sample_inputs = []
val = {"e1": args.input, "r": "Causes"}
sample_inputs.append(val)
if "bs" in opt.eval:
opt.eval.pop("bs")
if "k" in opt.eval:
opt.eval.pop("k")
if "beam" in args.sampling_algorithm:
opt.eval.sample = "beam"
opt.eval.bs = int(args.sampling_algorithm.split("-")[1])
sampler = BeamSampler(opt, data_loader)
else:
opt.eval.sample = "greedy"
opt.eval.bs = 1
sampler = GreedySampler(opt, data_loader)
outputs = []
for input_case in tqdm(sample_inputs):
e1 = input_case["e1"]
r = input_case["r"]
output = interactive.get_conceptnet_sequence(e1, model, sampler,
data_loader, text_encoder, r)
outputs.append(output)
json.dump(outputs, open(args.output_file, "w"))
relation = ['AtLocation', 'HasPrerequisite', 'Desires', 'UsedFor']
sampling_algorithm = args.sampling_algorithm
# %%
sampler = interactive.set_sampler(opt, sampling_algorithm, data_loader)
for rel in relation:
if rel not in data.conceptnet_data.conceptnet_relations:
relation = "all"
break
start = time.time()
logging.info('Getting Outputs from ConceptNet...')
outputs = [
interactive.get_conceptnet_sequence(input_event, comet_model, sampler,
data_loader, text_encoder, relation)
for input_event in input_events
]
logging.info('Total Time Used: ' + str(time.time() - start) + "s")
# %% [markdown]
# use bert-as-service to embed the output phrases from conceptnet
# %%
start = time.time()
logging.info('Getting Happy Elements Encodings...')
num_beam = int(sampling_algorithm.split('-')[1])
embeddings = [
beams for x in outputs for beams in [
def _get_result(self, event, category):
raw_result = interactive.get_conceptnet_sequence(
event, self._model, self._sampler, self._data_loader,
self._text_encoder, category)
return self.all_beams_cleanup(raw_result)
]
results_cache = {}
for sample in tqdm.tqdm(raw_data):
for turn in sample[0]:
content = turn.split(':')[1].strip()
sents = nltk.sent_tokenize(content)
for sent in sents:
sent = sent.strip()[:-1]
subsents = sent.split(',')
for subsent in subsents:
tokens = tokenizer.tokenize(subsent)
candidates = get_ngrams(tokens)
for cand in candidates:
if cand not in results_cache:
outputs = interactive.get_conceptnet_sequence(
cand, model, sampler, data_loader,
text_encoder, relation)
results_cache[cand] = outputs
for qa in sample[1]:
ques = qa['question'][:-1]
subqs = qa['question'].split(',')
for subq in subqs:
subq = subq.strip()
tokens = tokenizer.tokenize(subq)
candidates = get_ngrams(tokens)
for cand in candidates:
if cand not in results_cache:
outputs = interactive.get_conceptnet_sequence(
cand, model, sampler, data_loader, text_encoder,
relation)
results_cache[cand] = outputs
def _get_sequence(self, event, relations, sampler):
return interactive.get_conceptnet_sequence(event, self._model, sampler,
self._data_loader,
self._text_encoder,
relations)