def load_data(input_file):
raw_data = open(input_file,'r')
doc_list = []
doc = Document()
ne_dict = {}
for rline in raw_data.readlines():
if rline.strip():
i = 0
entry = rline.split()
if len(entry) == 14:
i = 1
docID = entry[i]
#new document
if docID != doc.docID:
#import pdb
#if doc.docID!='':
# pdb.set_trace()
#record the name entity dictionary we have created
doc.set_ne_dict(ne_dict)
doc_list.append(doc)
ne_dict = {}
doc = Document(docID)
first = Entity(entry[i+1],(entry[i+2],entry[i+3]),entry[i+4],entry[i+5],entry[i+6])
ne_dict[entry[i+5]] = (entry[i+1],entry[i+2])
second = Entity(entry[i+7],(entry[i+8],entry[i+9]),entry[i+10],entry[i+11],entry[i+12])
ne_dict[entry[i+11]] = (entry[i+7],entry[i+8])
pair = Pair(first,second)
if i:
pair.set_label(entry[0])
doc.add_pair(pair)
else:
first = Entity(entry[i+1],(entry[i+2],entry[i+3]),entry[i+4],entry[i+5],entry[i+6])
ne_dict[entry[i+5]] = (entry[i+1],entry[i+2])
second = Entity(entry[i+7],(entry[i+8],entry[i+9]),entry[i+10],entry[i+11],entry[i+12])
ne_dict[entry[i+11]] = (entry[i+7],entry[i+8])
pair = Pair(first,second)
if i:
pair.set_label(entry[0])
doc.add_pair(pair)
doc.set_ne_dict(ne_dict)
doc_list.append(doc)
return doc_list
def DoMark(self, request, context):
""" gec服务的处理函数
Args:
request: rpc client发送过来的请求
context: 没啥卵用
Returns: gec处理后的文本和相应的错误信息，json字符串格式
"""
gec_output = json.loads(request.marker_input.strip())
doc_sents = [
gec_output["sentence_" + str(i)]["corr_sent"]
for i in range(gec_output["sent_nums"])
]
doc_encodes = self.__bert_predictor.encode(doc_sents)
prompt_sents = sentence_tokenize(gec_output["title"])
prompt_encodes = self.__bert_predictor.encode(prompt_sents)
# article_id, article_set在推理的时候并没有什么作用，只是为了数据类型和train保持一致
features = dict()
features["doc_encodes"] = np.expand_dims(doc_encodes, 0)
features["article_set"] = np.array([9], np.int64)
features["domain1_score"] = np.array([0], np.float32)
features["article_id"] = np.array([0], np.int64)
features["doc_sent_num"] = np.array([gec_output["sent_nums"]],
np.int64)
features["prompt_encodes"] = np.expand_dims(prompt_encodes, 0)
try:
bsp_score = self.__bsp_stub.Predict(
self.__format_request("bsp", features), 10)
bsp_score = bsp_score.outputs["batch_scores"].float_val[0]
csp_score = self.__csp_stub.Predict(
self.__format_request("csp", features), 10)
csp_score = csp_score.outputs["batch_scores"].float_val[0]
psp_score = self.__psp_stub.Predict(
self.__format_request("psp", features), 10)
psp_score = psp_score.outputs["batch_scores"].float_val[0]
temp_doc = Document(gec_output)
handcrafted_features = temp_doc.features
doc_result = temp_doc.doc_result
handcrafted_features.extend([bsp_score, csp_score, psp_score])
dtest = xgb.DMatrix(handcrafted_features)
overall_score = self.__bst.predict(dtest)[0]
doc_result["score_lexical"] = bsp_score
doc_result["score_coherence"] = csp_score
doc_result["score_gramatical"] = bsp_score
doc_result["score_task"] = psp_score
doc_result["score_summary"] = float(overall_score)
gec_output["score_result"] = doc_result
return rpc_server_pb2.MarkerData(
marker_output=json.dumps(gec_output))
except:
raise ValueError("")
def load_data(input_file):
raw_data = open(input_file,'r')
doc_list = []
doc = Document()
for rline in raw_data.readlines():
if rline.strip():
entry = rline.split()
docID = entry[0]
if docID != doc.docID:
#import pdb
#if doc.docID!='':
# pdb.set_trace()
doc_list.append(doc)
doc = Document(docID)
first = Entity(entry[1],(entry[2],entry[3]),entry[4],entry[5])
second = Entity(entry[6],(entry[7],entry[8]),entry[9],entry[10])
pair = Pair(first,second)
if len(entry) == 12:
pair.set_label(entry[11])
doc.add_pair(pair)
else:
first = Entity(entry[1],(entry[2],entry[3]),entry[4],entry[5])
second = Entity(entry[6],(entry[7],entry[8]),entry[9],entry[10])
pair = Pair(first,second)
if len(entry) == 12:
pair.set_label(entry[11])
doc.add_pair(pair)
doc_list.append(doc)
return doc_list
def loadIndex(self, path):
"""
Loads the self.documents and self.invertedIndex dicts data from a file
created with the self.saveIndex method.
If it fails to open the file the exception is not handled.
param path: string containing the path to file to load.
return: None.
"""
print("Loading index from the file: {}".format(path))
fin = open(path)
# regex for parsing the self.documents
docRegex = re.compile(
r"(?P<id>\d+);(?P<year>\d+);(?P<title>.+);(?P<authors>.+)?;(?P<norm>.+)"
)
# parsing the documents data
for line in fin:
line = line.strip()
# if there's something in the line it must be data about a document
if line:
# if the line starts with a # ignore it
if not line.startswith("#"):
match = docRegex.match(line)
docId = int(match.group("id"))
year = match.group("year")
title = match.group("title")
authors = match.group("authors")
norm = float(match.group("norm"))
self.documents[docId] = Document(docId, year, title,
authors, norm)
# if there's an empty line the documents part have ended
else:
break
# regex for parsing the self.invertedIndex
indexRegex = re.compile(r"(?P<word>.+);(?P<idf>.+);(?P<lst>.+)")
# parsing the inverted index data
for line in fin:
line = line.strip()
if not line.startswith("#"):
match = indexRegex.match(line)
word = match.group("word")
idf = float(match.group("idf"))
lst = ast.literal_eval(match.group("lst"))
pair = (idf, lst)
self.invertedIndex[word] = pair
fin.close()
def treatLastDoc(self, lastDoc):
"""
Helper method that transforms the data in the lastDoc dict into a tuple
of util.Document object and a Counter containing the frequencies of the
words in the document
param lastDoc: a dict containing the data parsed.
return: a tuple(util.Document, collections.Counter). The counter is a
dict with word keys and frequency values.
"""
total = Counter()
# the list of relevant attributes to tokenize. Tokenize also
# removes stop words defined in the init method
relevant = ["TI", "AB", "EX", "MJ", "MN"]
for attr in relevant:
content = lastDoc[attr]
assert type(content) == str
words = self.tokenize(content)
counter = Counter(words)
total += counter
# form the Document object return
docId = int(lastDoc["RN"])
# get the year of publishment
regex = r"(?P<year>\d{2})(?P<idInYear>\d{3})"
sep = re.compile(regex)
match = sep.match(lastDoc["PN"])
year = int(match.group("year"))
title = lastDoc["TI"]
authors = lastDoc["AU"]
tempNorm = 0 # irrelevant norm to be udated in the future
doc = Document(docId, year, title, authors, tempNorm)
result = doc, total
return result
