def create_node_mi(name, parents, minum, milb, miub, mibins):
node_mi = Node(name, parents=parents, rvname='continuous')
minames = node_mi.discretize(milb, miub, minum, infinity='+-', bins=mibins)
node_insp = parents[0]
node_ai = parents[1]
ainum = node_ai.nstates()
labels = itertools.product(np.arange(node_insp.nstates()), np.arange(ainum))
labels = [label for label in labels]
for i,label in enumerate(labels):
if label[0] == 0: #no inspection
probs = 1./minum * np.ones(minum)
else: # with insepction
if label[0] == 1:
lmd = lmd1; beta = beta1
elif label[0] == 2:
lmd = lmd2; beta = beta2
elif label[0] == 3:
lmd = lmd3; beta = beta3
rvnames = ['Ai']
rvs = [node_ai.truncate_rv(label[1], lmd=trunclmd)]
aimean = rvs[0].stats('m')[()]
rv_am = stats.norm(aimean, sigmae)
pod = 1.-stats.norm.cdf((np.log(aimean)-lmd)/beta)
probs = rv_am.cdf(node_mi.bins[1:])-rv_am.cdf(node_mi.bins[:-1])
probs = probs/np.sum(probs)*pod
probs[0] = probs[0]+(1.-pod)
node_mi.assign_cpt(probs,label=np.asarray(label),statenames=node_mi.statenames)
# print 'labels: {}, progress: {}%, prob: {}'.format(label,
# float(i)/len(labels)*100, np.array_str(probs,precision=3))
return node_mi
def create_node_a(name,
parents,
ainum,
ailb,
aiub,
aiedges,
node_repair=None,
asmp0=None):
if node_repair is None:
node_ai = Node(name, parents=parents, rvname='continuous')
else:
node_ai = Node(name,
parents=parents + [node_repair],
rvname='continuous')
# dynamic discretization of nodes a and M
node_ap = parents[0]
knum = parents[1].nstates()
mnum = parents[2].nstates()
ainames = node_ai.discretize(ailb, aiub, ainum, infinity='+', bins=aiedges)
aibins = node_ai.bins
if node_repair is None:
labels = itertools.product(np.arange(node_ap.nstates()),
np.arange(knum), np.arange(mnum))
else:
labels = itertools.product(np.arange(node_ap.nstates()),
np.arange(knum), np.arange(mnum),
np.arange(2))
labels = [label for label in labels]
for i, label in enumerate(labels):
if len(label) == 4 and label[-1] == 1:
binnum, dummy = np.histogram(asmp0, aibins)
probs = binnum / np.sum(binnum, dtype=float)
node_ai.assign_cpt(probs,
label=np.asarray(label),
statenames=node_ai.statenames)
else:
truncrvs = []
for j, pstate in enumerate(label):
truncrvs.append(node_ai.parents[j].truncate_rv(pstate,
lmd=trunclmd))
rvnames = ['Ap', 'K', 'M']
rvs = truncrvs[:3]
probs, smpdb = mc2ai(rvnames, rvs, node_ai.bins, acrit, nsmp=nsmp)
# clean Ai states given Ai-1
apstate = label[0]
aplb = node_ap.bins[apstate]
aiubs = node_ai.bins[1:]
probs[aiubs <= aplb] = 0.
probs = probs / np.sum(probs)
node_ai.assign_cpt(probs,
label=np.asarray(label),
statenames=node_ai.statenames)
# print 'labels: {}, progress: {}%, prob: {}'.format(label,
# float(i)/len(labels)*100, np.array_str(probs,precision=3))
return node_ai
def create_node_a(name, parents, ainum, ailb, aiub, aiedges, node_repair=None, asmp0=None):
if node_repair is None:
node_ai = Node(name, parents=parents, rvname='continuous')
else:
node_ai = Node(name, parents=parents+[node_repair], rvname='continuous')
# dynamic discretization of nodes a and M
node_ap = parents[0]
knum = parents[1].nstates()
mnum = parents[2].nstates()
ainames = node_ai.discretize(ailb, aiub, ainum, infinity='+', bins=aiedges)
aibins = node_ai.bins
if node_repair is None:
labels = itertools.product(np.arange(node_ap.nstates()), np.arange(knum),np.arange(mnum))
else:
labels = itertools.product(np.arange(node_ap.nstates()), np.arange(knum),
np.arange(mnum), np.arange(2))
labels = [label for label in labels]
for i,label in enumerate(labels):
if len(label)==4 and label[-1] == 1:
binnum,dummy = np.histogram(asmp0, aibins)
probs = binnum/np.sum(binnum, dtype=float)
node_ai.assign_cpt(probs,label=np.asarray(label),statenames=node_ai.statenames)
else:
truncrvs=[]
for j,pstate in enumerate(label):
truncrvs.append(node_ai.parents[j].truncate_rv(pstate,lmd=trunclmd))
rvnames = ['Ap', 'K', 'M']
rvs = truncrvs[:3]
probs,smpdb = mc2ai(rvnames, rvs, node_ai.bins, acrit, nsmp=nsmp)
# clean Ai states given Ai-1
apstate = label[0]
aplb = node_ap.bins[apstate]
aiubs = node_ai.bins[1:]
probs[aiubs<=aplb] = 0.
probs = probs/np.sum(probs)
node_ai.assign_cpt(probs,label=np.asarray(label),statenames=node_ai.statenames)
# print 'labels: {}, progress: {}%, prob: {}'.format(label,
# float(i)/len(labels)*100, np.array_str(probs,precision=3))
return node_ai
def create_node_mi(name, parents, minum, milb, miub, mibins):
node_mi = Node(name, parents=parents, rvname='continuous')
minames = node_mi.discretize(milb, miub, minum, infinity='+-', bins=mibins)
node_insp = parents[0]
node_ai = parents[1]
ainum = node_ai.nstates()
labels = itertools.product(np.arange(node_insp.nstates()),
np.arange(ainum))
labels = [label for label in labels]
for i, label in enumerate(labels):
if label[0] == 0: #no inspection
probs = 1. / minum * np.ones(minum)
else: # with insepction
if label[0] == 1:
lmd = lmd1
beta = beta1
elif label[0] == 2:
lmd = lmd2
beta = beta2
elif label[0] == 3:
lmd = lmd3
beta = beta3
rvnames = ['Ai']
rvs = [node_ai.truncate_rv(label[1], lmd=trunclmd)]
aimean = rvs[0].stats('m')[()]
rv_am = stats.norm(aimean, sigmae)
pod = 1. - stats.norm.cdf((np.log(aimean) - lmd) / beta)
probs = rv_am.cdf(node_mi.bins[1:]) - rv_am.cdf(node_mi.bins[:-1])
probs = probs / np.sum(probs) * pod
probs[0] = probs[0] + (1. - pod)
node_mi.assign_cpt(probs,
label=np.asarray(label),
statenames=node_mi.statenames)
# print 'labels: {}, progress: {}%, prob: {}'.format(label,
# float(i)/len(labels)*100, np.array_str(probs,precision=3))
return node_mi
knum = 20 + 1
ksmp_prior, msmp_prior = ksmp_mc(nsmp, rv_C, rv_Sre, G, rv_m, rv_Na)
klb = np.percentile(ksmp_prior, 5)
kub = np.percentile(ksmp_prior, 95)
if klb > 0:
kbins = np.hstack((0., np.linspace(klb, kub, knum - 1)))
else:
kbins = np.linspace(0, kub, knum)
knames = node_k.discretize(klb, kub, knum, infinity='+', bins=kbins)
# calculate and assign CPT
# node a0
a0cpt = node_a0.rv.cdf(node_a0.bins[1:]) - node_a0.rv.cdf(
node_a0.bins[:-1])
a0cpt = a0cpt[np.newaxis, :]
node_a0.assign_cpt(a0cpt, statenames=a0names)
# node m
mcpt = node_m.rv.cdf(node_m.bins[1:]) - node_m.rv.cdf(node_m.bins[:-1])
mcpt = mcpt[np.newaxis, :]
node_m.assign_cpt(mcpt, statenames=mnames)
# node k
nstate = knum
labels = itertools.product(np.arange(node_m.nstates()))
labels = [label for label in labels]
for i, label in enumerate(labels):
truncrvs = []
for j, pstate in enumerate(label):
truncrvs.append(node_k.parents[j].truncate_rv(pstate,
lmd=trunclmd))
rvnames = ['M', 'C', 'Sre', 'Na']
rvs = truncrvs + [rv_C, rv_Sre, rv_Na]
# node k
knum = 20+1
ksmp_prior,msmp_prior = ksmp_mc(nsmp, rv_C, rv_Sre, G, rv_m, rv_Na)
klb = np.percentile(ksmp_prior, 5)
kub = np.percentile(ksmp_prior, 95)
if klb>0:
kbins = np.hstack((0., np.linspace(klb, kub, knum-1)))
else:
kbins = np.linspace(0, kub, knum)
knames = node_k.discretize(klb, kub, knum, infinity='+', bins=kbins)
# calculate and assign CPT
# node a0
a0cpt = node_a0.rv.cdf(node_a0.bins[1:]) - node_a0.rv.cdf(node_a0.bins[:-1])
a0cpt = a0cpt[np.newaxis, :]
node_a0.assign_cpt(a0cpt, statenames=a0names)
# node m
mcpt = node_m.rv.cdf(node_m.bins[1:]) - node_m.rv.cdf(node_m.bins[:-1])
mcpt = mcpt[np.newaxis, :]
node_m.assign_cpt(mcpt, statenames=mnames)
# node k
nstate = knum
labels = itertools.product(np.arange(node_m.nstates()))
labels = [label for label in labels]
for i,label in enumerate(labels):
truncrvs = []
for j, pstate in enumerate(label):
truncrvs.append(node_k.parents[j].truncate_rv(pstate,lmd=trunclmd))
rvnames = ['M', 'C', 'Sre', 'Na']
rvs = truncrvs+[rv_C, rv_Sre, rv_Na]
# rvnames = ['M', 'C', 'Sre', 'Na']
import numpy as np
# create new net
netp = Network("ChestClinic")
# define nodes
visitAsia = Node("VisitAsia", parents=None, rvname='discrete')
smoking = Node("Smoking", parents=None, rvname='discrete')
tuberculosis = Node("Tuberculosis", parents=[visitAsia], rvname='discrete')
cancer = Node("Cancer", parents=[smoking], rvname='discrete')
tbOrCa = Node("TbOrCa", parents=[tuberculosis, cancer], rvname='discrete')
xRay = Node("XRay", parents=[tbOrCa], rvname='discrete')
# assign CPT
visitAsiaCpt = np.array([0.01, 0.99])[np.newaxis, :]
visitAsia.assign_cpt(visitAsiaCpt, statenames=['visit', 'no_visit'])
smokingCpt = np.array([0.50, 0.50])[np.newaxis, :]
smoking.assign_cpt(smokingCpt, statenames=['smoker', 'nonsmoker'])
tuberCpt = np.array([[0.05, 0.95], [0.01, 0.99]])
tuberculosis.assign_cpt(tuberCpt, statenames=['present', 'absent'])
cancerCpt = np.array([[0.10, 0.90], [0.01, 0.99]])
cancer.assign_cpt(cancerCpt, statenames=['present', 'absent'])
tbOrCaCpt = np.array([[1.0, 0.0], [1.0, 0.0], [1.0, 0.0], [0.0, 1.0]])
tbOrCa.assign_cpt(tbOrCaCpt, statenames=['true', 'false'])
xRayCpt = np.array([[0.98, 0.02], [0.05, 0.95]])
xRay.assign_cpt(xRayCpt, statenames=['abnormal', 'normal'])
# add node
netp.add_nodes([visitAsia])
netp.add_nodes([smoking])
netp.add_nodes([tuberculosis])
uhnum = 50+1
uhlb = 0.; uhub = 150.
uhbins = np.hstack(np.linspace(uhlb, uhub, uhnum))
uhnames = uh.discretize(uhlb, uhub, uhnum, infinity='+', bins=uhbins)
# h
for h in harray:
hnum = 30+1
hlb = 0.; hub = 150.
hbins = np.hstack(np.linspace(hlb, hub, hnum))
hnames = h.discretize(hlb, hub, hnum, infinity='+', bins=hbins)
# calculate and assignCPT
# e0
probs = np.array([[1., 0.]])
e0.assign_cpt(probs, statenames=['safe', 'fail'])
# node R5
r5cpt = r5.rv.cdf(r5.bins[1:]) - r5.rv.cdf(r5.bins[:-1])
r5cpt = r5cpt[np.newaxis,:]
r5.assign_cpt(r5cpt, statenames=r5names)
# node Uh
uhcpt = uh.rv.cdf(uh.bins[1:]) - uh.rv.cdf(uh.bins[:-1])
uhcpt = uhcpt[np.newaxis,:]
uh.assign_cpt(uhcpt, statenames=uhnames)
# node R4
nstate = r4num
labels = itertools.product(np.arange(r5.nstates()))
mvnorm = stats.multivariate_normal(mean=None,cov=np.array([[1.,rolnR],[rolnR,1.]]))
for i,label in enumerate(labels):
probs=[]
plb = r5.bins[label[0]] #r5 lower bound
import numpy as np
# create new net
netp = Network("ChestClinic")
# define nodes
visitAsia = Node("VisitAsia", parents=None, rvname='discrete')
smoking = Node("Smoking", parents=None, rvname='discrete')
tuberculosis = Node("Tuberculosis", parents=[visitAsia], rvname='discrete')
cancer = Node("Cancer", parents=[smoking], rvname='discrete')
tbOrCa = Node("TbOrCa", parents=[tuberculosis, cancer], rvname='discrete')
xRay = Node("XRay", parents=[tbOrCa], rvname='discrete')
# assign CPT
visitAsiaCpt = np.array([0.01, 0.99])[np.newaxis, :]
visitAsia.assign_cpt(visitAsiaCpt, statenames=['visit','no_visit'])
smokingCpt = np.array([0.50, 0.50])[np.newaxis, :]
smoking.assign_cpt(smokingCpt, statenames=['smoker','nonsmoker'])
tuberCpt = np.array([[0.05, 0.95], [0.01, 0.99]])
tuberculosis.assign_cpt(tuberCpt, statenames=['present','absent'])
cancerCpt = np.array([[0.10, 0.90], [0.01, 0.99]])
cancer.assign_cpt(cancerCpt, statenames=['present','absent'])
tbOrCaCpt = np.array([[1.0, 0.0], [1.0, 0.0], [1.0, 0.0], [0.0, 1.0]])
tbOrCa.assign_cpt(tbOrCaCpt, statenames=['true','false'])
xRayCpt = np.array([[0.98, 0.02], [0.05, 0.95]])
xRay.assign_cpt(xRayCpt, statenames=['abnormal','normal'])
# add node
netp.add_nodes([visitAsia])
netp.add_nodes([smoking])
netp.add_nodes([tuberculosis])
m5num = 20+2
m = r5.rv.stats('m'); s = np.sqrt(r5.rv.stats('v'))
lb = 50.; ub = 250.
r4bins = np.hstack((0, np.linspace(lb, ub, r4num-1)))
r4names = r4.discretize(lb, ub, r4num, infinity='+', bins=r4bins)
m4names = m4.discretize(lb, ub, m4num, infinity='+-', bins=r4bins)
lb = 50.; ub = 250.
r5bins = np.hstack((0, np.linspace(lb, ub, r5num-1)))
r5names = r5.discretize(lb, ub, r5num, infinity='+', bins=r5bins)
m5names = m5.discretize(lb, ub, m5num, infinity='+-', bins=r5bins)
# calculate and assignCPT
# node R5
r5cpt = r5.rv.cdf(r5.bins[1:]) - r5.rv.cdf(r5.bins[:-1])
r5cpt = r5cpt[np.newaxis,:]
r5.assign_cpt(r5cpt, statenames=r5names)
# node R4
nstate = r4num
labels = itertools.product(np.arange(r5.nstates()))
mvnorm = stats.multivariate_normal(mean=None,cov=np.array([[1.,rolnR],[rolnR,1.]]))
for i,label in enumerate(labels):
probs=[]
plb = r5.bins[label[0]] #r5 lower bound
pub = r5.bins[label[0]+1] #r5 upper bound
z5lb = (np.log(plb)-logmean)/logstd
z5ub = (np.log(pub)-logmean)/logstd
for k in xrange(nstate):
clb = r4.bins[k] #r4 lower bound
cub = r4.bins[k+1] #r4 upper bound
z4lb = (np.log(clb)-logmean)/logstd
z4ub = (np.log(cub)-logmean)/logstd
# random variables
rvX1 = stats.lognorm(1., scale=np.exp(0))
rvX3 = stats.lognorm(1., scale=np.exp(3*np.sqrt(2)))
# create nodes
weather = Node("Weather", parents=None, rvname='discrete')
forecast = Node("Forecast", parents=[weather], rvname='discrete')
umbrella = Node("Umbrella", parents=[forecast], rvname='discrete')
satisfy = Node("Satisfaction", parents=[weather, umbrella], rvname='deterministic')
umbrella.set_node_kind(DECISION_NODE)
satisfy.set_node_kind(UTILITY_NODE)
# assign CPT
# node weather
weathercpt = np.array([[0.7, 0.3]])
weather.assign_cpt(weathercpt, statenames=['sunshine', 'rain'])
# node forecast
forecastcpt = np.array([[0.7, 0.2, 0.1], [0.15, 0.25, 0.6]])
forecast.assign_cpt(forecastcpt, statenames=['sunny', 'cloudy', 'rainy'])
# node umbrella
umbrella.set_node_state_name(['take_umbrella', 'dont_take_umbrella'])
# node satisfy
def calculate_util(pstate):
forecaststate,umbrellastate = pstate
if forecaststate==0 and umbrellastate==0:
return 20.
elif forecaststate==0 and umbrellastate==1:
return 100.
elif forecaststate==1 and umbrellastate==0:
return 70.
elif forecaststate==1 and umbrellastate==1:
uhlb = 0.
uhub = 150.
uhbins = np.hstack(np.linspace(uhlb, uhub, uhnum))
uhnames = uh.discretize(uhlb, uhub, uhnum, infinity='+', bins=uhbins)
# h
for h in harray:
hnum = 30 + 1
hlb = 0.
hub = 150.
hbins = np.hstack(np.linspace(hlb, hub, hnum))
hnames = h.discretize(hlb, hub, hnum, infinity='+', bins=hbins)
# calculate and assignCPT
# e0
probs = np.array([[1., 0.]])
e0.assign_cpt(probs, statenames=['safe', 'fail'])
# node R5
r5cpt = r5.rv.cdf(r5.bins[1:]) - r5.rv.cdf(r5.bins[:-1])
r5cpt = r5cpt[np.newaxis, :]
r5.assign_cpt(r5cpt, statenames=r5names)
# node Uh
uhcpt = uh.rv.cdf(uh.bins[1:]) - uh.rv.cdf(uh.bins[:-1])
uhcpt = uhcpt[np.newaxis, :]
uh.assign_cpt(uhcpt, statenames=uhnames)
# node R4
nstate = r4num
labels = itertools.product(np.arange(r5.nstates()))
mvnorm = stats.multivariate_normal(mean=None,
cov=np.array([[1., rolnR], [rolnR,
1.]]))
for i, label in enumerate(labels):
y = Node("Y", parents=[x2], rvname='discrete')
# discretize continuous rv
x1num = 5
x2num = 5
m = x1.rv.stats('m'); s = np.sqrt(x1.rv.stats('v'))
#lb = np.maximum(0, m-2*s); ub = m+2*s
lb = 0.; ub = m+1.5*s
x1names = x1.discretize(lb, ub, x1num, infinity='+')
x2names = x2.discretize(lb*10., ub*10., x2num, infinity='+')
# calculate and assign CPT
# node X1
x1cpt = x1.rv.cdf(x1.bins[1:]) - x1.rv.cdf(x1.bins[:-1])
x1cpt = x1cpt[np.newaxis,:]
x1.assign_cpt(x1cpt, statenames=x1names)
# node X2
lmd = 0.05
nstate = x2num
labels = itertools.product(np.arange(x1.nstates()))
for i, label in enumerate(labels):
rvs = []
probs=[]
for j, pstate in enumerate(label):
rvs.append(x2.parents[j].truncate_rv(pstate))
for k in xrange(nstate):
probs.append(rvs[0].cdf(x2.bins[k+1]/10.) -
rvs[0].cdf(x2.bins[k]/10.))
probs = np.asarray(probs)
x2.assign_cpt(probs,label=np.asarray(label),statenames=x2names)
# node Y
lb = 50.
ub = 250.
r4bins = np.hstack((0, np.linspace(lb, ub, r4num - 1)))
r4names = r4.discretize(lb, ub, r4num, infinity='+', bins=r4bins)
m4names = m4.discretize(lb, ub, m4num, infinity='+-', bins=r4bins)
lb = 50.
ub = 250.
r5bins = np.hstack((0, np.linspace(lb, ub, r5num - 1)))
r5names = r5.discretize(lb, ub, r5num, infinity='+', bins=r5bins)
m5names = m5.discretize(lb, ub, m5num, infinity='+-', bins=r5bins)
# calculate and assignCPT
# node R5
r5cpt = r5.rv.cdf(r5.bins[1:]) - r5.rv.cdf(r5.bins[:-1])
r5cpt = r5cpt[np.newaxis, :]
r5.assign_cpt(r5cpt, statenames=r5names)
# node R4
nstate = r4num
labels = itertools.product(np.arange(r5.nstates()))
mvnorm = stats.multivariate_normal(mean=None,
cov=np.array([[1., rolnR], [rolnR,
1.]]))
for i, label in enumerate(labels):
probs = []
plb = r5.bins[label[0]] #r5 lower bound
pub = r5.bins[label[0] + 1] #r5 upper bound
z5lb = (np.log(plb) - logmean) / logstd
z5ub = (np.log(pub) - logmean) / logstd
for k in xrange(nstate):
clb = r4.bins[k] #r4 lower bound
cub = r4.bins[k + 1] #r4 upper bound
rvX3 = stats.lognorm(1., scale=np.exp(3 * np.sqrt(2)))
# create nodes
weather = Node("Weather", parents=None, rvname='discrete')
forecast = Node("Forecast", parents=[weather], rvname='discrete')
umbrella = Node("Umbrella", parents=[forecast], rvname='discrete')
satisfy = Node("Satisfaction",
parents=[weather, umbrella],
rvname='deterministic')
umbrella.set_node_kind(DECISION_NODE)
satisfy.set_node_kind(UTILITY_NODE)
# assign CPT
# node weather
weathercpt = np.array([[0.7, 0.3]])
weather.assign_cpt(weathercpt, statenames=['sunshine', 'rain'])
# node forecast
forecastcpt = np.array([[0.7, 0.2, 0.1], [0.15, 0.25, 0.6]])
forecast.assign_cpt(forecastcpt, statenames=['sunny', 'cloudy', 'rainy'])
# node umbrella
umbrella.set_node_state_name(['take_umbrella', 'dont_take_umbrella'])
# node satisfy
def calculate_util(pstate):
forecaststate, umbrellastate = pstate
if forecaststate == 0 and umbrellastate == 0:
return 20.
elif forecaststate == 0 and umbrellastate == 1:
return 100.
elif forecaststate == 1 and umbrellastate == 0:
