def _run(self):
# supress output
#so = file('/dev/null', 'a+')
#stdout = os.dup(sys.stdout.fileno())
#os.dup2(so.fileno(), sys.stdout.fileno())
self.max_score = -1
attrs_selected = self.attrs_selected = []
attrs_left = range(self.num_attr)
cls_node = self.num_attr
_stop = self._stop
while len(attrs_selected) < self.num_attr and not _stop():
pick = -1
#p = 0.3
for i,a in enumerate(attrs_left):
tmp = attrs_selected + [a, cls_node]
tmp.sort()
#local_cpd_cache_ = SharedLocalCPDCache(self._cpd_cache._cpd_cache, tmp)
score = cross_validate(self.data.subset(tmp),
#local_cpd_cache=local_cpd_cache_,
score_type='WC',
runs=10)
if score > self.max_score:
self.max_score = score
pick = i
#if random() < p: break
if pick == -1: break
attrs_selected.append(attrs_left.pop(pick))
attrs_selected.sort()
attrs_selected.append(cls_node)
#self.attrs_selected = attrs_selected
tan_learner = TANClassifierLearner(self.data.subset(attrs_selected),
local_cpd_cache=SharedLocalCPDCache(self._cpd_cache._cpd_cache, attrs_selected))
tan_learner.run()
self.network = tan_learner.network
self.cpd = tan_learner.cpd
self._learner = tan_learner
self.result.add_network(self.network, self.max_score)
def _crossValidateScoreFunc(self, subset_idx, **cvargs):
data = self.data.subset(subset_idx)
return cross_validate(data, classifier_type=self.classifier_type, **cvargs)
parser.set_defaults(classifier_type="tan")
parser.add_option("-r", dest="test_ratio", type="float", help="Ratio of test data")
parser.set_defaults(test_ratio=0.3)
parser.add_option("-t", dest="runs", type="int", help="Number of runs")
parser.set_defaults(runs=10)
parser.add_option("-d", dest="numbins", type="int", help="Number of bins in discretizing")
parser.set_defaults(numbins=0)
parser.add_option("-v", dest="verbose", action="store_true", help="Report verbosely")
parser.set_defaults(verbose=False)
parser.add_option("-s", dest="score_type", help="Score type")
parser.set_defaults(score_type="BA")
(options, args) = parser.parse_args()
if len(args) < 1:
print "You did't specify a data file"
print
sys.exit(1)
else:
datafile = args[0]
dataset = data.fromfile(datafile)
if options.numbins:
dataset.discretize(numbins=options.numbins, excludevars=[dataset.variables.size-1])
cross_validate(dataset, options.classifier_type, options.test_ratio, options.runs, options.verbose, options.score_type)
