def main():
# 载入数据集
trainset, validset, testset = [], [], []
if args.inference: # 测试时只载入测试集
with open(args.testset_path, 'r', encoding='utf8') as fr:
for line in fr:
testset.append(json.loads(line))
print('载入测试集%d条' % len(testset))
else: # 训练时载入训练集和验证集
with open(args.trainset_path, 'r', encoding='utf8') as fr:
for line in fr:
trainset.append(json.loads(line))
print('载入训练集%d条' % len(trainset))
with open(args.validset_path, 'r', encoding='utf8') as fr:
for line in fr:
validset.append(json.loads(line))
print('载入验证集%d条' % len(validset))
# 载入词汇表，词向量
vocab, embeds = [], []
with open(args.embed_path, 'r', encoding='utf8') as fr:
for line in fr:
line = line.strip()
word = line[:line.find(' ')]
vec = line[line.find(' ') + 1:].split()
embed = [float(v) for v in vec]
assert len(embed) == config.embedding_size # 检测词向量维度
vocab.append(word)
embeds.append(embed)
print('载入词汇表: %d个' % len(vocab))
print('词向量维度: %d' % config.embedding_size)
# 通过词汇表构建一个word2index和index2word的工具
sentence_processor = SentenceProcessor(vocab, config.pad_id,
config.start_id, config.end_id,
config.unk_id)
# 创建模型
model = Model(config)
epoch = 0 # 训练集迭代次数
global_step = 0 # 参数更新次数
# 载入模型
if os.path.isfile(args.model_path): # 如果载入模型的位置存在则载入模型
epoch, global_step = model.load_model(args.model_path)
print('载入模型完成')
# 记录模型的文件夹
log_dir = os.path.split(args.model_path)[0]
elif args.inference: # 如果载入模型的位置不存在，但是又要测试，这是没有意义的
print('请测试一个训练过的模型!')
return
else: # 如果载入模型的位置不存在，重新开始训练，则载入预训练的词向量
model.embedding.embedding.weight = torch.nn.Parameter(
torch.FloatTensor(embeds))
print('初始化模型完成')
# 记录模型的文件夹
log_dir = os.path.join(args.log_path, 'run' + str(int(time.time())))
if not os.path.exists(log_dir):
os.makedirs(log_dir)
if args.gpu:
model.to('cuda') # 将模型参数转到gpu
model.print_parameters() # 输出模型参数个数
# 定义优化器参数
optim = Optim(config.method, config.lr, config.lr_decay,
config.weight_decay, config.max_grad_norm)
optim.set_parameters(model.parameters()) # 给优化器设置参数
optim.update_lr(epoch) # 每个epoch更新学习率
# 训练
if not args.inference:
summary_writer = SummaryWriter(os.path.join(
log_dir, 'summary')) # 创建tensorboard记录的文件夹
dp_train = DataProcessor(trainset, config.batch_size,
sentence_processor) # 数据的迭代器
dp_valid = DataProcessor(validset,
config.batch_size,
sentence_processor,
shuffle=False)
while epoch < args.max_epoch: # 最大训练轮数
model.train() # 切换到训练模式
for data in dp_train.get_batch_data():
start_time = time.time()
feed_data = prepare_feed_data(data)
nll_loss, ppl = train(model, feed_data)
optim.optimizer.zero_grad() # 清空梯度
nll_loss.mean().backward() # 反向传播
optim.step() # 更新参数
use_time = time.time() - start_time
global_step += 1 # 参数更新次数+1
# summary当前情况
if global_step % args.print_per_step == 0:
print(
'epoch: %d, global_step: %d, lr: %g, nll_loss: %.2f, ppl: %.2f, time: %.2fs'
%
(epoch, global_step, optim.lr, nll_loss.mean().item(),
ppl.mean().exp().item(), use_time))
summary_writer.add_scalar('train_nll',
nll_loss.mean().item(),
global_step)
summary_writer.add_scalar('train_ppl',
ppl.mean().exp().item(),
global_step)
summary_writer.flush() # 将缓冲区写入文件
if global_step % args.log_per_step == 0: # 保存模型
log_file = os.path.join(
log_dir, '%03d%012d.model' % (epoch, global_step))
model.save_model(epoch, global_step, log_file)
# 验证集上计算困惑度
model.eval()
nll_loss, ppl = valid(model, dp_valid)
model.train()
print('在验证集上的nll损失为: %g, 困惑度为: %g' %
(nll_loss, np.exp(ppl)))
summary_writer.add_scalar('valid_nll', nll_loss,
global_step)
summary_writer.add_scalar('valid_ppl', np.exp(ppl),
global_step)
summary_writer.flush() # 将缓冲区写入文件
epoch += 1 # 数据集迭代次数+1
optim.update_lr(epoch) # 调整学习率
# 保存模型
log_file = os.path.join(log_dir,
'%03d%012d.model' % (epoch, global_step))
model.save_model(epoch, global_step, log_file)
# 验证集上计算困惑度
model.eval()
nll_loss, ppl = valid(model, dp_valid)
print('在验证集上的nll损失为: %g, 困惑度为: %g' % (nll_loss, np.exp(ppl)))
summary_writer.add_scalar('valid_nll', nll_loss, global_step)
summary_writer.add_scalar('valid_ppl', np.exp(ppl), global_step)
summary_writer.flush() # 将缓冲区写入文件
summary_writer.close()
else: # 测试
if not os.path.exists(args.result_path): # 创建结果文件夹
os.makedirs(args.result_path)
result_file = os.path.join(args.result_path, '%03d%012d.txt' %
(epoch, global_step)) # 命名结果文件
fw = open(result_file, 'w', encoding='utf8')
dp_test = DataProcessor(testset,
config.batch_size,
sentence_processor,
shuffle=False)
model.eval() # 切换到测试模式，会停用dropout等等
nll_loss, ppl = valid(model, dp_test) # 评估困惑度
print('在测试集上的nll损失为: %g, 困惑度为: %g' % (nll_loss, np.exp(ppl)))
len_results = [] # 统计生成结果的总长度
for data in dp_test.get_batch_data():
posts = data['str_posts']
responses = data['str_responses']
feed_data = prepare_feed_data(data, inference=True)
results = test(model, feed_data) # 使用模型计算结果 [batch, len_decoder]
for idx, result in enumerate(results):
new_data = {}
new_data['post'] = posts[idx]
new_data['response'] = responses[idx]
new_data['result'] = sentence_processor.index2word(
result) # 将输出的句子转回单词的形式
len_results.append(len(new_data['result']))
fw.write(json.dumps(new_data) + '\n')
fw.close()
print('生成句子平均长度: %d' % (1.0 * sum(len_results) / len(len_results)))
def main():
trainset, validset, testset = [], [], []
if args.inference: # 测试时只载入测试集
with open(args.testset_path, 'r', encoding='utf8') as fr:
for line in fr:
testset.append(json.loads(line))
print(f'载入测试集{len(testset)}条')
else: # 训练时载入训练集和验证集
with open(args.trainset_path, 'r', encoding='utf8') as fr:
for line in fr:
trainset.append(json.loads(line))
print(f'载入训练集{len(trainset)}条')
with open(args.validset_path, 'r', encoding='utf8') as fr:
for line in fr:
validset.append(json.loads(line))
print(f'载入验证集{len(validset)}条')
vocab, embeds = [], []
with open(args.embed_path, 'r', encoding='utf8') as fr:
for line in fr:
line = line.strip()
word = line[:line.find(' ')]
vec = line[line.find(' ') + 1:].split()
embed = [float(v) for v in vec]
assert len(embed) == config.embedding_size # 检测词向量维度
vocab.append(word)
embeds.append(embed)
print(f'载入词汇表: {len(vocab)}个')
print(f'词向量维度: {config.embedding_size}')
vads = []
with open(args.vad_path, 'r', encoding='utf8') as fr:
for line in fr:
line = line.strip()
vad = line[line.find(' ') + 1:].split()
vad = [float(item) for item in vad]
assert len(vad) == config.affect_embedding_size
vads.append(vad)
print(f'载入vad字典: {len(vads)}个')
print(f'vad维度: {config.affect_embedding_size}')
# 通过词汇表构建一个word2index和index2word的工具
sentence_processor = SentenceProcessor(vocab, config.pad_id,
config.start_id, config.end_id,
config.unk_id)
model = Model(config)
model.print_parameters() # 输出模型参数个数
epoch = 0 # 训练集迭代次数
global_step = 0 # 参数更新次数
# 载入模型
if os.path.isfile(args.model_path): # 如果载入模型的位置存在则载入模型
epoch, global_step = model.load_model(args.model_path)
model.affect_embedding.embedding.weight.requires_grad = False
print('载入模型完成')
log_dir = os.path.split(args.model_path)[0]
elif args.inference: # 如果载入模型的位置不存在，但是又要测试，这是没有意义的
print('请测试一个训练过的模型!')
return
else: # 如果载入模型的位置不存在，重新开始训练，则载入预训练的词向量
model.embedding.embedding.weight = torch.nn.Parameter(
torch.FloatTensor(embeds))
model.affect_embedding.embedding.weight = torch.nn.Parameter(
torch.tensor(vads).float())
model.affect_embedding.embedding.weight.requires_grad = False
print('初始化模型完成')
log_dir = os.path.join(args.log_path, 'run' + str(int(time.time())))
if not os.path.exists(log_dir):
os.makedirs(log_dir)
if args.gpu:
model.to('cuda') # 将模型参数转到gpu
# 定义优化器参数
optim = Optim(config.method, config.lr, config.lr_decay,
config.weight_decay, config.max_grad_norm)
optim.set_parameters(model.parameters()) # 给优化器设置参数
optim.update_lr(epoch) # 每个epoch更新学习率
# 训练
if not args.inference:
dp_train = DataProcessor(trainset, config.batch_size,
sentence_processor) # 数据的迭代器
dp_valid = DataProcessor(validset,
config.batch_size,
sentence_processor,
shuffle=False)
while epoch < args.max_epoch: # 最大训练轮数
model.train() # 切换到训练模式
for data in dp_train.get_batch_data():
start_time = time.time()
feed_data = prepare_feed_data(data)
nll_loss, precision = train(model, feed_data)
nll_loss.mean().backward() # 反向传播
optim.step() # 更新参数
optim.optimizer.zero_grad() # 清空梯度
use_time = time.time() - start_time
global_step += 1 # 参数更新次数+1
if global_step % args.print_per_step == 0:
print(
'epoch: {:d}, global_step: {:d}, lr: {:g}, nll_loss: {:.2f}, precision: {:.2%},'
' time: {:.2f}s/step'.format(epoch, global_step,
optim.lr,
nll_loss.mean().item(),
precision.mean().item(),
use_time))
epoch += 1 # 数据集迭代次数+1
optim.update_lr(epoch) # 调整学习率
log_file = os.path.join(
log_dir, '{:03d}{:012d}.model'.format(epoch, global_step))
model.save_model(epoch, global_step, log_file)
model.eval()
nll_loss, precision = valid(model, dp_valid)
print('在验证集上的NLL损失为: {:g}, 准确率为: {:.2%}'.format(
nll_loss, precision))
else: # 测试
if not os.path.exists(args.result_path): # 创建结果文件夹
os.makedirs(args.result_path)
result_file = os.path.join(args.result_path,
'{:03d}{:012d}.txt'.format(
epoch, global_step)) # 命名结果文件
fw = open(result_file, 'w', encoding='utf8')
dp_test = DataProcessor(testset,
config.batch_size,
sentence_processor,
shuffle=False)
model.eval()
nll_loss, precision = valid(model, dp_test)
print('在测试集上的NLL损失为: {:g}, 准确率为: {:.2%}'.format(nll_loss, precision))
for data in dp_test.get_batch_data():
texts = data['str_texts']
emotions = data['emotions']
feed_data = prepare_feed_data(data)
results = test(model, feed_data) # 使用模型计算结果 [batch]
for idx, result in enumerate(results):
new_data = dict()
new_data['text'] = texts[idx]
new_data['emotion'] = emotions[idx]
new_data['result'] = result # 将输出的句子转回单词的形式
fw.write(json.dumps(new_data, ensure_ascii=False) + '\n')
fw.close()
def main():
trainset, validset, testset = [], [], []
if args.inference: # 测试时只载入测试集
with open(args.testset_path, 'r', encoding='utf8') as fr:
for line in fr:
testset.append(json.loads(line))
print(f'载入测试集{len(testset)}条')
else: # 训练时载入训练集和验证集
with open(args.trainset_path, 'r', encoding='utf8') as fr:
for line in fr:
trainset.append(json.loads(line))
print(f'载入训练集{len(trainset)}条')
with open(args.validset_path, 'r', encoding='utf8') as fr:
for line in fr:
validset.append(json.loads(line))
print(f'载入验证集{len(validset)}条')
vocab, embeds = [], []
with open(args.embed_path, 'r', encoding='utf8') as fr:
for line in fr:
line = line.strip()
word = line[:line.find(' ')]
vec = line[line.find(' ') + 1:].split()
embed = [float(v) for v in vec]
assert len(embed) == config.embedding_size # 检测词向量维度
vocab.append(word)
embeds.append(embed)
print(f'载入词汇表: {len(vocab)}个')
print(f'词向量维度: {config.embedding_size}')
vads = []
with open(args.vad_path, 'r', encoding='utf8') as fr:
for line in fr:
line = line.strip()
vad = line[line.find(' ') + 1:].split()
vad = [float(item) for item in vad]
assert len(vad) == config.affect_embedding_size
vads.append(vad)
print(f'载入vad字典: {len(vads)}个')
print(f'vad维度: {config.affect_embedding_size}')
# 通过词汇表构建一个word2index和index2word的工具
sentence_processor = SentenceProcessor(vocab, config.pad_id,
config.start_id, config.end_id,
config.unk_id)
model = Model(config)
model.print_parameters() # 输出模型参数个数
epoch = 0 # 训练集迭代次数
global_step = 0 # 参数更新次数
# 载入模型
if os.path.isfile(args.model_path): # 如果载入模型的位置存在则载入模型
epoch, global_step = model.load_model(args.model_path)
model.affect_embedding.embedding.weight.requires_grad = False
print('载入模型完成')
log_dir = os.path.split(args.model_path)[0]
elif args.inference: # 如果载入模型的位置不存在，但是又要测试，这是没有意义的
print('请测试一个训练过的模型!')
return
else: # 如果载入模型的位置不存在，重新开始训练，则载入预训练的词向量
model.embedding.embedding.weight = torch.nn.Parameter(
torch.FloatTensor(embeds))
model.affect_embedding.embedding.weight = torch.nn.Parameter(
torch.FloatTensor(vads))
model.affect_embedding.embedding.weight.requires_grad = False
print('初始化模型完成')
log_dir = os.path.join(args.log_path, 'run' + str(int(time.time())))
if not os.path.exists(log_dir):
os.makedirs(log_dir)
if args.gpu:
model.to('cuda') # 将模型参数转到gpu
# 定义优化器参数
optim = Optim(config.method, config.lr, config.lr_decay,
config.weight_decay, config.eps, config.max_grad_norm)
optim.set_parameters(model.parameters()) # 给优化器设置参数
optim.update_lr(epoch) # 每个epoch更新学习率
# 训练
if not args.inference:
summary_writer = SummaryWriter(os.path.join(
log_dir, 'summary')) # 创建tensorboard记录的文件夹
dp_train = DataProcessor(trainset, config.batch_size,
sentence_processor) # 数据的迭代器
dp_valid = DataProcessor(validset,
config.batch_size,
sentence_processor,
shuffle=False)
while epoch < args.max_epoch: # 最大训练轮数
model.train() # 切换到训练模式
for data in dp_train.get_batch_data():
start_time = time.time()
feed_data = prepare_feed_data(data)
rl_loss, reward, loss, nll_loss, kld_loss, kld_weight, ppl = \
train(model, feed_data, global_step)
if args.reinforce:
rl_loss.mean().backward()
else:
loss.mean().backward() # 反向传播
optim.step() # 更新参数
optim.optimizer.zero_grad() # 清空梯度
use_time = time.time() - start_time
global_step += 1 # 参数更新次数+1
if global_step % args.print_per_step == 0:
print(
'epoch: {:d}, global_step: {:d}, lr: {:g}, rl_loss: {:.2f}, reward: {:.2f}, nll_loss: {:.2f},'
' kld_loss: {:.2f}, kld_weight: {:g}, ppl: {:.2f}, time: {:.2f}s/step'
.format(epoch, global_step, optim.lr,
rl_loss.mean().item(),
reward.mean().item(),
nll_loss.mean().item(),
kld_loss.mean().item(), kld_weight,
ppl.mean().exp().item(), use_time))
summary_writer.add_scalar('train_rl',
rl_loss.mean().item(),
global_step)
summary_writer.add_scalar('train_reward',
reward.mean().item(),
global_step)
summary_writer.add_scalar('train_nll',
nll_loss.mean().item(),
global_step)
summary_writer.add_scalar('train_kld',
kld_loss.mean().item(),
global_step)
summary_writer.add_scalar('train_weight', kld_weight,
global_step)
summary_writer.add_scalar('train_ppl',
ppl.mean().exp().item(),
global_step)
summary_writer.flush() # 将缓冲区写入文件
if global_step % args.log_per_step == 0 \
or (global_step % (2*config.kl_step) - config.kl_step) == config.kl_step // 2:
log_file = os.path.join(
log_dir,
'{:03d}{:012d}.model'.format(epoch, global_step))
model.save_model(epoch, global_step, log_file)
model.eval()
reward, nll_loss, kld_loss, ppl = valid(
model, dp_valid, global_step - 1)
model.train()
print(
'在验证集上的REWARD为: {:g}, NLL损失为: {:g}, KL损失为: {:g}, PPL为: {:g}'
.format(reward, nll_loss, kld_loss, np.exp(ppl)))
summary_writer.add_scalar('valid_reward', reward,
global_step)
summary_writer.add_scalar('valid_nll', nll_loss,
global_step)
summary_writer.add_scalar('valid_kld', kld_loss,
global_step)
summary_writer.add_scalar('valid_ppl', np.exp(ppl),
global_step)
summary_writer.flush() # 将缓冲区写入文件
epoch += 1 # 数据集迭代次数+1
optim.update_lr(epoch) # 调整学习率
log_file = os.path.join(
log_dir, '{:03d}{:012d}.model'.format(epoch, global_step))
model.save_model(epoch, global_step, log_file)
model.eval()
reward, nll_loss, kld_loss, ppl = valid(model, dp_valid,
global_step - 1)
print('在验证集上的REWARD为: {:g}, NLL损失为: {:g}, KL损失为: {:g}, PPL为: {:g}'.
format(reward, nll_loss, kld_loss, np.exp(ppl)))
summary_writer.add_scalar('valid_reward', reward, global_step)
summary_writer.add_scalar('valid_nll', nll_loss, global_step)
summary_writer.add_scalar('valid_kld', kld_loss, global_step)
summary_writer.add_scalar('valid_ppl', np.exp(ppl), global_step)
summary_writer.flush() # 将缓冲区写入文件
summary_writer.close()
else: # 测试
if not os.path.exists(args.result_path): # 创建结果文件夹
os.makedirs(args.result_path)
result_file = os.path.join(args.result_path,
'{:03d}{:012d}.txt'.format(
epoch, global_step)) # 命名结果文件
fw = open(result_file, 'w', encoding='utf8')
dp_test = DataProcessor(testset,
config.batch_size,
sentence_processor,
shuffle=False)
model.eval() # 切换到测试模式，会停用dropout等等
reward, nll_loss, kld_loss, ppl = valid(model, dp_test,
global_step - 1)
print('在测试集上的REWARD为: {:g}, NLL损失为: {:g}, KL损失为: {:g}, PPL为: {:g}'.
format(reward, nll_loss, kld_loss, np.exp(ppl)))
len_results = [] # 统计生成结果的总长度
for data in dp_test.get_batch_data():
posts = data['str_posts']
responses = data['str_responses']
feed_data = prepare_feed_data(data, inference=True)
results = test(model, feed_data) # 使用模型计算结果 [batch, len_decoder]
for idx, result in enumerate(results):
new_data = dict()
new_data['post'] = posts[idx]
new_data['response'] = responses[idx]
new_data['result'] = sentence_processor.index2word(
result) # 将输出的句子转回单词的形式
len_results.append(len(new_data['result']))
fw.write(json.dumps(new_data, ensure_ascii=False) + '\n')
fw.close()
print(f'生成句子平均长度: {1.0 * sum(len_results) / len(len_results)}')
def filter_by_emotion(args):
vocab, embeds = [], []
with open(args.embed_path, 'r', encoding='utf8') as fr:
for line in fr:
line = line.strip()
word = line[:line.find(' ')]
vec = line[line.find(' ') + 1:].split()
embed = [float(v) for v in vec]
assert len(embed) == config.embedding_size # 检测词向量维度
vocab.append(word)
embeds.append(embed)
print(f'载入词汇表: {len(vocab)}个')
print(f'词向量维度: {config.embedding_size}')
vads = []
with open(args.vad_path, 'r', encoding='utf8') as fr:
for line in fr:
line = line.strip()
vad = line[line.find(' ') + 1:].split()
vad = [float(item) for item in vad]
assert len(vad) == config.affect_embedding_size
vads.append(vad)
print(f'载入vad字典: {len(vads)}个')
print(f'vad维度: {config.affect_embedding_size}')
sentence_processor = SentenceProcessor(vocab, config.pad_id,
config.start_id, config.end_id,
config.unk_id)
model = Model(config)
if args.gpu:
model.cuda()
num = 0
with open(args.data_path, 'r', encoding='utf8') as fr:
with open(args.save_path, 'w', encoding='utf8') as fw:
for line in fr:
data = json.loads(line.strip())
post = data['post']
response = data['response']
id_post, len_post = sentence_processor.word2index(post)
id_response, len_response = sentence_processor.word2index(
response)
max_len = max(len_post, len_response) + 2
id_post = sentence_processor.pad_sentence(id_post, max_len)
id_response = sentence_processor.pad_sentence(
id_response, max_len)
texts = [id_post, id_response]
lengths = [len_post + 2, len_response + 2]
feed_data = {
'x': torch.tensor(texts).long(),
'len_x': torch.tensor(lengths).long()
}
if args.gpu:
for key, value in feed_data.items():
feed_data[key] = value.cuda()
result = model(feed_data).argmax(1).detach().tolist()
if sum(result) != 0:
new_data = {
'post': post,
'post_e': result[0],
'response': response,
'response_e': result[1]
}
fw.write(json.dumps(new_data, ensure_ascii=False) + '\n')
num += 1
print(f'剩余{num}条数据')