def evaluate(data, model, name):
if name == "train":
instances = data.train_Ids
texts = data.train_texts
elif name == "dev":
instances = data.dev_Ids
texts = data.dev_texts
elif name == 'test':
instances = data.test_Ids
texts = data.test_texts
elif name == 'raw':
instances = data.raw_Ids
else:
print("Error: wrong evaluate name,", name)
right_token = 0
whole_token = 0
pred_results = []
gold_results = []
## set model in eval model
model.eval()
batch_size = 4
start_time = time.time()
train_num = len(instances)
total_batch = train_num // batch_size + 1
for batch_id in range(total_batch):
start = batch_id * batch_size
end = (batch_id + 1) * batch_size
if end > train_num:
end = train_num
instance = instances[start:end]
if not instance:
continue
gaz_list, reverse_gaz_list, batch_word, batch_biword, batch_wordlen, batch_wordrecover, batch_label, mask = batchify_with_label(
instance, data.HP_gpu, True)
tag_seq = model(gaz_list, reverse_gaz_list, batch_word, batch_wordlen,
mask)
pred_label, gold_label = recover_label(tag_seq, batch_label, mask,
data.label_alphabet,
batch_wordrecover)
pred_results += pred_label
gold_results += gold_label
decode_time = time.time() - start_time
speed = len(instances) / decode_time
if type == '-':
print('overall result of: {}'.format(name))
elif type == ".NAM":
print('NE result of: {}'.format(name))
elif type == ".NOM":
print('NM result of: {}'.format(name))
save_gold_pred(texts, pred_results, gold_results, name)
acc, p, r, f = get_ner_fmeasure(gold_results, pred_results, data.tagScheme)
print(
"time: %.2fs, speed: %.2fst/s; acc: %.4f, p: %.4f, r: %.4f, f: %.4f" %
(decode_time, speed, acc, p, r, f))
return speed, acc, p, r, f, pred_results
def evaluate(data, model, name):
if name == "train":
instances = data.train_Ids
elif name == "dev":
instances = data.dev_Ids
elif name == 'test':
instances = data.test_Ids
elif name == 'raw':
instances = data.raw_Ids
else:
print("Error: wrong evaluate name,", name)
right_token = 0
whole_token = 0
pred_results = []
gold_results = []
## set model in eval model
model.eval()
batch_size = 1
start_time = time.time()
train_num = len(instances)
total_batch = train_num // batch_size + 1
for batch_id in range(total_batch):
start = batch_id * batch_size
end = (batch_id + 1) * batch_size
if end > train_num:
end = train_num
instance = instances[start:end]
if not instance:
continue
gaz_list, batch_word, batch_biword, batch_wordlen, batch_wordrecover, batch_char, batch_charlen, batch_charrecover, batch_label, mask = batchify_with_label(
instance, data.HP_gpu, True)
tag_seq = model(gaz_list, batch_word, batch_biword, batch_wordlen,
batch_char, batch_charlen, batch_charrecover, mask)
# print "tag:",tag_seq
pred_label, gold_label = recover_label(tag_seq, batch_label, mask,
data.label_alphabet,
batch_wordrecover)
pred_results += pred_label
gold_results += gold_label
decode_time = time.time() - start_time
speed = len(instances) / decode_time
acc, p, r, f = get_ner_fmeasure(gold_results, pred_results, data.tagScheme)
return speed, acc, p, r, f, pred_results