, data)
groundbnet.set_cpds(joint)
obj = BayesNetCPD(states, data)
b2 = BayesNetSampler(obj, template, groundbnet, priorweight)
#s2 = SAMCRun(b2,burn,stepscale,refden,thin)
#s2.sample(iters, temperature)
#######################################
def test():
def close_enough(n1, n2):
return (n1 - n2) < 1e-5
return (n1 - n2) < np.finfo(float).eps
random.seed(123456) np.random.seed(123456) groundgraph = generateHourGlassGraph(nodes=N) #joint, states = generateJoint(groundgraph, method='dirichlet') joint, states = generateJoint(groundgraph, method='noisylogic') data = generateData(groundgraph, joint, numdata) template = sampleTemplate(groundgraph, numtemplate) print "Joint:" print joint random.seed() np.random.seed() ground = BayesNetCPD(states, data) ground.set_cpds(joint) obj = BayesNetCPD(states, data) b = BayesNetSampler(obj, template, ground, priorweight) s = SAMCRun(b,burn,stepscale,refden,thin) time() s.sample(iters, temperature) time() s.compute_means() #fname = '/tmp/test.h5' #fid = open(fname, 'w') #fid.write(zlib.decompress(s.read_db()))