PyTorch深度学习实践-P13循环神经网络（高级篇）

用RNN做一个分类器~

• 根据名字来判断国家18个语言地区，几千个名字，训练模型，输入新名字，然后告诉我们是使用哪种语言的人
• 复习：onehot通过嵌入层，然后RNN层，隐状态~做线性层，把输出映射成我们要的维度

•  hN是最后一个隐层

• 输入名字M c l e a n是一个序列，而且序列长短不一样

• 我们把嵌入层RNN 线性层 定义名叫RNNClassifier

• 主循环：

• 

• if __name__ == '__main__':
      #N_CHARS字符数量输入英文字母转换成onehot向量、隐层数量、country分类一共多少、layerGRU几层
      classifier = RNNClassifier(N_CHARS, HIDDEN_SIZE, N_COUNTRY, N_LAYER)
      if USE_GPU:
          device = torch.device("cuda:0")
          classifier.to(device)
      #损失函数、优化器
      criterion = torch.nn.CrossEntropyLoss()
      optimizer = torch.optim.Adam(classifier.parameters(), lr=0.001)
      #打印训练时间的长度
      start = time.time()
      print("Training for %d epochs..." % N_EPOCHS)
      #recording the accuracy of testing
      acc_list = []
      for epoch in range(1, N_EPOCHS + 1):
          # Train cycle 把训练和测试封装在两个函数里
          trainModel()
          acc = testModel()
          acc_list.append(acc)

  还有一个函数：

• #距离训练开始的时间，python中以秒为单位，除60变成分
      def time_since(since):
          s = time.time() - since
          m = math.floor(s / 60)
          s -= m * 60
          return '%dm %ds' % (m, s)
  

  1准备数据

• 名字的处理：

• 每个字符对应一个ASCII值0-127

• 序列长度不一样，做padding让他们变成一样长度

• 国家处理：

• 把国家名变成分类索引，从0开始索引标签~做一个对应的词典~

• 代码：

• # Parameters
  HIDDEN_SIZE = 100
  BATCH_SIZE = 256
  N_LAYER = 2
  N_EPOCHS = 100
  N_CHARS = 128
  USE_GPU = False
  
  
  #prepare data
  class NameDataset(Dataset):
      def __init__(self, is_train_set=True):
          #reading data frfom .gz file with package gzip and csv
          filename = 'data/names_train.csv.gz' if is_train_set else 'data/names_test.csv.gz'
          with gzip.open(filename, 'rt') as f:
              reader = csv.reader(f)
              rows = list(reader)
              #save names and countries in list
              self.names = [row[0] for row in rows]
              self.len = len(self.names)
              self.countries = [row[1] for row in rows]
              #save countries and its index in list and dictionary
              #sorted 去除重复的国家名
              self.country_list = list(sorted(set(self.countries)))
              #根据列表 转成词典
              self.country_dict = self.getCountryDict()
              self.country_num = len(self.country_list)
  
      # save countries and its index in list and dictionary
      def __getitem__(self, index):
          return self.names[index], self.country_dict[self.countries[index]]
      #return length of dataset
      def __len__(self):
          return self.len
  
      #convert list into dictionary
      def getCountryDict(self):
          country_dict = dict()
  
      for idx, country_name in enumerate(self.country_list, 0):
          country_dict[country_name] = idx
          return country_dict
      #根据索引返回国家名
      def idx2country(self, index):
          return self.country_list[index]
      #得到国家的数量
      def getCountriesNum(self):
          return self.country_nu
  
      trainset = NameDataset(is_train_set=True)
      trainloader = DataLoader(trainset, batch_size=BATCH_SIZE, shuffle=True)
      testset = NameDataset(is_train_set=False)
      testloader = DataLoader(testset, batch_size=BATCH_SIZE, shuffle=False)
      #N_COUNTRY is output size of our model
      N_COUNTRY = trainset.getCountriesNum()

  

   2模型设置

• 什么是Bi-direction RNN/LSTM/GRU
• 在自然语言中 考虑后文的词的影响
• forward沿着序列方向，反向一次，然后把正向反向的隐层做拼接~
• 输出：h0...hN
• hidden=[]
• 

 

 代码：

#model design
class RNNClassifier(torch.nn.Module):
    def __init__(self, input_size, hidden_size, output_size, n_layers=1, bidirectional=True):
        super(RNNClassifier, self).__init__()
        #hidden size n——layers是GRU的参数
        self.hidden_size = hidden_size
        self.n_layers = n_layers
        self.n_directions = 2 if bidirectional else 1

        #The input of Embedding layer with shape （, ℎ）
        #The ouput of Embedding layer with shape ( , ℎ, ℎ)
        self.embedding = torch.nn.Embedding(input_size, hidden_size)

        #GRU层 bidirectional是用来说明是单向还是双向
        self.gru = torch.nn.GRU(hidden_size, hidden_size, n_layers,
                            bidirectional=bidirectional)
        #bidirectional之后要乘self.n_directions
        self.fc = torch.nn.Linear(hidden_size * self.n_directions, output_size)


    def _init_hidden(self, batch_size):
        hidden = torch.zeros(self.n_layers * self.n_directions,
        batch_size, self.hidden_size)
        return create_tensor(hidden)

    def forward(self, input, seq_lengths):

        # input shape : B x S -> S x B transpose转置
        input = input.t()
        batch_size = input.size(1)

        hidden = self._init_hidden(batch_size)
        embedding = self.embedding(input)
        # pack them up GRU中的提速功能
        gru_input = pack_padded_sequence(embedding, seq_lengths)
        output, hidden = self.gru(gru_input, hidden)
        if self.n_directions == 2:
            hidden_cat = torch.cat([hidden[-1], hidden[-2]], dim=1)
        else:
            hidden_cat = hidden[-1]
        fc_output = self.fc(hidden_cat)
        return fc_output

3convert name to tensor

• 把字符转变成一个字符，字符变成ASCII码，然后填充，转置，排序
• 

def name2list(name):
    arr=[ord(c) for c in name]
    return arr,len(arr)

def make_tensors(names, countries):
    sequences_and_lengths = [name2list(name) for name in names]
    name_sequences = [sl[0] for sl in sequences_and_lengths]
    seq_lengths = torch.LongTensor([sl[1] for sl in sequences_and_lengths])
    countries = countries.long()

    # make tensor of name, BatchSize x SeqLen
    # 补零：先将所有的0yongtensor初始化，然后在每行前面填充名字
    seq_tensor = torch.zeros(len(name_sequences), seq_lengths.max()).long()
    # print("seq_lengths.max:", seq_lengths.max())
    for idx, (seq, seq_len) in enumerate(zip(name_sequences, seq_lengths), 0):
        seq_tensor[idx, :seq_len] = torch.LongTensor(seq)

    # sort by length to use pack_padded_sequence
    # 将名字长度降序排列，并且返回降序之后的长度在原tensor中的小标perm_idx
    seq_lengths, perm_idx = seq_lengths.sort(dim=0, descending=True)
    # 这个Tensor中的类似于列表中切片的方法神奇啊，直接返回下标对应的元素，相等于排序了
    seq_tensor = seq_tensor[perm_idx]
    countries = countries[perm_idx]

    # 返回排序之后名字Tensor，排序之后的名字长度Tensor，排序之后的国家名字Tensor
    return create_tensor(seq_tensor), \
        create_tensor(seq_lengths),\
        create_tensor(countries)

4 训练数据

classifier = RNNClassifier(N_CHARS, HIDDEN_SIZE, N_COUNTRY, N_LAYER) 

criterion = torch.nn.CrossEntropyLoss() 
optimizer = torch.optim.Adam(classifier.parameters(), lr=0.001)

import time
import math

def trainModel():
    def time_since(since):
        s = time.time() - since
        m = math.floor(s / 60)
        s -= m * 60
        return '%dm %ds' % (m, s)

    total_loss = 0 
    for i, (names, countries) in enumerate(trainLoader, 1):
        # print(type(names), type(countries))
        # print(len(names), countries.shape)
        inputs, seq_lengths, target = make_tensors(names, countries)
        
        output = classifier(inputs, seq_lengths) 
        # print("Shape:", output.shape, target.shape)
        # 注意输出和目标的维度：Shape: torch.Size([256, 18]) torch.Size([256])
        loss = criterion(output, target) 
        optimizer.zero_grad() 
        loss.backward() 
        optimizer.step()

        total_loss += loss.item() 
        if i % 10 == 0:
            print(f'[{time_since(start)}] Epoch {epoch} ', end='')
            print(f'[{i * len(inputs)}/{len(trainSet)}] ', end='')
            print(f'loss={total_loss / (i * len(inputs))}') 
    return total_loss

def testModel():
    correct = 0
    total = len(testSet)
    print("evaluating trained model ... ")
    with torch.no_grad():
        for i, (names, countries) in enumerate(testLoader):
            inputs, seq_lengths, target = make_tensors(names, countries)
            output = classifier(inputs, seq_lengths)
            # 注意这个keepdim的使用，为了直接和target计算loss
            pred = output.max(dim=1, keepdim=True)[1]
            # 注意这个view_as 和 eq
            correct += pred.eq(target.view_as(pred)).sum().item()
        
        percent = '%.2f' % (100 * correct / total) 
        print(f'Test set: Accuracy {correct}/{total} {percent}%')

    return correct / total
  
N_EPOCHS = 50
start = time.time() 
print("Training for %d epochs..." % N_EPOCHS) 
acc_list = [] 
for epoch in range(1, N_EPOCHS + 1):
    # Train cycle
    trainModel() 
    acc = testModel() 
    acc_list.append(acc)

训练结果：

Training for 50 epochs...
[0m 2s] Epoch 1 [2560/13374] loss=0.00895661092363298
[0m 4s] Epoch 1 [5120/13374] loss=0.007739758561365306
[0m 6s] Epoch 1 [7680/13374] loss=0.0069853457777450485
[0m 8s] Epoch 1 [10240/13374] loss=0.006530495395418256
[0m 11s] Epoch 1 [12800/13374] loss=0.006135637713596225
evaluating trained model ... 
Test set: Accuracy 4535/6700 67.69%
[0m 15s] Epoch 2 [2560/13374] loss=0.004228085093200207
[0m 17s] Epoch 2 [5120/13374] loss=0.0041014277492649855
[0m 19s] Epoch 2 [7680/13374] loss=0.004011582878107826
[0m 22s] Epoch 2 [10240/13374] loss=0.0038964587613008915
[0m 24s] Epoch 2 [12800/13374] loss=0.0038181920163333416
evaluating trained model ... 
......（中间省略若干）
[12m 46s] Epoch 49 [2560/13374] loss=0.00016357196727767587
[12m 49s] Epoch 49 [5120/13374] loss=0.0001682748734310735
[12m 51s] Epoch 49 [7680/13374] loss=0.00017566338913942067
[12m 54s] Epoch 49 [10240/13374] loss=0.0001776946208337904
[12m 57s] Epoch 49 [12800/13374] loss=0.00018831568930181676
evaluating trained model ... 
Test set: Accuracy 5627/6700 83.99%
[13m 2s] Epoch 50 [2560/13374] loss=0.00016892087151063607
[13m 5s] Epoch 50 [5120/13374] loss=0.00015529338124906645
[13m 7s] Epoch 50 [7680/13374] loss=0.00017500294488854707
[13m 11s] Epoch 50 [10240/13374] loss=0.00017692927776806754
[13m 14s] Epoch 50 [12800/13374] loss=0.00018558732335804962
evaluating trained model ... 
Test set: Accuracy 5613/6700 83.78%
 

 5可视化

import matplotlib.pyplot as plt
import numpy as np

epoch = np.arange(1, len(acc_list) + 1)
acc_list = np.array(acc_list)
plt.plot(epoch, acc_list)
plt.xlabel('Epoch')
plt.ylabel('Accuracy')
plt.grid()
plt.show()

 注意：本节没看完，实在看不懂了，所以后半程的代码参考另一个csdn大佬博客【PyTorch】深度学习实践 b站 第13讲RNN高级篇_像我这样的人-CSDN博客  若有侵犯联系删除

 作业：

链接：Sentiment Analysis on Movie Reviews | Kaggle