def __init__(self, relations, bag_size=40):
assert (isinstance(bag_size, int)
and bag_size > 0) # relations from relationsCollection
assert (isinstance(relations,
list)) # relations from relationsCollection
self.bags = []
self.bag_size = bag_size
pos_bag_ind = self._add_bag(PositiveLabel())
neg_bag_ind = self._add_bag(NegativeLabel())
neu_bag_ind = self._add_bag(NeutralLabel())
for r in relations:
assert (isinstance(r, ExtractedRelation))
self._optional_add_in_bag(pos_bag_ind, r.opinion_vector,
r.text_position, r.label)
self._optional_add_in_bag(neg_bag_ind, r.opinion_vector,
r.text_position, r.label)
self._optional_add_in_bag(neu_bag_ind, r.opinion_vector,
r.text_position, r.label)
if len(self.bags[pos_bag_ind]) == bag_size:
pos_bag_ind = self._add_bag(PositiveLabel())
if len(self.bags[neg_bag_ind]) == bag_size:
neg_bag_ind = self._add_bag(NegativeLabel())
if len(self.bags[neu_bag_ind]) == bag_size:
neu_bag_ind = self._add_bag(NeutralLabel())
def __init__(self, synonyms_filepath, user_answers_filepath, stemmer):
assert(isinstance(stemmer, Stemmer)) # for opinion collections
self.synonyms_filepath = synonyms_filepath
self.user_answers = user_answers_filepath
self.stemmer = stemmer
self.pos = PositiveLabel()
self.neg = NegativeLabel()
self.neu = NeutralLabel()
def from_file(cls, filepath, synonyms):
assert (isinstance(filepath, str))
assert (isinstance(synonyms, SynonymsCollection))
instance = cls(synonyms=synonyms)
with open(filepath, 'r') as f:
it = cls.iter_line_params(f)
for args in tqdm(it, desc="Init BasePrinter from file"):
pos_count, neg_count, source_id, target_id, _ = args
value_left = synonyms.get_group_by_index(int(source_id))[0]
value_right = synonyms.get_group_by_index(int(target_id))[0]
pos_opinion = Opinion(value_left=value_left,
value_right=value_right,
sentiment=PositiveLabel())
neg_opinion = Opinion(value_left=value_left,
value_right=value_right,
sentiment=NegativeLabel())
if pos_count > 0:
instance.register_extracted_opinion(pos_opinion,
count=pos_count)
if neg_count > 0:
instance.register_extracted_opinion(neg_opinion,
count=neg_count)
return instance
def debug_statistic(self):
labels_count = [0, 0, 0]
for r in self.relations:
labels_count[r.label.to_uint()] += 1
return {
'pos': labels_count[PositiveLabel().to_uint()],
'neg': labels_count[NegativeLabel().to_uint()],
'neu': labels_count[NeutralLabel().to_uint()]
}
def optional_invert_label(label, is_inverted):
assert (isinstance(label, Label))
assert (isinstance(is_inverted, bool))
if not is_inverted:
return label
if isinstance(label, PositiveLabel):
return NegativeLabel()
elif isinstance(label, NegativeLabel):
return PositiveLabel()
raise Exception("Not supported label")
def get_method_statistic(files_to_compare_list, synonyms_filepath, stemmer):
"""
Calculate statistic based on result files
files_to_compare_list: list
list of FilesToCompare objects
synonyms_filepath: str
stemmer: Stemmer
"""
assert(isinstance(stemmer, Stemmer))
columns = ["t_all", "t_pos", "t_neg", "e_all", "e_pos", "e_neg"]
df = pd.DataFrame(columns=columns)
for files_to_compare in files_to_compare_list:
assert(isinstance(files_to_compare, FilesToCompare))
test_opins = OpinionCollection.from_file(
files_to_compare.test_filepath, synonyms_filepath, stemmer=stemmer)
etalon_opins = OpinionCollection.from_file(
files_to_compare.etalon_filepath, synonyms_filepath, stemmer=stemmer)
df.loc[files_to_compare.index] = [
MethodStatistic.founded_opins(test_opins, etalon_opins),
MethodStatistic.founded_opins(test_opins, etalon_opins, PositiveLabel()),
MethodStatistic.founded_opins(test_opins, etalon_opins, NegativeLabel()),
len(etalon_opins),
len(list(etalon_opins.iter_sentiment(PositiveLabel()))),
len(list(etalon_opins.iter_sentiment(NegativeLabel())))]
df.loc['sum'] = [float(df[c].sum()) for c in columns]
df.loc['found'] = None
df.loc['found']['t_all'] = float(df.loc['sum']['t_all']) / df.loc['sum']['e_all']
df.loc['found']['t_pos'] = float(df.loc['sum']['t_pos']) / df.loc['sum']['e_pos']
df.loc['found']['t_neg'] = float(df.loc['sum']['t_neg']) / df.loc['sum']['e_neg']
return df
def __neg_labeling_callback(self, rc, dest_data_type):
assert(isinstance(rc, ExtractedRelationsCollection))
assert(isinstance(dest_data_type, unicode))
rch = self.get_relations_collection_helper(dest_data_type)
assert(isinstance(rch, ExtractedRelationsCollectionHelper))
stat, _ = rch.get_statistic()
neg_prec_bound = stat[2] / 100.0
for relation in rc:
assert(isinstance(relation, ExtractedRelation))
next = random.random()
label = NegativeLabel() if next <= neg_prec_bound else PositiveLabel()
rc.apply_label(label=label, relation_id=relation.RelationID)
return PredictVariables()
class Evaluator:
# Columns
C_POS_PREC = 'pos_prec'
C_NEG_PREC = 'neg_prec'
C_POS_RECALL = 'pos_recall'
C_NEG_RECALL = 'neg_recall'
C_F1_POS = 'f1_pos'
C_F1_NEG = 'f1_neg'
C_F1 = 'f1'
# Columns for differences
C_WHO = 'who'
C_TO = 'to'
C_ORIG = 'how_orig'
C_RES = 'how_results'
C_CMP = 'comparison'
def __init__(self, synonyms_filepath, user_answers_filepath, stemmer):
assert(isinstance(stemmer, Stemmer)) # for opinion collections
self.synonyms_filepath = synonyms_filepath
self.user_answers = user_answers_filepath
self.stemmer = stemmer
self.pos = PositiveLabel()
self.neg = NegativeLabel()
self.neu = NeutralLabel()
@staticmethod
def get_result_columns():
return [Evaluator.C_POS_PREC,
Evaluator.C_POS_PREC,
Evaluator.C_NEG_PREC,
Evaluator.C_POS_RECALL,
Evaluator.C_NEG_RECALL,
Evaluator.C_F1_POS,
Evaluator.C_F1_NEG,
Evaluator.C_F1]
@staticmethod
def _calcRecall(results, answers, label):
assert(isinstance(label, PositiveLabel) or isinstance(label, NegativeLabel))
if len(results[results[Evaluator.C_ORIG] == label.to_str()]) != 0:
return 1.0 * len(answers[(answers[Evaluator.C_CMP] == True)]) / len(results[results[Evaluator.C_ORIG] == label.to_str()])
else:
return 0.0
@staticmethod
def _calcPrecision(answers):
if len(answers) != 0:
return 1.0 * len(answers[(answers[Evaluator.C_CMP] == True)]) / len(answers)
else:
return 0.0
def _calcPrecisionAndRecall(self, results):
""" Расчет полноты и точности.
"""
pos_answers = results[(results[Evaluator.C_RES] == self.pos.to_str())]
neg_answers = results[(results[Evaluator.C_RES] == self.neg.to_str())]
pos_prec = self._calcPrecision(pos_answers)
neg_prec = self._calcPrecision(neg_answers)
pos_recall = self._calcRecall(results, pos_answers, self.pos)
neg_recall = self._calcRecall(results, neg_answers, self.neg)
assert(isinstance(pos_prec, float))
assert(isinstance(neg_prec, float))
assert(isinstance(pos_recall, float))
assert(isinstance(neg_recall, float))
return pos_prec, neg_prec, pos_recall, neg_recall
def _check(self, etalon_opins, test_opins):
assert(isinstance(etalon_opins, OpinionCollection))
assert(isinstance(test_opins, OpinionCollection))
df = pd.DataFrame(
columns=[self.C_WHO, self.C_TO, self.C_ORIG, self.C_RES, self.C_CMP])
r_ind = 0
# Append everithing that exist in etalon collection.
for o_etalon in etalon_opins:
comparison = False
has_opinion = test_opins.has_opinion_by_synonyms(o_etalon)
if has_opinion:
o_test = test_opins.get_opinion_by_synonyms(o_etalon)
comparison = o_test.sentiment == o_etalon.sentiment
df.loc[r_ind] = [o_etalon.value_left.encode('utf-8'),
o_etalon.value_right.encode('utf-8'),
o_etalon.sentiment.to_str(),
None if not has_opinion else o_test.sentiment.to_str(),
comparison]
r_ind += 1
# Append everithing that exist in test collection.
for o_test in test_opins:
has_opinion = etalon_opins.has_opinion_by_synonyms(o_test)
if has_opinion:
continue
df.loc[r_ind] = [o_test.value_left.encode('utf-8'),
o_test.value_right.encode('utf-8'),
None,
o_test.sentiment.to_str(),
False]
r_ind += 1
return df
# TODO. change it with the list of FilesToCompare objects.
def _calc_a_file(self, files_to_compare, debug):
assert(isinstance(files_to_compare, FilesToCompare))
# Reading test answers.
test_opins = OpinionCollection.from_file(
files_to_compare.test_filepath,
self.synonyms_filepath,
stemmer=self.stemmer)
# Reading etalon answers.
etalon_opins = OpinionCollection.from_file(
files_to_compare.etalon_filepath,
self.synonyms_filepath,
stemmer=self.stemmer)
if debug:
print "{} <-> {}, {}".format(
files_to_compare.test_filepath,
files_to_compare.etalon_filepath,
files_to_compare.index)
# Comparing test and etalon results.
results = self._check(etalon_opins, test_opins)
# Save result comparison into file.
# TODO. remove path declaration from here.
comparison_file = "{}/art{}.comp.txt".format(
self.user_answers, str(files_to_compare.index))
if debug:
print "Save comparison file: {}".format(comparison_file)
results.to_csv(comparison_file)
return self._calcPrecisionAndRecall(results)
def evaluate(self, files_to_compare_list, debug=False):
""" Main evaluation subprogram
"""
assert(isinstance(files_to_compare_list, list))
pos_prec, neg_prec, pos_recall, neg_recall = (0, 0, 0, 0)
for files_to_compare in files_to_compare_list:
[pos_prec1, neg_prec1, pos_recall1, neg_recall1] = self._calc_a_file(files_to_compare, debug=debug)
pos_prec += pos_prec1
neg_prec += neg_prec1
pos_recall += pos_recall1
neg_recall += neg_recall1
# print len(files_to_compare_list)
pos_prec /= len(files_to_compare_list)
neg_prec /= len(files_to_compare_list)
pos_recall /= len(files_to_compare_list)
neg_recall /= len(files_to_compare_list)
if pos_prec * pos_recall != 0:
f1_pos = 2 * pos_prec * pos_recall / (pos_prec + pos_recall)
else:
f1_pos = 0
if neg_prec * neg_recall != 0:
f1_neg = 2 * neg_prec * neg_recall / (neg_prec + neg_recall)
else:
f1_neg = 0
return {self.C_POS_PREC: pos_prec,
self.C_NEG_PREC: neg_prec,
self.C_POS_RECALL: pos_recall,
self.C_NEG_RECALL: neg_recall,
self.C_F1_POS: f1_pos,
self.C_F1_NEG: f1_neg,
self.C_F1: (f1_pos + f1_neg) / 2}