def evaluate(self, dev, avg_best, BLEU=False):
logging.info("STARTING EVALUATION")
acc_avg = 0.0
wer_avg = 0.0
bleu_avg = 0.0
acc_P = 0.0
acc_V = 0.0
ref = []
hyp = []
ref_s = ""
hyp_s = ""
pbar = tqdm(enumerate(dev), total=len(dev))
for j, data_dev in pbar:
words = self.evaluate_batch(len(data_dev[1]), data_dev[0], data_dev[1], data_dev[2])
acc = 0
w = 0
temp_gen = []
# print(words)
for i, row in enumerate(np.transpose(words)):
st = ''
for e in row:
if e == '<EOS>':
break
else:
st += e + ' '
temp_gen.append(st)
correct = data_dev[7][i]
if (correct.lstrip().rstrip() == st.lstrip().rstrip()):
acc += 1
# else:
# print("Correct:"+str(correct.lstrip().rstrip()))
# print("\tPredict:"+str(st.lstrip().rstrip()))
# print("\tFrom:"+str(self.from_whichs[:,i]))
w += wer(correct.lstrip().rstrip(), st.lstrip().rstrip())
ref.append(str(correct.lstrip().rstrip()))
hyp.append(str(st.lstrip().rstrip()))
ref_s += str(correct.lstrip().rstrip()) + "\n"
hyp_s += str(st.lstrip().rstrip()) + "\n"
acc_avg += acc / float(len(data_dev[1]))
wer_avg += w / float(len(data_dev[1]))
pbar.set_description("R:{:.4f},W:{:.4f}".format(acc_avg / float(len(dev)),
wer_avg / float(len(dev))))
if (BLEU):
bleu_score = moses_multi_bleu(np.array(hyp), np.array(ref), lowercase=True)
logging.info("BLEU SCORE:" + str(bleu_score))
if (bleu_score >= avg_best):
self.save_model(str(self.name) + str(bleu_score))
logging.info("MODEL SAVED")
return bleu_score
else:
acc_avg = acc_avg / float(len(dev))
if (acc_avg >= avg_best):
self.save_model(str(self.name) + str(acc_avg))
logging.info("MODEL SAVED")
return acc_avg
def evaluate(self, out_f, dev, avg_best, BLEU=False):
logging.info("STARTING EVALUATION")
acc_avg = 0.0
wer_avg = 0.0
ref = []
hyp = []
pbar = tqdm(enumerate(dev), total=len(dev))
for j, data_dev in pbar:
words = self.evaluate_batch(batch_size=len(data_dev[1]),
input_batches=data_dev[0],
input_lengths=data_dev[1],
target_batches=data_dev[2],
target_lengths=data_dev[3],
target_index=data_dev[4],
src_plain=data_dev[5])
acc = 0
w = 0
print("data_dev", data_dev[1])
temp_gen = []
for i, row in enumerate(np.transpose(words)):
st = ''
for e in row:
if e == '<EOS>': break
else: st += e + ' '
temp_gen.append(st)
correct = " ".join(data_dev[6][i])
### IMPORTANT
### WE NEED TO COMPARE THE PLAIN STRING, BECAUSE WE COPY THE WORDS FROM THE INPUT
### ====>> the index in the output gold can be UNK
if (correct.lstrip().rstrip() == st.lstrip().rstrip()):
acc += 1
w += wer(correct.lstrip().rstrip(), st.lstrip().rstrip())
ref.append(str(correct.lstrip().rstrip()))
hyp.append(str(st.lstrip().rstrip()))
# print('correct',correct)
# print('hyp',st)
acc_avg += acc / float(len(data_dev[1]))
wer_avg += w / float(len(data_dev[1]))
pbar.set_description("R:{:.4f},W:{:.4f}".format(
acc_avg / float(len(dev)), wer_avg / float(len(dev))))
bleu_score = moses_multi_bleu(np.array(hyp),
np.array(ref),
lowercase=True)
out_f.writelines(str(correct) + "===" + str(hyp) + "\n")
out_f.writelines("---------" + "\n")
logging.info("BLEU SCORE:" + str(bleu_score))
out_f.writelines("bleu_score" + str(bleu_score) + "\n")
if (bleu_score >= avg_best):
self.save_model(str(self.name) + str(bleu_score))
logging.info("MODEL SAVED")
return bleu_score
def eval_bleu(data):
hyps = []
refs = []
for dialog in data:
pred_str = dialog["result"]
gold_str = dialog["target"]
hyps.append(pred_str)
refs.append(gold_str)
assert len(hyps) == len(refs)
hyp_arrys = np.array(hyps)
ref_arrys = np.array(refs)
bleu_score = moses_multi_bleu(hyp_arrys, ref_arrys, lowercase=True)
return bleu_score
def evaluate(self, dev, matric_best, early_stop=None):
print("STARTING EVALUATION")
# Set to not-training mode to disable dropout
self.encoder.train(False)
self.extKnow.train(False)
self.decoder.train(False)
ref, hyp = [], []
acc, total = 0, 0
dialog_acc_dict = {}
F1_pred, F1_cal_pred, F1_nav_pred, F1_wet_pred = 0, 0, 0, 0
F1_count, F1_cal_count, F1_nav_count, F1_wet_count = 0, 0, 0, 0
pbar = tqdm(enumerate(dev),total=len(dev))
new_precision, new_recall, new_f1_score = 0, 0, 0
if args['dataset'] == 'kvr':
with open('data/KVR/kvret_entities.json') as f:
global_entity = json.load(f)
global_entity_list = []
for key in global_entity.keys():
if key != 'poi':
global_entity_list += [item.lower().replace(' ', '_') for item in global_entity[key]]
else:
for item in global_entity['poi']:
global_entity_list += [item[k].lower().replace(' ', '_') for k in item.keys()]
global_entity_list = list(set(global_entity_list))
for j, data_dev in pbar:
# Encode and Decode
_, _, decoded_fine, decoded_coarse, global_pointer = self.encode_and_decode(data_dev, self.max_resp_len, False, True)
decoded_coarse = np.transpose(decoded_coarse)
decoded_fine = np.transpose(decoded_fine)
for bi, row in enumerate(decoded_fine):
st = ''
for e in row:
if e == 'EOS': break
else: st += e + ' '
st_c = ''
for e in decoded_coarse[bi]:
if e == 'EOS': break
else: st_c += e + ' '
pred_sent = st.lstrip().rstrip()
pred_sent_coarse = st_c.lstrip().rstrip()
gold_sent = data_dev['response_plain'][bi].lstrip().rstrip()
ref.append(gold_sent)
hyp.append(pred_sent)
if args['dataset'] == 'kvr':
# compute F1 SCORE
single_f1, count = self.compute_prf(data_dev['ent_index'][bi], pred_sent.split(), global_entity_list, data_dev['kb_arr_plain'][bi])
F1_pred += single_f1
F1_count += count
single_f1, count = self.compute_prf(data_dev['ent_idx_cal'][bi], pred_sent.split(), global_entity_list, data_dev['kb_arr_plain'][bi])
F1_cal_pred += single_f1
F1_cal_count += count
single_f1, count = self.compute_prf(data_dev['ent_idx_nav'][bi], pred_sent.split(), global_entity_list, data_dev['kb_arr_plain'][bi])
F1_nav_pred += single_f1
F1_nav_count += count
single_f1, count = self.compute_prf(data_dev['ent_idx_wet'][bi], pred_sent.split(), global_entity_list, data_dev['kb_arr_plain'][bi])
F1_wet_pred += single_f1
F1_wet_count += count
else:
# compute Dialogue Accuracy Score
current_id = data_dev['ID'][bi]
if current_id not in dialog_acc_dict.keys():
dialog_acc_dict[current_id] = []
if gold_sent == pred_sent:
dialog_acc_dict[current_id].append(1)
else:
dialog_acc_dict[current_id].append(0)
# compute Per-response Accuracy Score
total += 1
if (gold_sent == pred_sent):
acc += 1
if args['genSample']:
self.print_examples(bi, data_dev, pred_sent, pred_sent_coarse, gold_sent)
# Set back to training mode
self.encoder.train(True)
self.extKnow.train(True)
self.decoder.train(True)
bleu_score = moses_multi_bleu(np.array(hyp), np.array(ref), lowercase=True)
acc_score = acc / float(total)
print("ACC SCORE:\t"+str(acc_score))
if args['dataset'] == 'kvr':
F1_score = F1_pred / float(F1_count)
print("F1 SCORE:\t{}".format(F1_pred/float(F1_count)))
print("\tCAL F1:\t{}".format(F1_cal_pred/float(F1_cal_count)))
print("\tWET F1:\t{}".format(F1_wet_pred/float(F1_wet_count)))
print("\tNAV F1:\t{}".format(F1_nav_pred/float(F1_nav_count)))
print("BLEU SCORE:\t"+str(bleu_score))
else:
dia_acc = 0
for k in dialog_acc_dict.keys():
if len(dialog_acc_dict[k])==sum(dialog_acc_dict[k]):
dia_acc += 1
print("Dialog Accuracy:\t"+str(dia_acc*1.0/len(dialog_acc_dict.keys())))
if (early_stop == 'BLEU'):
if (bleu_score >= matric_best):
self.save_model('BLEU-'+str(bleu_score))
print("MODEL SAVED")
return bleu_score
elif (early_stop == 'ENTF1'):
if (F1_score >= matric_best):
self.save_model('ENTF1-{:.4f}'.format(F1_score))
print("MODEL SAVED")
return F1_score
else:
if (acc_score >= matric_best):
self.save_model('ACC-{:.4f}'.format(acc_score))
print("MODEL SAVED")
return acc_score
def evaluate(self, dev, avg_best, BLEU=False):
logging.info("STARTING EVALUATION")
acc_avg = 0.0
wer_avg = 0.0
bleu_avg = 0.0
acc_P = 0.0
acc_V = 0.0
microF1_PRED,microF1_PRED_cal,microF1_PRED_nav,microF1_PRED_wet = [],[],[],[]
microF1_TRUE,microF1_TRUE_cal,microF1_TRUE_nav,microF1_TRUE_wet = [],[],[],[]
ref = []
hyp = []
ref_s = ""
hyp_s = ""
pbar = tqdm(enumerate(dev), total=len(dev))
for j, data_dev in pbar:
words = self.evaluate_batch(len(data_dev[1]), data_dev[0],
data_dev[1], data_dev[2])
acc = 0
w = 0
temp_gen = []
#print(words)
for i, row in enumerate(np.transpose(words)):
st = ''
for e in row:
if e == '<EOS>':
break
else:
st += e + ' '
temp_gen.append(st)
correct = data_dev[7][i]
### compute F1 SCORE
if (len(data_dev) > 10):
f1_true, f1_pred = computeF1(data_dev[8][i],
st.lstrip().rstrip(),
correct.lstrip().rstrip())
microF1_TRUE += f1_true
microF1_PRED += f1_pred
f1_true, f1_pred = computeF1(data_dev[9][i],
st.lstrip().rstrip(),
correct.lstrip().rstrip())
microF1_TRUE_cal += f1_true
microF1_PRED_cal += f1_pred
f1_true, f1_pred = computeF1(data_dev[10][i],
st.lstrip().rstrip(),
correct.lstrip().rstrip())
microF1_TRUE_nav += f1_true
microF1_PRED_nav += f1_pred
f1_true, f1_pred = computeF1(data_dev[11][i],
st.lstrip().rstrip(),
correct.lstrip().rstrip())
microF1_TRUE_wet += f1_true
microF1_PRED_wet += f1_pred
if (correct.lstrip().rstrip() == st.lstrip().rstrip()):
acc += 1
#else:
# print("Correct:"+str(correct.lstrip().rstrip()))
# print("\tPredict:"+str(st.lstrip().rstrip()))
# print("\tFrom:"+str(self.from_whichs[:,i]))
w += wer(correct.lstrip().rstrip(), st.lstrip().rstrip())
ref.append(str(correct.lstrip().rstrip()))
hyp.append(str(st.lstrip().rstrip()))
ref_s += str(correct.lstrip().rstrip()) + "\n"
hyp_s += str(st.lstrip().rstrip()) + "\n"
acc_avg += acc / float(len(data_dev[1]))
wer_avg += w / float(len(data_dev[1]))
pbar.set_description("R:{:.4f},W:{:.4f}".format(
acc_avg / float(len(dev)), wer_avg / float(len(dev))))
if (len(data_dev) > 10):
logging.info(
"F1 SCORE:\t" +
str(f1_score(microF1_TRUE, microF1_PRED, average='micro')))
logging.info("F1 CAL:\t" + str(
f1_score(microF1_TRUE_cal, microF1_PRED_cal, average='micro')))
logging.info("F1 WET:\t" + str(
f1_score(microF1_TRUE_wet, microF1_PRED_wet, average='micro')))
logging.info("F1 NAV:\t" + str(
f1_score(microF1_TRUE_nav, microF1_PRED_nav, average='micro')))
if (BLEU):
bleu_score = moses_multi_bleu(np.array(hyp),
np.array(ref),
lowercase=True)
logging.info("BLEU SCORE:" + str(bleu_score))
if (bleu_score >= avg_best):
self.save_model(str(self.name) + str(bleu_score))
logging.info("MODEL SAVED")
return bleu_score
else:
acc_avg = acc_avg / float(len(dev))
if (acc_avg >= avg_best):
self.save_model(str(self.name) + str(acc_avg))
logging.info("MODEL SAVED")
return acc_avg
def evaluate(self, dev, matric_best, output=False, early_stop=None):
print("STARTING EVALUATION")
# Set to not-training mode to disable dropout
self.encoder.train(False)
self.extKnow.train(False)
self.decoder.train(False)
ref, hyp = [], []
ids = []
acc, total = 0, 0
if args['dataset'] == 'kvr':
F1_pred, F1_cal_pred, F1_nav_pred, F1_wet_pred = 0, 0, 0, 0
F1_count, F1_cal_count, F1_nav_count, F1_wet_count = 0, 0, 0, 0
TP_all, FP_all, FN_all = 0, 0, 0
TP_sche, FP_sche, FN_sche = 0, 0, 0
TP_wea, FP_wea, FN_wea = 0, 0, 0
TP_nav, FP_nav, FN_nav = 0, 0, 0
elif args['dataset'] == 'woz':
F1_pred, F1_police_pred, F1_restaurant_pred, F1_hospital_pred, F1_attraction_pred, F1_hotel_pred = 0, 0, 0, 0, 0, 0
F1_count, F1_police_count, F1_restaurant_count, F1_hospital_count, F1_attraction_count, F1_hotel_count = 0, 0, 0, 0, 0, 0
TP_all, FP_all, FN_all = 0, 0, 0
TP_restaurant, FP_restaurant, FN_restaurant = 0, 0, 0
TP_attraction, FP_attraction, FN_attraction = 0, 0, 0
TP_hotel, FP_hotel, FN_hotel = 0, 0, 0
elif args['dataset'] == 'cam':
F1_pred, F1_count = 0, 0
TP_all, FP_all, FN_all = 0, 0, 0
pbar = tqdm(enumerate(dev), total=len(dev))
if args['dataset'] == 'kvr':
entity_path = '../dataset/KVR/kvret_entities.json'
elif args['dataset'] == 'woz':
entity_path = '../dataset/MULTIWOZ2.1/global_entities.json'
elif args['dataset'] == 'cam':
entity_path = '../dataset/CamRest676/CamRest676_entities.json'
with open(entity_path) as f:
global_entity = json.load(f)
global_entity_type = {}
global_entity_list = []
for key in global_entity.keys():
if key != 'poi':
entity_arr = [
item.lower().replace(' ', '_')
for item in global_entity[key]
]
global_entity_list += entity_arr
for entity in entity_arr:
global_entity_type[entity] = key
else:
for item in global_entity['poi']:
entity_arr = [
item[k].lower().replace(' ', '_')
for k in item.keys()
]
global_entity_list += entity_arr
for key in item:
global_entity_type[item[key].lower().replace(
' ', '_')] = key
global_entity_list = list(set(global_entity_list))
for j, data_dev in pbar:
ids.extend(data_dev['id'])
# Encode and Decode
_, _, decoded_fine, decoded_coarse, global_pointer, _, _, _, _ = self.encode_and_decode(
data_dev, self.max_resp_len, False, True, global_entity_type)
decoded_coarse = np.transpose(decoded_coarse)
decoded_fine = np.transpose(decoded_fine)
for bi, row in enumerate(decoded_fine):
st = ''
for e in row:
if e == 'EOS':
break
else:
st += e + ' '
st_c = ''
for e in decoded_coarse[bi]:
if e == 'EOS':
break
else:
st_c += e + ' '
pred_sent = st.lstrip().rstrip()
pred_sent_coarse = st_c.lstrip().rstrip()
gold_sent = data_dev['response_plain'][bi].lstrip().rstrip()
ref.append(gold_sent)
hyp.append(pred_sent)
if args['dataset'] == 'kvr':
# compute F1 SCORE
single_tp, single_fp, single_fn, single_f1, count = self.compute_prf(
data_dev['ent_index'][bi], pred_sent.split(),
global_entity_list, data_dev['kb_arr_plain'][bi])
F1_pred += single_f1
F1_count += count
TP_all += single_tp
FP_all += single_fp
FN_all += single_fn
single_tp, single_fp, single_fn, single_f1, count = self.compute_prf(
data_dev['ent_idx_cal'][bi], pred_sent.split(),
global_entity_list, data_dev['kb_arr_plain'][bi])
F1_cal_pred += single_f1
F1_cal_count += count
TP_sche += single_tp
FP_sche += single_fp
FN_sche += single_fn
single_tp, single_fp, single_fn, single_f1, count = self.compute_prf(
data_dev['ent_idx_nav'][bi], pred_sent.split(),
global_entity_list, data_dev['kb_arr_plain'][bi])
F1_nav_pred += single_f1
F1_nav_count += count
TP_nav += single_tp
FP_nav += single_fp
FN_nav += single_fn
single_tp, single_fp, single_fn, single_f1, count = self.compute_prf(
data_dev['ent_idx_wet'][bi], pred_sent.split(),
global_entity_list, data_dev['kb_arr_plain'][bi])
F1_wet_pred += single_f1
F1_wet_count += count
TP_wea += single_tp
FP_wea += single_fp
FN_wea += single_fn
elif args['dataset'] == 'woz':
# coimpute F1 SCORE
single_tp, single_fp, single_fn, single_f1, count = self.compute_prf(
data_dev['ent_index'][bi], pred_sent.split(),
global_entity_list, data_dev['kb_arr_plain'][bi])
F1_pred += single_f1
F1_count += count
TP_all += single_tp
FP_all += single_fp
FN_all += single_fn
single_tp, single_fp, single_fn, single_f1, count = self.compute_prf(
data_dev['ent_idx_restaurant'][bi], pred_sent.split(),
global_entity_list, data_dev['kb_arr_plain'][bi])
F1_restaurant_pred += single_f1
F1_restaurant_count += count
TP_restaurant += single_tp
FP_restaurant += single_fp
FN_restaurant += single_fn
single_tp, single_fp, single_fn, single_f1, count = self.compute_prf(
data_dev['ent_idx_attraction'][bi], pred_sent.split(),
global_entity_list, data_dev['kb_arr_plain'][bi])
F1_attraction_pred += single_f1
F1_attraction_count += count
TP_attraction += single_tp
FP_attraction += single_fp
FN_attraction += single_fn
single_tp, single_fp, single_fn, single_f1, count = self.compute_prf(
data_dev['ent_idx_hotel'][bi], pred_sent.split(),
global_entity_list, data_dev['kb_arr_plain'][bi])
F1_hotel_pred += single_f1
F1_hotel_count += count
TP_hotel += single_tp
FP_hotel += single_fp
FN_hotel += single_fn
elif args['dataset'] == 'cam':
# compute F1 SCORE
single_tp, single_fp, single_fn, single_f1, count = self.compute_prf(
data_dev['ent_index'][bi], pred_sent.split(),
global_entity_list, data_dev['kb_arr_plain'][bi])
F1_pred += single_f1
F1_count += count
TP_all += single_tp
FP_all += single_fp
FN_all += single_fn
# compute Per-response Accuracy Score
total += 1
if (gold_sent == pred_sent):
acc += 1
if args['genSample']:
self.print_examples(bi, data_dev, pred_sent,
pred_sent_coarse, gold_sent)
# Set back to training mode
self.encoder.train(True)
self.extKnow.train(True)
self.decoder.train(True)
bleu_score = moses_multi_bleu(np.array(hyp),
np.array(ref),
lowercase=True)
acc_score = acc / float(total)
print("ACC SCORE:\t" + str(acc_score))
if args['dataset'] == 'kvr':
print("BLEU SCORE:\t" + str(bleu_score))
print("F1-macro SCORE:\t{}".format(F1_pred / float(F1_count)))
print("F1-macro-sche SCORE:\t{}".format(F1_cal_pred /
float(F1_cal_count)))
print("F1-macro-wea SCORE:\t{}".format(F1_wet_pred /
float(F1_wet_count)))
print("F1-macro-nav SCORE:\t{}".format(F1_nav_pred /
float(F1_nav_count)))
P_score = TP_all / float(TP_all + FP_all) if (TP_all +
FP_all) != 0 else 0
R_score = TP_all / float(TP_all + FN_all) if (TP_all +
FN_all) != 0 else 0
P_nav_score = TP_nav / float(TP_nav + FP_nav) if (
TP_nav + FP_nav) != 0 else 0
P_sche_score = TP_sche / float(TP_sche + FP_sche) if (
TP_sche + FP_sche) != 0 else 0
P_wea_score = TP_wea / float(TP_wea + FP_wea) if (
TP_wea + FP_wea) != 0 else 0
R_nav_score = TP_nav / float(TP_nav + FN_nav) if (
TP_nav + FN_nav) != 0 else 0
R_sche_score = TP_sche / float(TP_sche + FN_sche) if (
TP_sche + FN_sche) != 0 else 0
R_wea_score = TP_wea / float(TP_wea + FN_wea) if (
TP_wea + FN_wea) != 0 else 0
F1_score = self.compute_F1(P_score, R_score)
print("F1-micro SCORE:\t{}".format(F1_score))
print("F1-micro-sche SCORE:\t{}".format(
self.compute_F1(P_sche_score, R_sche_score)))
print("F1-micro-wea SCORE:\t{}".format(
self.compute_F1(P_wea_score, R_wea_score)))
print("F1-micro-nav SCORE:\t{}".format(
self.compute_F1(P_nav_score, R_nav_score)))
print("BLEU SCORE:" + str(bleu_score))
print("F1-macro SCORE:\t{}".format(F1_pred / float(F1_count)))
print("F1-micro SCORE:\t{}".format(F1_score))
elif args['dataset'] == 'woz':
print("BLEU SCORE:\t" + str(bleu_score))
print("F1-macro SCORE:\t{}".format(F1_pred / float(F1_count)))
print("F1-macro-restaurant SCORE:\t{}".format(
F1_restaurant_pred / float(F1_restaurant_count)))
print("F1-macro-attraction SCORE:\t{}".format(
F1_attraction_pred / float(F1_attraction_count)))
print("F1-macro-hotel SCORE:\t{}".format(F1_hotel_pred /
float(F1_hotel_count)))
P_score = TP_all / float(TP_all + FP_all) if (TP_all +
FP_all) != 0 else 0
R_score = TP_all / float(TP_all + FN_all) if (TP_all +
FN_all) != 0 else 0
P_restaurant_score = TP_restaurant / float(
TP_restaurant + FP_restaurant) if (TP_restaurant +
FP_restaurant) != 0 else 0
P_attraction_score = TP_attraction / float(
TP_attraction + FP_attraction) if (TP_attraction +
FP_attraction) != 0 else 0
P_hotel_score = TP_hotel / float(TP_hotel + FP_hotel) if (
TP_hotel + FP_hotel) != 0 else 0
R_restaurant_score = TP_restaurant / float(
TP_restaurant + FN_restaurant) if (TP_restaurant +
FN_restaurant) != 0 else 0
R_attraction_score = TP_attraction / float(
TP_attraction + FN_attraction) if (TP_attraction +
FN_attraction) != 0 else 0
R_hotel_score = TP_hotel / float(TP_hotel + FN_hotel) if (
TP_hotel + FN_hotel) != 0 else 0
F1_score = self.compute_F1(P_score, R_score)
print("F1-micro SCORE:\t{}".format(F1_score))
print("F1-micro-restaurant SCORE:\t{}".format(
self.compute_F1(P_restaurant_score, R_restaurant_score)))
print("F1-micro-attraction SCORE:\t{}".format(
self.compute_F1(P_attraction_score, R_attraction_score)))
print("F1-micro-hotel SCORE:\t{}".format(
self.compute_F1(P_hotel_score, R_hotel_score)))
elif args['dataset'] == 'cam':
print("BLEU SCORE:\t" + str(bleu_score))
print("F1-macro SCORE:\t{}".format(F1_pred / float(F1_count)))
P_score = TP_all / float(TP_all + FP_all) if (TP_all +
FP_all) != 0 else 0
R_score = TP_all / float(TP_all + FN_all) if (TP_all +
FN_all) != 0 else 0
F1_score = self.compute_F1(P_score, R_score)
print("F1-micro SCORE:\t{}".format(F1_score))
if output:
print('Test Finish!')
with open(args['output'], 'w+') as f:
if args['dataset'] == 'kvr':
print("ACC SCORE:\t" + str(acc_score), file=f)
print("BLEU SCORE:\t" + str(bleu_score), file=f)
print("F1-macro SCORE:\t{}".format(F1_pred /
float(F1_count)),
file=f)
print("F1-micro SCORE:\t{}".format(
self.compute_F1(P_score, R_score)),
file=f)
print("F1-macro-sche SCORE:\t{}".format(
F1_cal_pred / float(F1_cal_count)),
file=f)
print("F1-macro-wea SCORE:\t{}".format(
F1_wet_pred / float(F1_wet_count)),
file=f)
print("F1-macro-nav SCORE:\t{}".format(
F1_nav_pred / float(F1_nav_count)),
file=f)
print("F1-micro-sche SCORE:\t{}".format(
self.compute_F1(P_sche_score, R_sche_score)),
file=f)
print("F1-micro-wea SCORE:\t{}".format(
self.compute_F1(P_wea_score, R_wea_score)),
file=f)
print("F1-micro-nav SCORE:\t{}".format(
self.compute_F1(P_nav_score, R_nav_score)),
file=f)
elif args['dataset'] == 'woz':
print("ACC SCORE:\t" + str(acc_score), file=f)
print("BLEU SCORE:\t" + str(bleu_score), file=f)
print("F1-macro SCORE:\t{}".format(F1_pred /
float(F1_count)),
file=f)
print("F1-micro SCORE:\t{}".format(
self.compute_F1(P_score, R_score)),
file=f)
print("F1-macro-restaurant SCORE:\t{}".format(
F1_restaurant_pred / float(F1_restaurant_count)),
file=f)
print("F1-macro-attraction SCORE:\t{}".format(
F1_attraction_pred / float(F1_attraction_count)),
file=f)
print("F1-macro-hotel SCORE:\t{}".format(
F1_hotel_pred / float(F1_hotel_count)),
file=f)
print("F1-micro SCORE:\t{}".format(F1_score), file=f)
print("F1-micro-restaurant SCORE:\t{}".format(
self.compute_F1(P_restaurant_score,
R_restaurant_score)),
file=f)
print("F1-micro-attraction SCORE:\t{}".format(
self.compute_F1(P_attraction_score,
R_attraction_score)),
file=f)
print("F1-micro-hotel SCORE:\t{}".format(
self.compute_F1(P_hotel_score, R_hotel_score)),
file=f)
elif args['dataset'] == 'cam':
print("ACC SCORE:\t" + str(acc_score), file=f)
print("BLEU SCORE:\t" + str(bleu_score), file=f)
print("F1-macro SCORE:\t{}".format(F1_pred /
float(F1_count)),
file=f)
print("F1-micro SCORE:\t{}".format(
self.compute_F1(P_score, R_score)),
file=f)
if (early_stop == 'BLEU'):
if (bleu_score >= matric_best):
self.save_model('BLEU-' + str(bleu_score) + 'F1-' +
str(F1_score))
print("MODEL SAVED")
return bleu_score
elif (early_stop == 'ENTF1'):
if (F1_score >= matric_best):
# self.save_model('ENTF1-{:.4f}'.format(F1_score))
self.save_model('ENTF1')
print("MODEL SAVED")
return F1_score
else:
if (acc_score >= matric_best):
self.save_model('ACC-{:.4f}'.format(acc_score))
print("MODEL SAVED")
return acc_score
def evaluate(self, dev, avg_best, BLEU=False):
logging.info("STARTING EVALUATION")
acc_avg = 0.0
wer_avg = 0.0
bleu_avg = 0.0
acc_P = 0.0
acc_V = 0.0
microF1_PRED, microF1_PRED_cal, microF1_PRED_nav, microF1_PRED_wet = 0, 0, 0, 0
microF1_TRUE, microF1_TRUE_cal, microF1_TRUE_nav, microF1_TRUE_wet = 0, 0, 0, 0
ref = []
hyp = []
ref_s = ""
hyp_s = ""
dialog_acc_dict = {}
if args['dataset'] == 'kvr':
with open('data/KVR/kvret_entities.json') as f:
global_entity = json.load(f)
global_entity_list = []
for key in global_entity.keys():
if key != 'poi':
global_entity_list += [
item.lower().replace(' ', '_')
for item in global_entity[key]
]
else:
for item in global_entity['poi']:
global_entity_list += [
item[k].lower().replace(' ', '_')
for k in item.keys()
]
global_entity_list = list(set(global_entity_list))
else:
if int(args["task"]) != 6:
global_entity_list = entityList(
'data/dialog-bAbI-tasks/dialog-babi-kb-all.txt',
int(args["task"]))
else:
global_entity_list = entityList(
'data/dialog-bAbI-tasks/dialog-babi-task6-dstc2-kb.txt',
int(args["task"]))
pbar = tqdm(enumerate(dev), total=len(dev))
for j, data_dev in pbar:
if args['dataset'] == 'kvr':
words = self.evaluate_batch(len(data_dev[1]), data_dev[0],
data_dev[1], data_dev[2],
data_dev[3], data_dev[4],
data_dev[5], data_dev[6])
else:
words = self.evaluate_batch(len(data_dev[1]), data_dev[0],
data_dev[1], data_dev[2],
data_dev[3], data_dev[4],
data_dev[5], data_dev[6])
acc = 0
w = 0
temp_gen = []
for i, row in enumerate(np.transpose(words)):
st = ''
for e in row:
if e == '<EOS>': break
else: st += e + ' '
temp_gen.append(st)
correct = data_dev[7][i]
### compute F1 SCORE
st = st.lstrip().rstrip()
correct = correct.lstrip().rstrip()
if args['dataset'] == 'kvr':
f1_true, count = self.compute_prf(data_dev[8][i],
st.split(),
global_entity_list,
data_dev[14][i])
microF1_TRUE += f1_true
microF1_PRED += count
f1_true, count = self.compute_prf(data_dev[9][i],
st.split(),
global_entity_list,
data_dev[14][i])
microF1_TRUE_cal += f1_true
microF1_PRED_cal += count
f1_true, count = self.compute_prf(data_dev[10][i],
st.split(),
global_entity_list,
data_dev[14][i])
microF1_TRUE_nav += f1_true
microF1_PRED_nav += count
f1_true, count = self.compute_prf(data_dev[11][i],
st.split(),
global_entity_list,
data_dev[14][i])
microF1_TRUE_wet += f1_true
microF1_PRED_wet += count
elif args['dataset'] == 'babi' and int(args["task"]) == 6:
f1_true, count = self.compute_prf(data_dev[10][i],
st.split(),
global_entity_list,
data_dev[12][i])
microF1_TRUE += f1_true
microF1_PRED += count
if args['dataset'] == 'babi':
if data_dev[11][i] not in dialog_acc_dict.keys():
dialog_acc_dict[data_dev[11][i]] = []
if (correct == st):
acc += 1
dialog_acc_dict[data_dev[11][i]].append(1)
else:
dialog_acc_dict[data_dev[11][i]].append(0)
else:
if (correct == st):
acc += 1
# print("Correct:"+str(correct))
# print("\tPredict:"+str(st))
# print("\tFrom:"+str(self.from_whichs[:,i]))
w += wer(correct, st)
ref.append(str(correct))
hyp.append(str(st))
ref_s += str(correct) + "\n"
hyp_s += str(st) + "\n"
acc_avg += acc / float(len(data_dev[1]))
wer_avg += w / float(len(data_dev[1]))
pbar.set_description("R:{:.4f},W:{:.4f}".format(
acc_avg / float(len(dev)), wer_avg / float(len(dev))))
# dialog accuracy
if args['dataset'] == 'babi':
dia_acc = 0
for k in dialog_acc_dict.keys():
if len(dialog_acc_dict[k]) == sum(dialog_acc_dict[k]):
dia_acc += 1
logging.info("Dialog Accuracy:\t" +
str(dia_acc * 1.0 / len(dialog_acc_dict.keys())))
if args['dataset'] == 'kvr':
logging.info("F1 SCORE:\t{}".format(microF1_TRUE /
float(microF1_PRED)))
logging.info("\tCAL F1:\t{}".format(microF1_TRUE_cal /
float(microF1_PRED_cal)))
logging.info("\tWET F1:\t{}".format(microF1_TRUE_wet /
float(microF1_PRED_wet)))
logging.info("\tNAV F1:\t{}".format(microF1_TRUE_nav /
float(microF1_PRED_nav)))
elif args['dataset'] == 'babi' and int(args["task"]) == 6:
logging.info("F1 SCORE:\t{}".format(microF1_TRUE /
float(microF1_PRED)))
bleu_score = moses_multi_bleu(np.array(hyp),
np.array(ref),
lowercase=True)
logging.info("BLEU SCORE:" + str(bleu_score))
if (BLEU):
if (bleu_score >= avg_best):
self.save_model(str(self.name) + str(bleu_score))
logging.info("MODEL SAVED")
return bleu_score
else:
acc_avg = acc_avg / float(len(dev))
if (acc_avg >= avg_best):
self.save_model(str(self.name) + str(acc_avg))
logging.info("MODEL SAVED")
return acc_avg
def evaluate(self, dev, avg_best, BLEU=False):
logging.info("STARTING EVALUATION")
acc_avg = 0.0
wer_avg = 0.0
bleu_avg = 0.0
acc_P = 0.0
acc_V = 0.0
microF1_PRED,microF1_PRED_cal,microF1_PRED_nav,microF1_PRED_wet = [],[],[],[]
microF1_TRUE,microF1_TRUE_cal,microF1_TRUE_nav,microF1_TRUE_wet = [],[],[],[]
ref = []
hyp = []
ref_s = ""
hyp_s = ""
dialog_acc_dict = {}
pbar = tqdm(enumerate(dev), total=len(dev))
for j, data_dev in pbar:
if args['dataset'] == 'kvr':
words = self.evaluate_batch(len(data_dev[1]), data_dev[0],
data_dev[1], data_dev[2],
data_dev[3], data_dev[4],
data_dev[5], data_dev[6],
data_dev[-2], data_dev[-1])
else:
words = self.evaluate_batch(len(data_dev[1]), data_dev[0],
data_dev[1], data_dev[2],
data_dev[3], data_dev[4],
data_dev[5], data_dev[6],
data_dev[-4], data_dev[-3])
# acc_P += acc_ptr
# acc_V += acc_vac
acc = 0
w = 0
temp_gen = []
for i, row in enumerate(np.transpose(words)):
st = ''
for e in row:
if e == '<EOS>': break
else: st += e + ' '
temp_gen.append(st)
correct = data_dev[7][i]
### compute F1 SCORE
if args['dataset'] == 'kvr':
f1_true, f1_pred = computeF1(data_dev[8][i],
st.lstrip().rstrip(),
correct.lstrip().rstrip())
microF1_TRUE += f1_true
microF1_PRED += f1_pred
f1_true, f1_pred = computeF1(data_dev[9][i],
st.lstrip().rstrip(),
correct.lstrip().rstrip())
microF1_TRUE_cal += f1_true
microF1_PRED_cal += f1_pred
f1_true, f1_pred = computeF1(data_dev[10][i],
st.lstrip().rstrip(),
correct.lstrip().rstrip())
microF1_TRUE_nav += f1_true
microF1_PRED_nav += f1_pred
f1_true, f1_pred = computeF1(data_dev[11][i],
st.lstrip().rstrip(),
correct.lstrip().rstrip())
microF1_TRUE_wet += f1_true
microF1_PRED_wet += f1_pred
elif args['dataset'] == 'babi' and int(self.task) == 6:
f1_true, f1_pred = computeF1(data_dev[-2][i],
st.lstrip().rstrip(),
correct.lstrip().rstrip())
microF1_TRUE += f1_true
microF1_PRED += f1_pred
if args['dataset'] == 'babi':
if data_dev[-1][i] not in dialog_acc_dict.keys():
dialog_acc_dict[data_dev[-1][i]] = []
if (correct.lstrip().rstrip() == st.lstrip().rstrip()):
acc += 1
dialog_acc_dict[data_dev[-1][i]].append(1)
else:
dialog_acc_dict[data_dev[-1][i]].append(0)
else:
if (correct.lstrip().rstrip() == st.lstrip().rstrip()):
acc += 1
# print("Correct:"+str(correct.lstrip().rstrip()))
# print("\tPredict:"+str(st.lstrip().rstrip()))
# print("\tFrom:"+str(self.from_whichs[:,i]))
w += wer(correct.lstrip().rstrip(), st.lstrip().rstrip())
ref.append(str(correct.lstrip().rstrip()))
hyp.append(str(st.lstrip().rstrip()))
ref_s += str(correct.lstrip().rstrip()) + "\n"
hyp_s += str(st.lstrip().rstrip()) + "\n"
acc_avg += acc / float(len(data_dev[1]))
wer_avg += w / float(len(data_dev[1]))
pbar.set_description("R:{:.4f},W:{:.4f}".format(
acc_avg / float(len(dev)), wer_avg / float(len(dev))))
# dialog accuracy
if args['dataset'] == 'babi':
dia_acc = 0
for k in dialog_acc_dict.keys():
if len(dialog_acc_dict[k]) == sum(dialog_acc_dict[k]):
dia_acc += 1
logging.info("Dialog Accuracy:\t" +
str(dia_acc * 1.0 / len(dialog_acc_dict.keys())))
if args['dataset'] == 'kvr':
logging.info(
"F1 SCORE:\t" +
str(f1_score(microF1_TRUE, microF1_PRED, average='micro')))
logging.info("F1 CAL:\t" + str(
f1_score(microF1_TRUE_cal, microF1_PRED_cal, average='micro')))
logging.info("F1 WET:\t" + str(
f1_score(microF1_TRUE_wet, microF1_PRED_wet, average='micro')))
logging.info("F1 NAV:\t" + str(
f1_score(microF1_TRUE_nav, microF1_PRED_nav, average='micro')))
elif args['dataset'] == 'babi' and int(self.task) == 6:
logging.info(
"F1 SCORE:\t" +
str(f1_score(microF1_TRUE, microF1_PRED, average='micro')))
bleu_score = moses_multi_bleu(np.array(hyp),
np.array(ref),
lowercase=True)
logging.info("BLEU SCORE:" + str(bleu_score))
if (BLEU):
if (bleu_score >= avg_best):
self.save_model(str(self.name) + str(bleu_score))
logging.info("MODEL SAVED")
return bleu_score
else:
acc_avg = acc_avg / float(len(dev))
if (acc_avg >= avg_best):
self.save_model(str(self.name) + str(acc_avg))
logging.info("MODEL SAVED")
return acc_avg
def evaluate(self, dev, metric_best, early_stop=None):
print('\nSTARTING EVALUATING...')
self.encoder.train(False)
self.ext_know.train(False)
self.decoder.train(False)
label, pred = [], []
acc, total = 0, 0
# kvr数据集的评价指标
F1_pred, F1_cal_pred, F1_nav_pred, F1_wet_pred = 0, 0, 0, 0
F1_count, F1_cal_count, F1_nav_count, F1_wet_count = 0, 0, 0, 0
dialogue_acc_dict = defaultdict(list)
if args['dataset'] == 'kvr':
with open('../data/KVR/kvret_entities.json') as f:
global_entity = json.load(f)
global_entity_list = []
for key in global_entity:
if key != 'poi':
global_entity_list += [
item.lower().replace(' ', '_')
for item in global_entity[key]
]
else:
for item in global_entity[key]:
global_entity_list += [
item[x].lower().replace(' ', '_') for x in item
]
global_entity_list = list(set(global_entity_list))
for i, data_item in tqdm(enumerate(dev), total=len(dev)):
max_target_length = max(data_item['response_lengths'])
_, _, decoded_fine, decoded_coarse, global_ptr = self.encode_and_decode(
data_item, max_target_length, evaluating=True)
# decoded_fine是以一个batch的一个单词组成的列表为最内维度，所以倒置转化成行为一个完整的句子的预测疏输出
decoded_fine, decoded_coarse = map(
lambda x: np.transpose(np.array(x)),
(decoded_fine, decoded_coarse))
for bi, word_fine in enumerate(decoded_fine):
response_fine = ''
for e in word_fine:
if e == 'EOS':
break
response_fine += (e + ' ')
st_c = ''
for e in decoded_coarse[bi]:
if e == 'EOS':
break
else:
st_c += e + ' '
pred_sentence = response_fine.strip()
pred_sentence_coarse = st_c.strip()
pred.append(pred_sentence)
label_sentence = data_item['response_plain'][bi].strip(
) # 有一次bi会越界
label.append(label_sentence)
# 打印输出样例
# print('Context:')
# print(data_item['context_arr_plain'][bi])
# print('Predictive response:')
# print(pred_sentence)
# print('Label sentence:')
# print(label_sentence)
if args['dataset'] == 'kvr':
# compute F1 SCORE
single_f1, count = self.compute_prf(
data_item['ent_index'][bi], pred_sentence.split(),
global_entity_list, data_item['kb_info_plain'][bi])
F1_pred += single_f1
F1_count += count
single_f1, count = self.compute_prf(
data_item['ent_idx_cal'][bi], pred_sentence.split(),
global_entity_list, data_item['kb_info_plain'][bi])
F1_cal_pred += single_f1
F1_cal_count += count
single_f1, count = self.compute_prf(
data_item['ent_idx_nav'][bi], pred_sentence.split(),
global_entity_list, data_item['kb_info_plain'][bi])
F1_nav_pred += single_f1
F1_nav_count += count
single_f1, count = self.compute_prf(
data_item['ent_idx_wet'][bi], pred_sentence.split(),
global_entity_list, data_item['kb_info_plain'][bi])
F1_wet_pred += single_f1
F1_wet_count += count
else:
if pred_sentence == label_sentence:
acc += 1
dialogue_acc_dict[data_item['ID'][bi]].append(1)
else:
dialogue_acc_dict[data_item['ID'][bi]].append(0)
total += 1
if args['genSample']:
self.print_examples(bi, data_item, pred_sentence,
pred_sentence_coarse, label_sentence)
self.encoder.train(True)
self.ext_know.train(True)
self.decoder.train(True)
acc_score = acc / float(total)
print('TRAIN ACC SCORE:\t{}'.format(acc_score))
bleu_score = moses_multi_bleu(np.array(pred),
np.array(label),
lowercase=True) # 暂时无法使用
if args['dataset'] == 'kvr':
F1_score = F1_pred / float(F1_count)
print("F1 SCORE:\t{}".format(F1_pred / float(F1_count)))
print("\tCAL F1:\t{}".format(F1_cal_pred / float(F1_cal_count)))
print("\tWET F1:\t{}".format(F1_wet_pred / float(F1_wet_count)))
print("\tNAV F1:\t{}".format(F1_nav_pred / float(F1_nav_count)))
print("BLEU SCORE:\t" + str(bleu_score))
else:
dialogue_acc = 0
for key in dialogue_acc_dict:
if len(dialogue_acc_dict[key]) == sum(dialogue_acc_dict[key]):
dialogue_acc += 1
print("Dialog Accuracy:\t{}".format(dialogue_acc * 1.0 /
len(dialogue_acc_dict)))
if early_stop == 'BLEU':
if bleu_score >= metric_best:
self.save_model('BLEU-' + str(bleu_score))
print('MODEL SAVED')
return bleu_score
else:
if acc_score >= metric_best:
self.save_model('ACC-' + str(acc_score))
print('MODEL SAVED')
return acc_score
def evaluate(self, dev, avg_best, BLEU=False):
logging.info("STARTING EVALUATION")
acc_avg = 0.0
wer_avg = 0.0
bleu_avg = 0.0
acc_P = 0.0
acc_V = 0.0
microF1_PRED,microF1_PRED_cal,microF1_PRED_nav,microF1_PRED_wet = [],[],[],[]
microF1_TRUE,microF1_TRUE_cal,microF1_TRUE_nav,microF1_TRUE_wet = [],[],[],[]
# 在whole eval_dataset上计算的
ref = []
hyp = []
ref_s = ""
hyp_s = ""
dialog_acc_dict = {} # 在whole eval_dataset上计算的
pbar = tqdm(enumerate(dev), total=len(dev))
for j, data_dev in pbar:
if args['dataset'] == 'kvr':
'''
batch_size,
input_batches, input_lengths,
target_batches, target_lengths,
target_index,target_gate,
src_plain,
conv_seqs, conv_lengths'''
# output shape (T,B)
words = self.evaluate_batch(len(data_dev[1]), data_dev[0],
data_dev[1], data_dev[2],
data_dev[3], data_dev[4],
data_dev[5], data_dev[6],
data_dev[-2], data_dev[-1])
else:
words = self.evaluate_batch(len(data_dev[1]), data_dev[0],
data_dev[1], data_dev[2],
data_dev[3], data_dev[4],
data_dev[5], data_dev[6],
data_dev[-4], data_dev[-3])
# acc_P += acc_ptr
# acc_V += acc_vac
acc = 0 # 在one batch里计算的
w = 0
temp_gen = []
# Permute the dimensions of an array
for i, row in enumerate(np.transpose(words)): # shape (B,T)
st = ''
for e in row:
if e == '<EOS>': break
else: st += e + ' '
temp_gen.append(st)
# data_dev[7] may be the correct sentences; shape(B,T)
correct = data_dev[7][i] # this is response sentences
# compute F1 SCORE
if args['dataset'] == 'kvr':
f1_true, f1_pred = computeF1(data_dev[8][i],
st.lstrip().rstrip(),
correct.lstrip().rstrip())
microF1_TRUE += f1_true
microF1_PRED += f1_pred # 全部是1,估计用来做分母的,多余的？？
f1_true, f1_pred = computeF1(data_dev[9][i],
st.lstrip().rstrip(),
correct.lstrip().rstrip())
microF1_TRUE_cal += f1_true
microF1_PRED_cal += f1_pred
f1_true, f1_pred = computeF1(data_dev[10][i],
st.lstrip().rstrip(),
correct.lstrip().rstrip())
microF1_TRUE_nav += f1_true
microF1_PRED_nav += f1_pred
f1_true, f1_pred = computeF1(data_dev[11][i],
st.lstrip().rstrip(),
correct.lstrip().rstrip())
microF1_TRUE_wet += f1_true
microF1_PRED_wet += f1_pred
elif args['dataset'] == 'babi' and int(self.task) == 6:
f1_true, f1_pred = computeF1(data_dev[-2][i],
st.lstrip().rstrip(),
correct.lstrip().rstrip())
microF1_TRUE += f1_true
microF1_PRED += f1_pred
if args['dataset'] == 'babi':
# ID
if data_dev[-1][i] not in dialog_acc_dict.keys():
dialog_acc_dict[data_dev[-1][i]] = []
if (correct.lstrip().rstrip() == st.lstrip().rstrip()):
acc += 1 # 在one batch里计算的
dialog_acc_dict[data_dev[-1][i]].append(1)
else: # 在whole eval_dataset上计算的
dialog_acc_dict[data_dev[-1][i]].append(0)
else:
if (correct.lstrip().rstrip() == st.lstrip().rstrip()):
acc += 1
# print("Correct:"+str(correct.lstrip().rstrip()))
# print("\tPredict:"+str(st.lstrip().rstrip()))
# print("\tFrom:"+str(self.from_whichs[:,i]))
w += wer(correct.lstrip().rstrip(), st.lstrip().rstrip())
ref.append(str(correct.lstrip().rstrip()))
hyp.append(str(st.lstrip().rstrip()))
ref_s += str(correct.lstrip().rstrip()) + "\n"
hyp_s += str(st.lstrip().rstrip()) + "\n"
acc_avg += acc / float(len(
data_dev[1])) # len(data_dev[1]) = batch_size
wer_avg += w / float(len(data_dev[1])) # len(dev) = num of batches
# TODO: 有点不合理啊; 除以j应该比较合理;
pbar.set_description("R:{:.4f},W:{:.4f}".format(
acc_avg / float(len(dev)), wer_avg / float(len(dev))))
# dialog accuracy
if args['dataset'] == 'babi':
dia_acc = 0
for k in dialog_acc_dict.keys():
if len(dialog_acc_dict[k]) == sum(dialog_acc_dict[k]):
dia_acc += 1
logging.info("Dialog Accuracy:\t" +
str(dia_acc * 1.0 / len(dialog_acc_dict.keys())))
if args['dataset'] == 'kvr':
logging.info(
"F1 SCORE:\t" +
str(f1_score(microF1_TRUE, microF1_PRED, average='micro')))
logging.info("F1 CAL:\t" + str(
f1_score(microF1_TRUE_cal, microF1_PRED_cal, average='micro')))
logging.info("F1 WET:\t" + str(
f1_score(microF1_TRUE_wet, microF1_PRED_wet, average='micro')))
logging.info("F1 NAV:\t" + str(
f1_score(microF1_TRUE_nav, microF1_PRED_nav, average='micro')))
elif args['dataset'] == 'babi' and int(self.task) == 6:
logging.info(
"F1 SCORE:\t" +
str(f1_score(microF1_TRUE, microF1_PRED, average='micro')))
bleu_score = moses_multi_bleu(np.array(hyp),
np.array(ref),
lowercase=True)
logging.info("BLEU SCORE:" + str(bleu_score))
if (BLEU):
if (bleu_score >= avg_best):
self.save_model(str(self.name) + str(bleu_score))
logging.info("MODEL SAVED")
return bleu_score
else:
acc_avg = acc_avg / float(len(dev))
if (acc_avg >= avg_best):
self.save_model(str(self.name) + str(acc_avg))
logging.info("MODEL SAVED")
return acc_avg
def evaluate(self,
dev,
avg_best,
epoch,
BLEU=False,
Analyse=False,
type='dev'):
assert type == 'dev' or type == 'test'
logging.info("STARTING EVALUATION:{}".format(type))
acc_avg = 0.0
wer_avg = 0.0
bleu_avg = 0.0
acc_P = 0.0
acc_V = 0.0
microF1_PRED,microF1_PRED_cal,microF1_PRED_nav,microF1_PRED_wet = [],[],[],[]
microF1_TRUE,microF1_TRUE_cal,microF1_TRUE_nav,microF1_TRUE_wet = [],[],[],[]
ref = []
hyp = []
ref_s = ""
hyp_s = ""
dialog_acc_dict = {}
pbar = tqdm(enumerate(dev), total=len(dev))
if Analyse == True:
write_fp = write_to_disk('./vanilla-seq-generate.txt')
for j, data_dev in pbar:
words = self.evaluate_batch(len(data_dev[1]), data_dev[0],
data_dev[1], data_dev[2])
acc = 0
w = 0
temp_gen = []
#print(words)
for i, row in enumerate(np.transpose(words)):
st = ''
for e in row:
if e == '<EOS>':
break
else:
st += e + ' '
temp_gen.append(st)
correct = data_dev[7][i]
### compute F1 SCORE
if args['dataset'] == 'kvr':
f1_true, f1_pred = computeF1(data_dev[8][i],
st.lstrip().rstrip(),
correct.lstrip().rstrip())
microF1_TRUE += f1_true
microF1_PRED += f1_pred
f1_true, f1_pred = computeF1(data_dev[9][i],
st.lstrip().rstrip(),
correct.lstrip().rstrip())
microF1_TRUE_cal += f1_true
microF1_PRED_cal += f1_pred
f1_true, f1_pred = computeF1(data_dev[10][i],
st.lstrip().rstrip(),
correct.lstrip().rstrip())
microF1_TRUE_nav += f1_true
microF1_PRED_nav += f1_pred
f1_true, f1_pred = computeF1(data_dev[11][i],
st.lstrip().rstrip(),
correct.lstrip().rstrip())
microF1_TRUE_wet += f1_true
microF1_PRED_wet += f1_pred
elif args['dataset'] == 'babi' and int(self.task) == 6:
f1_true, f1_pred = computeF1(data_dev[-2][i],
st.lstrip().rstrip(),
correct.lstrip().rstrip())
microF1_TRUE += f1_true
microF1_PRED += f1_pred
if args['dataset'] == 'babi':
if data_dev[-1][i] not in dialog_acc_dict.keys():
dialog_acc_dict[data_dev[-1][i]] = []
if (correct.lstrip().rstrip() == st.lstrip().rstrip()):
acc += 1
dialog_acc_dict[data_dev[-1][i]].append(1)
else:
dialog_acc_dict[data_dev[-1][i]].append(0)
else:
if (correct.lstrip().rstrip() == st.lstrip().rstrip()):
acc += 1
#else:
# print("Correct:"+str(correct.lstrip().rstrip()))
# print("\tPredict:"+str(st.lstrip().rstrip()))
# print("\tFrom:"+str(self.from_whichs[:,i]))
# w += wer(correct.lstrip().rstrip(),st.lstrip().rstrip())
ref.append(str(correct.lstrip().rstrip()))
hyp.append(str(st.lstrip().rstrip()))
ref_s += str(correct.lstrip().rstrip()) + "\n"
hyp_s += str(st.lstrip().rstrip()) + "\n"
# write batch data to disk
if Analyse == True:
for gen, gold in zip(temp_gen, data_dev[7]):
write_fp.write(gen + '\t' + gold + '\n')
acc_avg += acc / float(len(data_dev[1]))
wer_avg += w / float(len(data_dev[1]))
pbar.set_description("R:{:.4f},W:{:.4f}".format(
acc_avg / float(len(dev)), wer_avg / float(len(dev))))
if Analyse == True:
write_fp.close()
if args['dataset'] == 'babi':
# TODO:计算平均的对话准确度
dia_acc = 0
for k in dialog_acc_dict.keys():
if len(dialog_acc_dict[k]) == sum(dialog_acc_dict[k]):
dia_acc += 1
logging.info("Dialog Accuracy:\t" +
str(dia_acc * 1.0 / len(dialog_acc_dict.keys())))
self._send_metrics(epoch,
type,
acc=dia_acc * 1.0 / len(dialog_acc_dict.keys()))
self._send_metrics(epoch,
type,
acc_response=acc_avg / float(len(dev)))
if args['dataset'] == 'kvr':
f1 = f1_score(microF1_TRUE, microF1_PRED, average='micro')
f1_cal = f1_score(microF1_TRUE_cal,
microF1_PRED_cal,
average='micro')
f1_wet = f1_score(microF1_TRUE_wet,
microF1_PRED_wet,
average='micro')
f1_nav = f1_score(microF1_TRUE_nav,
microF1_PRED_nav,
average='micro')
logging.info("F1 SCORE:\t" + str(f1))
logging.info("F1 CAL:\t" + str(f1_cal))
logging.info("F1 WET:\t" + str(f1_wet))
logging.info("F1 NAV:\t" + str(f1_nav))
self._send_metrics(epoch,
type,
f1=f1,
f1_cal=f1_cal,
f1_wet=f1_wet,
f1_nav=f1_nav)
elif args['dataset'] == 'babi' and int(self.task) == 6:
f1 = f1_score(microF1_TRUE, microF1_PRED, average='micro')
logging.info("F1 SCORE:\t" + str(f1))
self._send_metrics(epoch, type, babi_6_f1=f1)
self._send_metrics(epoch, type, total_loss=self.print_loss_avg)
bleu_score = moses_multi_bleu(np.array(hyp),
np.array(ref),
lowercase=True)
logging.info("BLEU SCORE:" + str(bleu_score))
self._send_metrics(epoch, type, bleu=bleu_score)
if (BLEU):
if (bleu_score >= avg_best):
if type == 'dev':
self.save_model(str(self.name) + str(bleu_score))
logging.info("MODEL SAVED")
return bleu_score
else:
acc_avg = acc_avg / float(len(dev))
if (acc_avg >= avg_best):
if type == 'dev':
self.save_model(str(self.name) + str(acc_avg))
logging.info("MODEL SAVED")
return acc_avg
def evaluate(self, dev, avg_best, epoch, BLEU=False, Analyse=False, type='dev'):
# Analyse 是在分析Mem中每个attn的score使用的
assert type == 'dev' or type == 'test'
logging.info("STARTING EVALUATION:{}".format(type))
acc_avg = 0.0
wer_avg = 0.0
bleu_avg = 0.0
acc_P = 0.0
acc_V = 0.0
microF1_PRED, microF1_PRED_cal, microF1_PRED_nav, microF1_PRED_wet = [], [], [], []
microF1_TRUE, microF1_TRUE_cal, microF1_TRUE_nav, microF1_TRUE_wet = [], [], [], []
ref = []
hyp = []
ref_s = ""
hyp_s = ""
dialog_acc_dict = {}
pbar = tqdm(enumerate(dev), total=len(dev))
if Analyse == True:
write_fp = write_to_disk('./multi-mem-generate.txt')
# 统计有多少数据是从memory中复制出来的
for j, data_dev in pbar:
words = self.evaluate_batch(batch_size=len(data_dev[1]),
input_batches=data_dev[0],
input_lengths=data_dev[1],
target_batches=data_dev[2],
target_lengths=data_dev[3],
target_index=data_dev[4],
target_gate=data_dev[5],
src_plain=data_dev[6],
conv_seqs=data_dev[-6],
conv_lengths=data_dev[-5],
kb_seqs=data_dev[-4],
kb_lengths=data_dev[-3],
kb_target_index=data_dev[-2],
kb_plain=data_dev[-1],
step=j,
Analyse=Analyse)
# acc_P += acc_ptr
# acc_V += acc_vac
acc = 0
w = 0
temp_gen = []
for i, row in enumerate(np.transpose(words)):
st = ''
for e in row:
if e == '<EOS>':
break
else:
st += e + ' '
temp_gen.append(st)
correct = data_dev[7][i]
### compute F1 SCORE
if args['dataset'] == 'kvr':
# TODO:Check this
f1_true, f1_pred = computeF1(data_dev[8][i], st.lstrip().rstrip(), correct.lstrip().rstrip())
microF1_TRUE += f1_true
microF1_PRED += f1_pred
f1_true, f1_pred = computeF1(data_dev[9][i], st.lstrip().rstrip(), correct.lstrip().rstrip())
microF1_TRUE_cal += f1_true
microF1_PRED_cal += f1_pred
f1_true, f1_pred = computeF1(data_dev[10][i], st.lstrip().rstrip(), correct.lstrip().rstrip())
microF1_TRUE_nav += f1_true
microF1_PRED_nav += f1_pred
f1_true, f1_pred = computeF1(data_dev[11][i], st.lstrip().rstrip(), correct.lstrip().rstrip())
microF1_TRUE_wet += f1_true
microF1_PRED_wet += f1_pred
elif args['dataset'] == 'babi' and int(self.task) == 6:
f1_true, f1_pred = computeF1(data_dev[-6][i], st.lstrip().rstrip(), correct.lstrip().rstrip())
microF1_TRUE += f1_true
microF1_PRED += f1_pred
if args['dataset'] == 'babi':
if data_dev[-5][i] not in dialog_acc_dict.keys():
dialog_acc_dict[data_dev[-5][i]] = []
if (correct.lstrip().rstrip() == st.lstrip().rstrip()):
acc += 1
dialog_acc_dict[data_dev[-5][i]].append(1)
else:
dialog_acc_dict[data_dev[-5][i]].append(0)
else:
if (correct.lstrip().rstrip() == st.lstrip().rstrip()):
acc += 1
# print("Correct:"+str(correct.lstrip().rstrip()))
# print("\tPredict:"+str(st.lstrip().rstrip()))
# print("\tFrom:"+str(self.from_whichs[:,i]))
# w += wer(correct.lstrip().rstrip(), st.lstrip().rstrip())
ref.append(str(correct.lstrip().rstrip()))
hyp.append(str(st.lstrip().rstrip()))
ref_s += str(correct.lstrip().rstrip()) + "\n"
hyp_s += str(st.lstrip().rstrip()) + "\n"
# write batch data to disk
if Analyse == True:
for gen, gold in zip(temp_gen, data_dev[7]):
write_fp.write(gen + '\t' + gold + '\n')
acc_avg += acc / float(len(data_dev[1]))
wer_avg += w / float(len(data_dev[1]))
pbar.set_description("R:{:.4f},W:{:.4f}".format(acc_avg / float(len(dev)),
wer_avg / float(len(dev))))
if Analyse == True:
write_fp.close()
# dialog accuracy
if args['dataset'] == 'babi':
dia_acc = 0
for k in dialog_acc_dict.keys():
if len(dialog_acc_dict[k]) == sum(dialog_acc_dict[k]):
dia_acc += 1
logging.info("Dialog Accuracy:\t" + str(dia_acc * 1.0 / len(dialog_acc_dict.keys())))
self._send_metrics(epoch, type, acc=dia_acc * 1.0 / len(dialog_acc_dict.keys()))
if args['dataset'] == 'kvr':
f1 = f1_score(microF1_TRUE, microF1_PRED, average='micro')
f1_cal = f1_score(microF1_TRUE_cal, microF1_PRED_cal, average='micro')
f1_wet = f1_score(microF1_TRUE_wet, microF1_PRED_wet, average='micro')
f1_nav = f1_score(microF1_TRUE_nav, microF1_PRED_nav, average='micro')
logging.info("F1 SCORE:\t" + str(f1))
logging.info("F1 CAL:\t" + str(f1_cal))
logging.info("F1 WET:\t" + str(f1_wet))
logging.info("F1 NAV:\t" + str(f1_nav))
self._send_metrics(epoch, type, f1=f1, f1_cal=f1_cal, f1_wet=f1_wet, f1_nav=f1_nav)
elif args['dataset'] == 'babi' and int(self.task) == 6:
f1 = f1_score(microF1_TRUE, microF1_PRED, average='micro')
logging.info("F1 SCORE:\t" + str(f1))
self._send_metrics(epoch, type, babi_6_f1=f1)
# Report Bleu score
bleu_score = moses_multi_bleu(np.array(hyp), np.array(ref), lowercase=True)
# Report Acc per response
self._send_metrics(epoch, type, acc_response=acc_avg / float(len(dev)))
logging.info("BLEU SCORE:" + str(bleu_score))
if Analyse == False:
# Send loss
self._send_metrics(epoch, type, total_loss=self.print_loss_avg,
ptr_loss=self.print_loss_kb,
vocab_loss=self.print_loss_vocabulary,
his_loss=self.print_loss_memory,
bleu_score=bleu_score)
if (BLEU):
if (bleu_score >= avg_best and bleu_score != 0):
if type == 'dev':
directory = self.save_model(str(self.name) + str(bleu_score))
locals_var = locals()
logging.info("MODEL SAVED")
return bleu_score
else:
acc_avg = acc_avg / float(len(dev))
if (acc_avg >= avg_best):
if type == 'dev':
locals_var = locals()
directory = self.save_model(str(self.name) + str(acc_avg))
logging.info("MODEL SAVED")
return acc_avg
else:
if (BLEU):
return bleu_score
else:
return acc_avg
def evaluate(self, dev, dev_length, matric_best, early_stop=None):
print('STARTING EVALUATION:')
ref, hyp = [], []
acc, total = 0, 0
dialog_acc_dict = {}
F1_pred, F1_cal_pred, F1_nav_pred, F1_wet_pred, F1_restaurant_pred, F1_hotel_pred, F1_attraction_pred, F1_train_pred, F1_hospital_pred = 0, 0, 0, 0, 0, 0, 0, 0, 0
F1_count, F1_cal_count, F1_nav_count, F1_wet_count, F1_restaurant_count, F1_hotel_count, F1_attraction_count, F1_train_count, F1_hospital_count = 0, 0, 0, 0, 0, 0, 0, 0, 0
pbar = tqdm(enumerate(dev.take(-1)), total=(dev_length))
new_precision, new_recall, new_f1_score = 0, 0, 0
if args['dataset'] == 'kvr':
with open('data/KVR/kvret_entities.json') as f:
global_entity = json.load(f)
global_entity_list = []
for key in global_entity.keys():
if key != 'poi':
global_entity_list += [
item.lower().replace(' ', '_')
for item in global_entity[key]
]
else:
for item in global_entity['poi']:
global_entity_list += [
item[k].lower().replace(' ', '_')
for k in item.keys()
]
global_entity_list = list(set(global_entity_list))
elif args['dataset'] == 'multiwoz':
with open('data/MULTIWOZ2.1/multiwoz_entities.json') as f:
global_entity = json.load(f)
global_entity_list = []
for key in global_entity.keys():
global_entity_list += [
item.lower().replace(' ', '_')
for item in global_entity[key]
]
global_entity_list = list(set(global_entity_list))
for j, data_dev in pbar:
# Encode and Decode
_, _, decoded_fine, decoded_coarse, global_pointer, global_pointer_logits = self.encode_and_decode(
data_dev, self.max_resp_len, False, True, False)
decoded_coarse = np.transpose(decoded_coarse)
decoded_fine = np.transpose(decoded_fine)
for bi, row in enumerate(decoded_fine):
st = ''
for e in row:
if e == 'EOS':
break
else:
st += e + ' '
st_c = ''
for e in decoded_coarse[bi]:
if e == 'EOS':
break
else:
st_c += e + ' '
pred_sent = st.lstrip().rstrip()
pred_sent_coarse = st_c.lstrip().rstrip()
gold_sent = data_dev[8][bi][0].numpy().decode().lstrip(
).rstrip() # data[8]: response_plain.
ref.append(gold_sent)
hyp.append(pred_sent)
if args['dataset'] == 'kvr':
# compute F1 SCORE
single_f1, count = self.compute_prf(
data_dev[14][bi], pred_sent.split(),
global_entity_list, data_dev[9]
[bi]) # data[14]: ent_index, data[9]: kb_arr_plain.
F1_pred += single_f1
F1_count += count
single_f1, count = self.compute_prf(
data_dev[16][bi], pred_sent.split(),
global_entity_list, data_dev[9]
[bi]) # data[16]: ent_idx_cal, data[9]: kb_arr_plain.
F1_cal_pred += single_f1
F1_cal_count += count
single_f1, count = self.compute_prf(
data_dev[17][bi], pred_sent.split(),
global_entity_list, data_dev[9]
[bi]) # data[17]: ent_idx_nav, data[9]: kb_arr_plain.
F1_nav_pred += single_f1
F1_nav_count += count
single_f1, count = self.compute_prf(
data_dev[18][bi], pred_sent.split(),
global_entity_list, data_dev[9]
[bi]) # data[18]: ent_idx_wet, data[9]: kb_arr_plain.
F1_wet_pred += single_f1
F1_wet_count += count
elif args['dataset'] == 'multiwoz':
# compute F1 SCORE
single_f1, count = self.compute_prf(
data_dev[14][bi], pred_sent.split(),
global_entity_list, data_dev[9]
[bi]) # data[14]: ent_index, data[9]: kb_arr_plain.
F1_pred += single_f1
F1_count += count
single_f1, count = self.compute_prf(
data_dev[28][bi], pred_sent.split(),
global_entity_list, data_dev[9][bi]
) # data[28]: ent_idx_restaurant, data[9]: kb_arr_plain.
F1_restaurant_pred += single_f1
F1_restaurant_count += count
single_f1, count = self.compute_prf(
data_dev[29][bi], pred_sent.split(),
global_entity_list, data_dev[9][bi]
) # data[29]: ent_idx_hotel, data[9]: kb_arr_plain.
F1_hotel_pred += single_f1
F1_hotel_count += count
single_f1, count = self.compute_prf(
data_dev[30][bi], pred_sent.split(),
global_entity_list, data_dev[9][bi]
) # data[30]: ent_idx_attraction, data[9]: kb_arr_plain.
F1_attraction_pred += single_f1
F1_attraction_count += count
single_f1, count = self.compute_prf(
data_dev[31][bi], pred_sent.split(),
global_entity_list, data_dev[9][bi]
) # data[31]: ent_idx_train, data[9]: kb_arr_plain.
F1_train_pred += single_f1
F1_train_count += count
else:
# compute Dialogue Accuracy Score
current_id = data_dev[22][bi]
if current_id not in dialog_acc_dict.keys():
dialog_acc_dict[current_id] = []
if gold_sent == pred_sent:
dialog_acc_dict[current_id].append(1)
else:
dialog_acc_dict[current_id].append(0)
# compute Per-response Accuracy Score
total += 1
if (gold_sent == pred_sent):
acc += 1
if args['genSample']:
self.print_examples(bi, data_dev, pred_sent,
pred_sent_coarse, gold_sent)
bleu_score = moses_multi_bleu(np.array(hyp),
np.array(ref),
lowercase=True)
acc_score = acc / float(total)
if args['dataset'] == 'kvr':
F1_score = F1_pred / float(F1_count)
print("F1 SCORE:\t{}".format(F1_pred / float(F1_count)))
print("\tCAL F1:\t{}".format(F1_cal_pred / float(F1_cal_count)))
print("\tWET F1:\t{}".format(F1_wet_pred / float(F1_wet_count)))
print("\tNAV F1:\t{}".format(F1_nav_pred / float(F1_nav_count)))
print("BLEU SCORE:\t" + str(bleu_score))
elif args['dataset'] == 'multiwoz':
F1_score = F1_pred / float(F1_count)
print("F1 SCORE:\t{}".format(F1_pred / float(F1_count)))
print("\tRES F1:\t{}".format(F1_restaurant_pred /
float(F1_restaurant_count)))
print("\tHOT F1:\t{}".format(F1_hotel_pred /
float(F1_hotel_count)))
print("\tATT F1:\t{}".format(F1_attraction_pred /
float(F1_attraction_count)))
print("\tTRA F1:\t{}".format(F1_train_pred /
float(F1_train_count)))
print("BLEU SCORE:\t" + str(bleu_score))
else:
dia_acc = 0
for k in dialog_acc_dict.keys():
if len(dialog_acc_dict[k]) == sum(dialog_acc_dict[k]):
dia_acc += 1
print("Dialog Accuracy:\t" +
str(dia_acc * 1.0 / len(dialog_acc_dict.keys())))
if (early_stop == 'BLEU'):
if (bleu_score >= matric_best):
self.save_model('BLEU-' + str(bleu_score))
print("MODEL SAVED")
return bleu_score
elif (early_stop == 'ENTF1'):
if (F1_score >= matric_best):
self.save_model('ENTF1-{:.4f}'.format(F1_score))
print("MODEL SAVED")
return F1_score
else:
if (acc_score >= matric_best):
self.save_model('ACC-{:.4f}'.format(acc_score))
print("MODEL SAVED")
return acc_score
def evaluate(self, dev, avg_best, BLEU=False):
logging.info("STARTING EVALUATION")
acc_avg = 0.0
wer_avg = 0.0
bleu_avg = 0.0
acc_P = 0.0
acc_V = 0.0
microF1_PRED, microF1_PRED_cal, microF1_PRED_nav, microF1_PRED_wet = 0, 0, 0, 0
microF1_TRUE, microF1_TRUE_cal, microF1_TRUE_nav, microF1_TRUE_wet = 0, 0, 0, 0
query = []
ref = []
hyp = []
ref_s = ""
hyp_s = ""
query_s = ""
dialog_acc_dict = {}
# only use for chitchat data
if args['dataset'] == 'chitchat':
global_entity_list = []
else:
raise ValueError("Must use chitchat data.")
pbar = tqdm(enumerate(dev), total=len(dev))
for j, data_dev in pbar:
# ???
if args['dataset'] == 'kvr':
words = self.evaluate_batch(len(data_dev[1]), data_dev[0], data_dev[1],
data_dev[2], data_dev[3], data_dev[4], data_dev[5], data_dev[6])
else:
words = self.evaluate_batch(len(data_dev[1]), data_dev[0], data_dev[1],
data_dev[2], data_dev[3], data_dev[4], data_dev[5], data_dev[6])
acc = 0
w = 0
temp_gen = []
for i, row in enumerate(np.transpose(words)):
st = ''
for e in row:
if e == '<EOS>':
break
else:
st += e + ' '
temp_gen.append(st)
# correct = " ".join([self.lang.index2word[item] for item in data_dev[2]])
correct = " ".join(data_dev[7][i])
user_query = " ".join(data_dev[6][i])
### compute F1 SCOR
st = st.lstrip().rstrip()
correct = correct.lstrip().rstrip()
user_query = user_query.strip()
'''
if args['dataset'] == 'kvr':
f1_true, count = self.compute_prf(data_dev[8][i], st.split(), global_entity_list, data_dev[14][i])
microF1_TRUE += f1_true`
microF1_PRED += count
f1_true, count = self.compute_prf(data_dev[9][i], st.split(), global_entity_list, data_dev[14][i])
microF1_TRUE_cal += f1_true
microF1_PRED_cal += count
f1_true, count = self.compute_prf(data_dev[10][i], st.split(), global_entity_list, data_dev[14][i])
microF1_TRUE_nav += f1_true
microF1_PRED_nav += count
f1_true, count = self.compute_prf(data_dev[11][i], st.split(), global_entity_list, data_dev[14][i])
microF1_TRUE_wet += f1_true
microF1_PRED_wet += count
elif args['dataset'] == 'babi' and int(args["task"]) == 6:
f1_true, count = self.compute_prf(data_dev[10][i], st.split(), global_entity_list, data_dev[12][i])
microF1_TRUE += f1_true
microF1_PRED += count
if args['dataset'] == 'babi':
if data_dev[11][i] not in dialog_acc_dict.keys():
dialog_acc_dict[data_dev[11][i]] = []
if (correct == st):
acc += 1
dialog_acc_dict[data_dev[11][i]].append(1)
else:
dialog_acc_dict[data_dev[11][i]].append(0)
else:
if (correct == st):
acc += 1
# print("Correct:"+str(correct))
# print("\tPredict:"+str(st))
# print("\tFrom:"+str(self.from_whichs[:,i]))
'''
w += wer(correct, st)
ref.append(str(correct))
hyp.append(str(st))
query.append(str(user_query))
ref_s += str(correct) + "\n"
hyp_s += str(st) + "\n"
query_s += str(user_query) + "\n"
# acc_avg += acc / float(len(data_dev[1]))
wer_avg += w / float(len(data_dev[1]))
pbar.set_description("W:{:.4f}".format(wer_avg / float(len(dev))))
# dialog accuracy
if args['dataset'] == 'babi':
dia_acc = 0
for k in dialog_acc_dict.keys():
if len(dialog_acc_dict[k]) == sum(dialog_acc_dict[k]):
dia_acc += 1
logging.info("Dialog Accuracy:\t" + str(dia_acc * 1.0 / len(dialog_acc_dict.keys())))
'''
if args['dataset'] == 'kvr':
logging.info("F1 SCORE:\t{}".format(microF1_TRUE / float(microF1_PRED)))
logging.info("\tCAL F1:\t{}".format(microF1_TRUE_cal / float(microF1_PRED_cal)))
logging.info("\tWET F1:\t{}".format(microF1_TRUE_wet / float(microF1_PRED_wet)))
logging.info("\tNAV F1:\t{}".format(microF1_TRUE_nav / float(microF1_PRED_nav)))
elif args['dataset'] == 'babi' and int(args["task"]) == 6:
logging.info("F1 SCORE:\t{}".format(microF1_TRUE / float(microF1_PRED)))
'''
# save decoding results
bleu_score = moses_multi_bleu(np.array(hyp), np.array(ref), lowercase=True)
logging.info("BLEU SCORE:" + str(bleu_score))
# always return with bleu score.
BLEU = True
if (BLEU):
if (bleu_score >= avg_best):
self.save_model(str(self.name) + str(bleu_score))
logging.info("MODEL SAVED")
self.save_decode_results(query, ref, hyp, bleu_score)
return bleu_score
'''
