def bilstm_train_and_eval(train_data,
dev_data,
test_data,
charEmbedding,
word2id,
tag2id,
crf=True,
remove_O=False):
train_word_lists, train_tag_lists = train_data
dev_word_lists, dev_tag_lists = dev_data
test_word_lists, test_tag_lists = test_data
start = time.time()
vocab_size = len(word2id)
out_size = len(tag2id)
bilstm_model = BILSTM_Model(vocab_size, charEmbedding, out_size, crf=crf)
bilstm_model.train(train_word_lists, train_tag_lists, dev_word_lists,
dev_tag_lists, word2id, tag2id)
model_name = "bilstm_crf" if crf else "bilstm"
save_model(
bilstm_model,
"./ckpts/" + str(bilstm_model.best_val_loss)[:5] + model_name + ".pkl")
print("训练完毕,共用时{}秒.".format(int(time.time() - start)))
print("评估{}模型中...".format(model_name))
pred_tag_lists, test_tag_lists = bilstm_model.test(test_word_lists,
test_tag_lists, word2id,
tag2id)
f = open(
'next_dev_result/' + str(bilstm_model.best_val_loss)[:5] +
'_bilstmcrf_result.txt', 'w')
for pred_tag_list in pred_tag_lists:
f.write(' '.join(pred_tag_list) + '\n')
f.close()
def bilstm_train_and_eval(train_data,
dev_data,
test_data,
word2id,
tag2id,
output_dir,
crf=True,
remove_O=False):
train_word_lists, train_tag_lists = train_data
dev_word_lists, dev_tag_lists = dev_data
test_word_lists, test_tag_lists = test_data
start = time.time()
vocab_size = len(word2id)
out_size = len(tag2id)
bilstm_model = BILSTM_Model(vocab_size, out_size, crf=crf)
bilstm_model.train(train_word_lists, train_tag_lists, dev_word_lists,
dev_tag_lists, word2id, tag2id)
model_name = "bilstm_crf" if crf else "bilstm"
save_model(bilstm_model, os.path.join(output_dir, model_name + ".pkl"))
print("训练完毕,共用时{}秒.".format(int(time.time() - start)))
print("评估{}模型中...".format(model_name))
pred_tag_lists, test_tag_lists = bilstm_model.test(test_word_lists,
test_tag_lists, word2id,
tag2id)
metrics = Metrics(test_tag_lists, pred_tag_lists, remove_O=remove_O)
metrics.report_scores()
metrics.report_confusion_matrix()
return pred_tag_lists
def bilstm_train_and_eval(train_data,
dev_data,
test_data,
word2id,
tag2id,
crf=True):
train_word_lists, train_tag_lists = train_data
dev_word_lists, dev_tag_lists = dev_data
test_word_lists, test_tag_lists = test_data
start = time.time()
vocab_size = len(word2id)
out_size = len(tag2id)
bilstm_model = BILSTM_Model(vocab_size, out_size, crf=crf)
bilstm_model.train(train_word_lists, train_tag_lists, dev_word_lists,
dev_tag_lists, word2id, tag2id)
model_name = "bilstm_crf" if crf else "bilstm"
save_model(bilstm_model, "./ckpts/" + model_name + ".pkl")
print("训练完毕,共用时{}秒.".format(int(time.time() - start)))
print("评估{}模型中...".format(model_name))
pred_tag_lists, test_tag_lists = bilstm_model.test(test_word_lists,
test_tag_lists, word2id,
tag2id)
metrics = Metrics(test_tag_lists, pred_tag_lists)
return pred_tag_lists
def bilstm_train_and_eval(train_data,
dev_data,
test_data,
word2id,
tag2id,
crf=True,
remove_O=False):
train_word_lists, train_tag_lists = train_data
dev_word_lists, dev_tag_lists = dev_data
test_word_lists, test_tag_lists = test_data
start = time.time()
vocab_size = len(word2id)
out_size = len(tag2id)
bilstm_model = BILSTM_Model(vocab_size, out_size, crf=crf)
bilstm_model.train(train_word_lists, train_tag_lists, dev_word_lists,
dev_tag_lists, word2id, tag2id)
model_name = "bilstm_crf" if crf else "bilstm"
save_model(bilstm_model, "./ckpts/" + model_name + ".pkl")
print("Training completed, {} seconds when sharing.".format(
int(time.time() - start)))
print("Evaluation{} model:...".format(model_name))
pred_tag_lists, test_tag_lists = bilstm_model.test(test_word_lists,
test_tag_lists, word2id,
tag2id)
metrics = Metrics(test_tag_lists, pred_tag_lists, remove_O=remove_O)
metrics.report_scores()
metrics.report_confusion_matrix()
return pred_tag_lists
def bilstm_train_and_eval(train_loader, dev_loader, eval_loader,
test_loader, token2id, tag2id, method):
"""训练并保存模型"""
vocab_size = len(token2id)
out_size = len(tag2id)
meta = get_meta([TrainingConfig.__dict__, LSTMConfig.__dict__])
model = BILSTM_Model(vocab_size, out_size, token2id, tag2id, method=method)
model.train(train_loader, dev_loader, eval_loader)
try:
# 保存模型的信息
root_dir = "/home/luopx/share_folders/Sohu"
model_dir = 'ckpts/{}/{}-{}-Len{}-{:.2f}-{:.4f}'.format(
model.method,
meta['token_method'],
meta['tag_schema'],
meta['max_len'],
model.best_val_loss,
model.best_f1_score
)
model_dir = join(root_dir, model_dir)
if not os.path.isdir(model_dir):
os.mkdir(model_dir)
save_model(model, join(model_dir, "model.pkl"))
# 保存word2id tag2id 以及模型设置的信息
with open(join(model_dir, 'meta.json'), 'w') as w:
w.write(json.dumps(meta, indent=4))
# 在验证集上面观察模型的效果、特点
print("评估{}模型中...".format(method))
# 分析结果
print("分析在验证集上的结果...")
metrics = model.cal_scores(eval_loader, use_model='best_f1')
with open(join(model_dir, 'dev_result.txt'), 'w') as outfile:
metrics.report_details(outfile=outfile)
# 加载测试集，解码，将结果保存成文件
print("在val_loss最小的模型上解码...")
test_result = join(model_dir, 'min_devLoss_result.txt')
decoding(model, test_loader, test_result)
print("在f1分数值最大的模型上解码...")
test_result = join(model_dir, 'max_f1_result.txt')
decoding(model, test_loader, test_result, use_model="best_f1")
except:
import pdb
pdb.set_trace()
def bilstm_pred(train_word_lists, train_tag_lists, dev_word_lists,
dev_tag_lists, test_word_lists, test_tag_lists):
start = time.time()
vocab_size = len(bilstm_word2id)
out_size = len(bilstm_tag2id)
model = BILSTM_Model(vocab_size, out_size, crf=False)
model.train(train_word_lists, train_tag_lists, dev_word_lists,
dev_tag_lists, bilstm_word2id, bilstm_tag2id)
save_model(model, "./ckpts/lstm.pkl")
print("训练完毕,共用时{}秒.".format(int(time.time() - start)))
pred_tag_lists, test_tag_lists = model.test(test_word_lists,
test_tag_lists, word2id,
tag2id)
return pred_tag_lists
def bilstm_train_and_eval(train_data,
dev_data,
test_data,
word2id,
tag2id,
crf=True,
remove_O=False,
reload_model=False):
# data
train_word_lists, train_tag_lists = train_data
dev_word_lists, dev_tag_lists = dev_data
test_word_lists, test_tag_lists = test_data
# training
start = time.time()
vocab_size = len(word2id)
out_size = len(tag2id)
# get model_file
if crf:
model_name = "bilstm_crf"
else:
model_name = "bilstm"
emb_size = LSTMConfig.emb_size
hidden_size = LSTMConfig.hidden_size
model_file = "./weights/" + model_name + '_' + str(emb_size) + '_' + str(
hidden_size) + ".pkl"
if reload_model:
# reload trained model!
bilstm_model = load_model(model_file)
else:
# train and save model!
bilstm_model = BILSTM_Model(vocab_size, out_size, crf=crf)
bilstm_model.train(train_word_lists, train_tag_lists, dev_word_lists,
dev_tag_lists, word2id, tag2id)
save_model(
bilstm_model, model_file
) # re-thinking when to save the model? after valid for each epoch?
print("Training finished, taken {} seconds!".format(
int(time.time() - start)))
print("Evaluating {} model:".format(model_name))
pred_tag_lists, test_tag_lists = bilstm_model.test(test_word_lists,
test_tag_lists, word2id,
tag2id)
results_print(test_tag_lists, pred_tag_lists, remove_O=remove_O)
return pred_tag_lists
def bilstm_train_and_eval(train_data,
dev_data,
test_data,
word2id,
data2id,
crf=True,
remove_O=False):
train_word_lists, train_data_lists, train_wordlabel_lists, train_datalabel_lists, train_dataptr_lists = train_data
dev_word_lists, dev_data_lists, dev_wordlabel_lists, dev_datalabel_lists, dev_dataptr_lists = dev_data
test_word_lists, test_data_lists, test_wordlabel_lists, test_datalabel_lists, test_dataptr_lists = test_data
start = time.time()
vocab_size = len(word2id)
data_size = len(data2id)
bilstm_model = BILSTM_Model(vocab_size, data_size, crf=crf)
bilstm_model.train(train_word_lists, train_data_lists,
train_wordlabel_lists, train_datalabel_lists,
train_dataptr_lists, dev_word_lists, dev_data_lists,
dev_wordlabel_lists, dev_datalabel_lists,
dev_dataptr_lists, word2id, data2id)
model_name = "bilstm_crf" if crf else "bilstm"
save_model(bilstm_model, "./ckpts/" + model_name + ".pkl")
print("训练完毕,共用时{}秒.".format(int(time.time() - start)))
print("评估{}模型中...".format(model_name))
pred_tag_lists, pred_tag_lists = bilstm_model.test(
test_word_lists, test_data_lists, test_wordlabel_lists,
test_datalabel_lists, test_dataptr_lists, word2id, data2id)
#for pred, gold in zip(pred_tag_lists, test_tag_lists):
#print(pred, gold)
allnum = 0
correct = 0
for pred, gold in zip(pred_tag_lists, pred_tag_lists):
pred = pred[:len(gold)]
for x, y in zip(pred, gold):
if x == y:
correct += 1
allnum += 1
print(correct / allnum)
return correct / allnum
def bilstm_train_and_eval(train_data,
dev_data,
test_data,
word2id,
tag2id,
crf=True,
remove_O=False):
train_word_lists, train_tag_lists = train_data
dev_word_lists, dev_tag_lists = dev_data
test_word_lists, test_tag_lists = test_data
start = time.time()
vocab_size = len(word2id)
out_size = len(tag2id)
bilstm_model = BILSTM_Model(vocab_size, out_size, crf=crf)
bilstm_model.train(train_word_lists, train_tag_lists, dev_word_lists,
dev_tag_lists, word2id, tag2id)
model_name = "bilstm_crf" if crf else "bilstm"
save_model(bilstm_model, "./ckpts/" + model_name + ".pkl")
print("训练完毕,共用时{}秒.".format(int(time.time() - start)))
print("评估{}模型中...".format(model_name))
pred_tag_lists, test_tag_lists = bilstm_model.test(test_word_lists,
test_tag_lists, word2id,
tag2id)
with open("./result.txt", "a+") as f:
for i in range(len(pred_tag_lists)):
f.write(pred_tag_lists[i] + " " + pred_tag_lists[i] + "\n")
metrics = Metrics(test_tag_lists, pred_tag_lists, remove_O=remove_O)
metrics.report_scores()
metrics.report_confusion_matrix()
return pred_tag_lists