def seed():
es = WikiSearchAdmin(
config_path='fact_verification_system/search/config.yaml').es
print(es)
print(dir(es))
print(es.ping)
print(es.info())
def main():
wsa = WikiSearchAdmin(
config_path='fact_verification_system/search/config.yaml')
# mock_input_claim = "Robert Downey jr is famous for playing Ironman."
mock_input_claim = "Wildcats"
res = wsq.query(wsa.es, mock_input_claim)
# pp.pprint(res.results)
pp.pprint(res.get_hits(limit=10))
def eval(args):
config = 'fact_verification_system/search/config.yaml'
ss = SentenceSelection()
es = WikiSearchAdmin(config).es
total = 0
total_matched = 0
with open(
'../dataset/devset.json', 'r'
) as fp: # NOTE: change to train.json when you receive the dataset
train_json = json.load(fp)
for i, data in enumerate(train_json.values()):
claim = data.get('claim')
# get RETRIEVED (page_id, sent_idx) tuples from returned search results
limit = 10
res = wsq.query(es, claim)
rel_pageid_sentidx = res.get_page_id_sent_idx(limit=limit)
sentences = res.get_sentences(limit=limit)
# sentence selection
filtered_sentences = ss.filtered_sentences(claim, sentences)
indices = (i for (i, sent) in filtered_sentences)
# modified retrieved data
filtered_rel = set(
filter(lambda x: x[0] in indices, enumerate(rel_pageid_sentidx)
)) #x[0] is an index, x[1] is (page_id, sent_idx)
# get TRUE (page_id, sent_idx) tuples from devset.json
true_pageid_sentidx = set([(ev[0], ev[1])
for ev in data.get('evidence')])
matched = len(filtered_rel.intersection(true_pageid_sentidx))
total_matched += matched
total += matched
total += len(data.get('evidence'))
if args.debug:
print("retrieved:\n{}".format(rel_pageid_sentidx))
print("filtered_sentences:\n{}".format(filtered_sentences))
print("filtered:\n{}".format(filtered_rel))
print("true:\n{}".format(true_pageid_sentidx))
if i >= args.debug - 1:
print("Eval ended. (DEBUG MODE)")
return
# recall = (retrieved & true) / true
recall = float(total_matched) / float(total)
print("Recall: {:.2f}".format(recall))
def setup():
print("Setting up elasticsearch database...")
index_name = 'wiki'
config = 'fact_verification_system/search/config.yaml'
wsa = WikiSearchAdmin(config)
try:
print("Creating index: {}".format(index_name))
wsa.es.indices.create(index=index_name)
except RequestError as e:
print(e)
print("[setup.py] Index: {} already exists.".format(index_name))
return None
except Exception as e:
print("[setup.py] Error while setting up elasticsearch index.")
raise e
def seed():
index_name = 'wiki'
config = 'fact_verification_system/search/config.yaml'
es = WikiSearchAdmin(config).es
prompt_str = "Are you sure you want to seed the elastic search database?\n\
(This will delete the existing 'wiki' index if it exists)\n\
Enter y/n: "
confirm_prompt = input(prompt_str)
if confirm_prompt != 'y':
exit("Exited.")
# ES 7.0 has removed mapping types, API calls are now Typeless
print("Creating new index: {}...".format(index_name))
if es.indices.exists(index=index_name):
es.indices.delete(index=index_name)
es.indices.create(index=index_name)
print("Seeding newly created index: {}...".format(index_name))
cpu_count = multiprocessing.cpu_count()
# deque(parallel_bulk(
# client=wsa.es,
# actions=_iter_wiki_data_for_es(index_name),
# thread_count=cpu_count
# ), maxlen=0)
pb = parallel_bulk(
client=es,
actions=_iter_wiki_data_for_es(index_name),
thread_count=cpu_count
)
for success, info in pb:
if not success:
print("A document failed: {}".format(info))
es.indices.refresh()
print("Bulk insert complete.")
def eval(args):
config = 'fact_verification_system/search/config.yaml'
print("Running evaluation script for search engine...")
print("Using elasticsearch config path: {}".format(config))
total = 0
total_matched = 0
es = WikiSearchAdmin(config_path=config).es
with open(
'../dataset/devset.json', 'r'
) as fp: # NOTE: change to train.json when you receive the dataset
train_json = json.load(fp)
for i, data in tqdm(enumerate(train_json.values())):
claim = data.get('claim')
# get RETRIEVED (page_id, sent_idx) tuples from returned search results
rel_pageid_sentidx = wsq.query(
es, claim).get_page_id_sent_idx(limit=10)
# get TRUE (page_id, sent_idx) tuples from devset.json
true_pageid_sentidx = set([(ev[0], ev[1])
for ev in data.get('evidence')])
matched = len(rel_pageid_sentidx.intersection(true_pageid_sentidx))
total_matched += matched
total += matched
total += len(data.get('evidence'))
if args.debug:
print("retrieved:\n {}".format(rel_pageid_sentidx))
print("true:\n {}".format(true_pageid_sentidx))
if i >= args.debug:
print("Eval ended. (DEBUG MODE)")
return
# recall = (retrieved & true) / true
recall = float(total_matched) / float(total)
print("Recall: {:.2f}".format(recall))
from flask import request, abort, Response
from flask_restplus import Resource, Api, fields
import pprint
pp = pprint.PrettyPrinter(indent=2).pprint
from fact_verification_system.search.wiki_search_admin import WikiSearchAdmin
from fact_verification_system.search.wiki_search_query import WikiSearchQuery as wsq
from fact_verification_system.sentence_selection.sentence_selection import SentenceSelection
from fact_verification_system.classifier.tfx.predict import TFXPredict
api = Api(app)
config = 'fact_verification_system/search/config.yaml'
es = WikiSearchAdmin(config).es
ss = SentenceSelection()
tfxp = TFXPredict()
@api.route('/health')
class Health(Resource):
def get(self):
# TODO: send a query to es, sentence selection then classifier.
return formattedResponse(None, "Fact Verification System is active.")
@api.route('/evidence')
class Evidence(Resource):
def post(self):