def main(bdtfile, ess=1.0, outfile=None, constraint_file=""):
bn, sc = bnsearch.empty_net(bdtfile, ess)
bestforests.kruskal(bn, sc)
sc.score_new(bn)
print sc.score()
for forest in bestforests.Forest(bn):
sc.score_new(forest)
print sc.score()
def main(bdtfile, ess=1.0, outfile=None):
bn, sc = bnsearch.empty_net(bdtfile, ess)
if outfile:
bn.save(outfile)
print sc.score()
def main(bdtfile, scoretype='BDeu', ess=1.0, outfile=None, cachefile=None):
bn, sc = bnsearch.empty_net(bdtfile, scoretype, ess, cachefile=cachefile)
if outfile:
bn.save(outfile)
print sc.score()
def main(bdtfile,
scoretype='BDeu',
ess=1.0,
time=None,
iters=None,
outfile=None,
constraint_file="",
startbn=None,
cachefile=None):
global searchtime
cstrs = Constraints(constraint_file)
if startbn != None:
bn = bnmodule.load(startbn, do_pic=False)
sc = scorefactory.getscorer(bdtfile,
scoretype,
ess,
cachefile=cachefile)
else:
bn, sc = bnsearch.empty_net(bdtfile,
scoretype,
ess,
cachefile=cachefile)
if constraint_file: # should check if compatible with start
for a in cstrs.must:
bn.addarc(a)
sc.score_new(bn)
bn.picall()
searchtime = sigpool.str2time(time)
endtime = tim.time() + searchtime
t0 = find_initial_temperature(bn, sc)
isss = {
"searchtime": searchtime,
"endtime": endtime,
"t0": t0,
"t": t0
}
isss.update(bnsearch.initial_search_status(bn, sc))
bnsearch.localsearch(isss, sa_step, sa_stop)
if outfile:
isss["best_bn"].save(outfile)
print isss["best_score"]
def main(bdtfile,
scoretype='BDeu',
ess=1.0,
outfile=None,
constraint_file="",
cachefile=None):
bn, sc = bnsearch.empty_net(bdtfile,
scoretype,
ess,
cachefile=cachefile)
cstrs = constraints.Constraints(constraint_file)
kruskal(bn, sc, cstrs)
if outfile:
bn.save(outfile)
sc.score_new(bn)
print sc.score()
def main(bdtfile,
scoretype='BDeu',
ess=1.0,
cachefile=None,
bnfile=None,
direction='up',
constraint_file=""):
cstrs = Constraints(constraint_file)
if bnfile == None:
bns, scr = empty_net(bdtfile, scoretype, ess, cachefile=cachefile)
else:
bns = bn.bn.load(bnfile, do_pic=True)
bdt = data.Data(bdtfile)
scr = scorefactory.getscorer(bdt, scoretype, ess, cachefile=cachefile)
scr.score_new(bns)
updown = {
'up': {
'changef': bns.addarc,
'cancelf': bns.delarc,
'tabuset': cstrs.no,
'candarcs': bns.new_dagarcs,
'actname': 'add'
},
'down': {
'changef': bns.delarc,
'cancelf': bns.addarc,
'tabuset': cstrs.must,
'candarcs': bns.arcs,
'actname': 'del'
}
}
print 'init', -1, -1, scr.score()
params = updown[direction]
for sdiff, arc in gsteps(bns, scr, params):
scr.score_new(bns) # needed?
print params['actname'], arc[0], arc[1], scr.score()
def main(bdtfile,
scoretype='BDeu',
ess=1.0,
time=None,
iters=None,
outfile=None,
constraint_file="",
startbn=None,
cachefile=None):
sigpool.watch('SIGUSR2')
sigpool.watch('SIGUSR1')
if time:
sigpool.wait_n_raise(sigpool.str2time(time), 'SIGUSR2')
cstrs = Constraints(constraint_file)
if startbn != None:
bn = bnmodule.load(startbn, do_pic=False)
sc = scorefactory.getscorer(bdtfile,
scoretype,
ess,
cachefile=cachefile)
forests_left = False
else:
bn, sc = bnsearch.empty_net(bdtfile,
scoretype,
ess,
cachefile=cachefile)
bestforests.kruskal(bn, sc, cstrs)
fry = bestforests.Forest(bn)
forests_left = True
bn = fry.next()
if constraint_file: # should check if compatible with start
for a in cstrs.must:
bn.addarc(a)
sc.score_new(bn)
bn.picall()
good_nets = [(sc.score(), bn.copy())]
t = 0L
while True:
greedysearch.greedysearch(bn, sc, 1000, cstrs)
gs = sc.score()
t += 1
if gs > good_nets[0][0] and bn not in [gn for (sn, gn) in good_nets]:
good_nets.append((gs, bn.copy()))
good_nets.sort()
if len(good_nets) > 2 * 10:
good_nets = good_nets[10:]
start_from_forest = random.choice((0, 1)) # stupid if out of forests
if forests_left and start_from_forest:
try:
bn = fry.next()
bn.picall()
except StopIteration:
forests_left = False
if (not start_from_forest) or (not forests_left):
bn = random.choice(good_nets)[1].copy()
# bn = good_nets[0][1].copy()
sc.score_new(bn)
sas = bnsearch.score_arcs(bn, sc)
sas.reverse()
eas = list(enumerate(sas))
for x in xrange(len(sas) / 2):
i, n, sa = wheelselect(eas)
ii, (s, a) = eas.pop(i)
if not a in cstrs.must:
#print 'DEL', a
bn.delarc(a)
#print 'ADEL', bn.arcs()
#for v in bn.vars(): print v, bn.path_in_counts[v]
sc.score_new(bn)
if (iters and t > iters): break
if 'SIGUSR2' in sigpool.flags: break
if 'SIGUSR1' in sigpool.flags:
if outfile: good_nets[-1][1].save(outfile)
print good_nets[-1][0]
sigpool.flags.remove('SIGUSR1')
if outfile:
good_nets[-1][1].save(outfile)
print good_nets[-1][0]
def main(bdtfile,
ess=1.0, time=None, iters=None, outfile=None, constraint_file=""):
sigpool.raise_on_signal(signal.SIGUSR2)
if time:
sigpool.wait_n_raise(sigpool.str2time(time), signal.SIGUSR2)
bn, sc = bnsearch.empty_net(bdtfile, ess)
if constraint_file:
cstrs = Constraints(constraint_file)
forests_left = False
for a in cstrs.must:
bn.addarc(a)
else:
cstrs = Constraints()
bestforests.kruskal(bn,sc)
bestforests.prune_1(bn,sc)
fry = bestforests.forests(bn)
forests_left = True
bn = fry.next()
sc.score_new(bn)
good_nets = [(sc.score(),bn.copy())]
t = 0L
while True:
greedysearch.greedysearch(bn, sc, 1000, cstrs)
gs = sc.score()
t += 1
if gs > good_nets[0][0] and bn not in [gn for (sn,gn) in good_nets]:
good_nets.append((gs,bn.copy(False)))
good_nets.sort()
if len(good_nets) > 2 * 10 :
good_nets = good_nets[10:]
start_from_forest = random.choice((0,1)) # stupid if out of forests
if forests_left and start_from_forest:
try:
bn = fry.next()
except StopIteration:
forests_left = False
if (not start_from_forest) or (not forests_left):
bn = random.choice(good_nets)[1].copy(False)
# bn = good_nets[0][1].copy()
sc.score_new(bn)
arcs = bnsearch.score_arcs(bn,sc)
arcs.reverse()
eas = list(enumerate(arcs))
for x in xrange(len(arcs) / 2) :
i, n, sa = wheelselect(eas)
ii, (s,a) = eas.pop(i)
if not a in cstrs.must: bn.delarc(a)
sc.score_new(bn)
if (iters and t > iters): break
if signal.SIGUSR2 in sigpool.flags: break
if outfile:
good_nets[-1][1].save(outfile)
print good_nets[-1][0]