def test():
gspec = AbstractGuesser.list_enabled_guessers()[0]
guesser_dir = AbstractGuesser.output_path(gspec.guesser_module,
gspec.guesser_class, '')
guesser = ElasticSearchWikidataGuesser.load(guesser_dir)
torch.cuda.set_device(0)
predictor = Predictor()
predictor.cuda()
dataset = BonusPairsDataset()
examples = [x for x in dataset.examples if x['start'] != -1]
guesses = []
for example in tqdm(examples):
document = example['content']
question = example['query']
answer = example['answer']
predictions = predictor.predict(document, question, top_n=1)
prediction = predictions[0][0]
gs = guesser.guess_single(example['query'])
gs = sorted(gs.items(), key=lambda x: x[1])[::-1]
guess = gs[0][0].replace('_', ' ')
guesses.append((prediction, guess, example['answer']))
with open('results.pkl', 'wb') as f:
pickle.dump(guesses, f)
def process(document, question, candidates=None,top_n=1):
predictor = Predictor(None,'spacy',num_workers=0,normalize=True)
predictions = predictor.predict(document,question,candidates,top_n)
#table = prettytable.PrettyTable(['Rank','Span','Score'])
val = []
for i,p in enumerate(predictions,1):
val.append(p[0])
return val[0]
def process():
data = request.json
document = data['document']
#print(document)
question = data['question']
#print(question)
predictor = Predictor(None, 'spacy', num_workers=0, normalize=True)
predictions = predictor.predict(document, question, None, 1)
val = []
for i, p in enumerate(predictions, 1):
val.append(p[0])
return jsonify(val)