import importlib.util
import os
from nltk import word_tokenize
from common.dataset.formatter import Formatter
from common.dataset.label_schema import LabelSchema
from retrieval.filter_uninformative import uninformative
def preprocess(p):
return p.replace(" ","_").replace("(","-LRB-").replace(")","-RRB-").replace(":","-COLON-").split("#")[0]
class FeverFormatter(Formatter):
def __init__(self, index, label_schema, tokenizer=None,filtering=None):
super().__init__(label_schema)
self.index=index
self.tokenize = tokenizer if tokenizer is not None else self.nltk_tokenizer
self.filtering = None
def import_module(filename):
spec = importlib.util.spec_from_file_location('filter_doc', filename)
module = importlib.util.module_from_spec(spec)
spec.loader.exec_module(module)
return module
if filtering is not None:
if filtering:
self.filtering = import_module(filtering).preprocess
def nltk_tokenizer(self,text):
return " ".join(word_tokenize(text))
class FEVERGoldFormatter(FeverFormatter):
def format_line(self,line):
annotation = None
if "label" in line:
annotation = line["label"]
pages = []
if 'predicted_sentences' in line:
pages.extend([(ev[0], ev[1]) for ev in line["predicted_sentences"]])
elif 'predicted_pages' in line:
pages.extend([(ev[0],-1) for ev in line["predicted_pages"]])
else:
for evidence_group in line["evidence"]:
pages.extend([(ev[2],ev[3]) for ev in evidence_group])
if self.filtering is not None:
for page,_ in pages:
if self.filtering({"id":page}) is None:
return None
if annotation is not None:
return {"claim":self.tokenize(line["claim"]), "evidence": pages, "label":self.label_schema.get_id(annotation),"label_text":annotation}
else:
return {"claim":self.tokenize(line["claim"]), "evidence": pages, "label":None,"label_text":None}
class FEVERPredictionsFormatter(FeverFormatter):
def format_line(self,line):
annotation = line["label"]
if annotation is None:
annotation = line["verifiable"]
pages = [preprocess(ev[0]) for ev in line["predicted_pages"]]
return {"claim":self.tokenize(line["claim"]), "evidence": pages, "label":self.label_schema.get_id(annotation),"label_text":annotation}
class FEVERPredictions2Formatter(FeverFormatter):
def format_line(self,line):
annotation = line["label"]
if annotation is None:
annotation = line["verifiable"]
if 'predicted_pages' in line:
pages = [ev[0] for ev in line["predicted_pages"]]
elif 'evidence' in line:
pages = [ev[1] for ev in line["evidence"]]
else:
pages = []
return {"claim":self.tokenize(line["claim"]), "evidence": pages, "label":self.label_schema.get_id(annotation),"label_text":annotation}
class FEVERLabelSchema(LabelSchema):
def __init__(self):
super().__init__(["supports","refutes","not enough info"])
