def run(self):
dataset_dir = self.pass_in
output_dir = self.pass_out
for root, dirs, files in os.walk(dataset_dir):
for fname in files:
matchObj = re.match(r"(train\d+).*", fname)
if matchObj is None:
continue
dataset = dpDataset.load(os.path.join(dataset_dir, fname))
# The returned data structure will be in the following
# format:
# [effective feature set for a class] * 4
# and the each feature set is again
# [user_id] * 10
e = Effective(dataset).getFeatureList(n_count=10)
out_prefix = matchObj.group(1)
out_fname = out_prefix + "_effectives.yaml"
yout = open(os.path.join(output_dir, out_fname), "w")
yout.write(yaml.dump(e))
yout.close()
print "Finished %s" % out_fname
def trainFrom(self, fname):
'''
@param fname the file which includes feature dataset from which we
will build a trained maxent model and return
'''
dataset = dpDataset.load(fname)
trainer = meSimple.METrainer()
model = trainer.trainedModelOn(dataset)
return model
def getSeeds():
fname = '../feature_set2/ver2.8-efollowing.libsvm'
dataset = dp_dataset.load(fname)
dataset = dp_dataset.shuffle(dataset)
e = Effective(dataset)
nested_seeds = e.getFeatureList(n_count = 10)
def _flatten(l):
for el in l:
if (isinstance(el, collections.Iterable) and
not isinstance(el, basestring)):
for sub in _flatten(el):
yield sub
else:
yield el
print nested_seeds
return list(set([x for x in _flatten(nested_seeds)]))
def getFeatureList(self, n_count = 10):
'''
Return a list of n_count effective features for each class. Therefore,
the total number of returned features will be n_count * n_class.
'''
dataset = self.dataset
m = self.trainedModelOn(dataset)
features = self._extractFeatures(dataset)
e_features = self._mostEffective(m, features, n_count)
# you might want to call reduce(lambda x, y: x + y, effectives) to
# flatten the returned value
return e_features
if __name__ == '__main__':
fname = '../feature_set2/ver2.8-etext.libsvm'
dataset = dp_dataset.load(fname)
e = Effective(dataset)
print e.getFeatureList(n_count = 10)
