def train(model):
with qacache.open_cache('qa-mert_input.txt','w') as f:
questions = defaultdict(lambda: []) # empty list as default value
features_names = {}
# build input for tmert
for _model in model:
for i in range(0,len(_model)):
for j in range (0, len(_model[i]['q'])):
qs = _model[i]['q'][j]['answer']
for candidate in qs:
_id = str(i) + str(j)
feats = candidate['score']
feature_names = {name: 1 for name in feats.keys()}
correct = 1 if 'correct' in candidate and candidate['correct'] else 0
questions[_id].append((int(correct), feats))
f.write(_id + ' ||| ' + str(correct) + ' ||| ' + ' '.join([key + "=" + str(value) for (key, value) in feats.items()]) + '\n')
best_weight = mert_training(questions,feature_names)
qacache.store_weight(best_weight)
return best_weight
def write_result(testdoc, name): f = cache.open_cache(name,'w') f.write(json.dumps(testdoc, sort_keys=True, indent=4 * ' ')) f.close()