def calc_q_link_accuracy(q_info_in, q_manual_info_in):
h_qid_info = load_query_info(q_info_in)
h_qid_manual_info = load_query_info(q_manual_info_in)
h_q_f1 = {}
for qid, h_info in h_qid_info.items():
l_e = []
if 'tagme' in h_info:
l_e = [ana[0] for ana in h_info['tagme']['query']]
elif 'cmns' in h_info:
l_e = [ana[0] for ana in h_info['cmns']['query']]
l_label_e = []
if qid in h_qid_manual_info:
l_label_e = [
ana[0] for ana in h_qid_manual_info[qid]['manual']['query']
]
if len(l_e) == 0 & len(l_label_e) == 0:
h_q_f1[qid] = 1
continue
s_e = set(l_e)
s_true = set(l_label_e)
prec = 0
recall = 0
overlap = float(len(s_e.intersection(s_true)))
if s_e:
prec = overlap / len(s_e)
if s_true:
recall = overlap / len(s_true)
if (prec == 0) | (recall == 0):
f1 = 0
else:
f1 = 2.0 * prec * recall / (prec + recall)
h_q_f1[qid] = f1
print json.dumps(h_q_f1, indent=1)
return h_q_f1
def process(q_info_in, out_name):
h_q_info = load_query_info(q_info_in)
bow_len, boe_len = avg_len(h_q_info)
out = open(out_name, 'w')
print >> out, 'bow_avg_len: %f\nboe_avg_len: %f' % (bow_len, boe_len)
out.close()
def __init__(self, **kwargs):
super(FusionAnalysis, self).__init__(**kwargs)
self.h_q_info = load_query_info(self.q_info)
self.out_dir = os.path.join(
self.out_dir, ntpath.basename(self.target_eva.split('.')[0]))
if not os.path.exists(self.out_dir):
os.makedirs(self.out_dir)
self.h_q_eva = dict(load_gdeval_res(self.target_eva)[0])
self.h_base_q_eva = dict(load_gdeval_res(self.base_eva)[0])
def get_target_surfaceforms(q_info_in):
h_qid_info = load_query_info(q_info_in)
h_surface = dict()
for qid, h_info in h_qid_info.items():
query = h_info['query']
for ana in h_info['tagme']['query']:
sf = query[ana[1]:ana[2]]
h_surface[sf.lower()] = []
print "total [%d] sf" % (len(h_surface))
return h_surface
def __init__(self, **kwargs):
super(RankComponentAna, self).__init__(**kwargs)
self.external_info = LeToRFeatureExternalInfo(**kwargs)
self.embedding = self.external_info.l_embedding[0]
self.h_entity_texts = self.external_info.h_entity_texts
self.h_field_h_df = self.external_info.h_field_h_df
self.h_corpus_stat = self.external_info.h_corpus_stat
self.h_q_info = load_query_info(self.q_info_in)
self.ll_qid_ranked_doc = load_trec_ranking_with_info(
self.trec_with_info_in)
self.h_qrel = load_trec_labels_dict(self.qrel_in)
if not os.path.exists(self.out_dir):
os.makedirs(self.out_dir)
def pipe_extract(self, q_info_in=None, out_name=None):
if not q_info_in:
q_info_in = self.q_info_in
if not out_name:
out_name = self.out_name
h_q_info = load_query_info(q_info_in)
l_h_feature = []
l_y = []
l_q_info = h_q_info.items()
l_q_info.sort(key=lambda item: int(item[0]))
l_qid = []
for qid, h_info in l_q_info:
h_feature = self._extract(qid, h_info)
y = -1
if qid in self.h_label:
y = self.h_label[qid]
l_h_feature.append(h_feature)
l_y.append(y)
l_qid.append(qid)
self._dump_feature_svm(l_y, l_h_feature, l_qid, out_name)
logging.info('q att feature extracted to [%s]', out_name)
return
def _load_data(self):
if self.ref_q_info_in:
self.h_ref_q_info = load_query_info(self.ref_q_info_in)
def __init__(self, **kwargs):
super(QLenPerformanceAna, self).__init__(**kwargs)
self.h_q_info = load_query_info(self.q_info_in)
self.h_rel_ndcg = get_rel_ndcg(self.eva_in, self.base_eva_in)
self.h_base_eva = dict(load_gdeval_res(self.base_eva_in, False))
self.h_eva = dict(load_gdeval_res(self.eva_in, False))