Hung-Yi Lee homework[4]:RNN

（一）作业描述

  输入英文句子，输出0或1（如果句子是正面的，标1；如果句子是负面的，标0）。要求采用RNN。
  下载的数据中包含三个文件：training_label.txt，training_nolabel.txt，testing_data.txt。
  training_label.txt的样式如下，最开始的1/0是正负倾向；+++&+++是分隔符，不用在意；分隔符后是需要判断倾向的句子。

  training_molabel.txt的样式如下：

  training_molabel.txt中的数据在训练过程中不进行使用，但在进行词的向量化时进行使用。
  testing_data.txt的样式如下：

  testing_data.txt分为两列，第一列是句子的id，第二列是句子，两列用’，'进行分隔。

（二）实现过程

  代码结构：

utils.py

  定义各种基本函数

import warnings
warnings.filterwarnings('ignore')
import torch
import numpy as np
import pandas as pd
import torch.optim as optim
import torch.nn.functional as Fdef load_training_data(path='training_label.txt'):# 把 training 时需要的 data 读进来# 如果是 'training_label.txt'，需要读取 label，如果是 'training_nolabel.txt'，不需要读取 labelif 'training_label' in path:with open(path, 'rb') as f:lines = f.readlines()lines = [line.decode().strip('\n').split(' ') for line in lines]x = [line[2:] for line in lines]y = [line[0] for line in lines]return x, yelse:with open(path, 'rb') as f:lines = f.readlines()x = [line.decode().strip('\n').split(' ') for line in lines]return xdef load_testing_data(path='testing_data'):# 把 testing 时需要的 data 读进来with open(path, 'rb') as f:lines = f.readlines()X = ["".join(line.decode().strip('\n').split(",")[1:]).strip() for line in lines[1:]]X = [sen.split(' ') for sen in X]return Xdef evaluation(outputs, labels):# outputs => probability (float)# labels => labelsoutputs[outputs >= 0.5] = 1  # 大于等于 0.5 为正面outputs[outputs < 0.5] = 0  # 小于 0.5 为负面correct = torch.sum(torch.eq(outputs, labels)).item()return correct

word2vec.py

  词的向量化

import os
import numpy as np
import pandas as pd
import argparse
from gensim.models import word2vec
from utils import *def train_word2vec(x):model = word2vec.Word2Vec(x, size=250, window=5, min_count=5, workers=12, iter=10, sg=1)return modelif __name__ == "__main__":print("loading training data ...")train_x, y = load_training_data('training_label.txt')train_x_no_label = load_training_data('training_nolabel.txt')print("loading testing data ...")test_x = load_testing_data('testing_data.txt')model = train_word2vec(train_x + test_x)print("saving model ...")model.save(os.path.join('w2v_all.model'))

data_preprocess.py

  数据预处理

from torch import nn
from gensim.models import Word2Vec
import torchclass Preprocess():def __init__(self, sentences, sen_len, w2v_path="./w2v.model"):self.w2v_path = w2v_pathself.sentences = sentencesself.sen_len = sen_lenself.idx2word = []self.word2idx = {}self.embedding_matrix = []def get_w2v_model(self):# 把之前训练好的 word to vec 模型读进来self.embedding = Word2Vec.load(self.w2v_path)self.embedding_dim = self.embedding.vector_sizedef add_embedding(self, word):# 把 word 加进 embedding, 并赋予它一个随机生成的 representation vectorvector = torch.empty(1, self.embedding_dim)torch.nn.init.uniform_(vector)self.word2idx[word] = len(self.word2idx)self.idx2word.append(word)self.embedding_matrix = torch.cat([self.embedding_matrix, vector], 0)def make_embedding(self, load=True):print("Get embedding ...")# 取得训练好的 Word2vec word embeddingif load:print("loading word to vec model ...")self.get_w2v_model()else:raise NotImplementedError# 制作一个 word2idx 的 dictionary# 制作一个 idx2word 的 list# 制作一个 word2vector 的 listfor i, word in enumerate(self.embedding.wv.vocab):print('get words #{}'.format(i+1), end='\r')self.word2idx[word] = len(self.word2idx)self.idx2word.append(word)self.embedding_matrix.append(self.embedding[word])print('')self.embedding_matrix = torch.tensor(self.embedding_matrix)# 将 "<PAD>" 跟 "<UNK>" 加进 embedding 里面self.add_embedding("<PAD>")self.add_embedding("<UNK>")print("total words: {}".format(len(self.embedding_matrix)))return self.embedding_matrixdef pad_sequence(self, sentence):# 将每个句子变成一样的长度if len(sentence) > self.sen_len:sentence = sentence[:self.sen_len]else:pad_len = self.sen_len - len(sentence)for _ in range(pad_len):sentence.append(self.word2idx["<PAD>"])assert len(sentence) == self.sen_lenreturn sentencedef sentence_word2idx(self):# 把句子里面的字转成相对应的 indexsentence_list = []for i, sen in enumerate(self.sentences):print('sentence count #{}'.format(i+1), end='\r')sentence_idx = []for word in sen:if (word in self.word2idx.keys()):sentence_idx.append(self.word2idx[word])else:sentence_idx.append(self.word2idx["<UNK>"])# 将每个句子变成一样的长度sentence_idx = self.pad_sequence(sentence_idx)sentence_list.append(sentence_idx)return torch.LongTensor(sentence_list)def labels_to_tensor(self, y):# 把 labels 转成 tensory = [int(label) for label in y]return torch.LongTensor(y)

dataset.py

  定义数据集结构。

import torch
from torch.utils import dataclass TwitterDataset(data.Dataset):"""Expected data shape like:(data_num, data_len)Data can be a list of numpy array or a list of listsinput data shape : (data_num, seq_len, feature_dim)__len__ will return the number of data"""def __init__(self, X, y):self.data = Xself.label = ydef __getitem__(self, idx):if self.label is None: return self.data[idx]return self.data[idx], self.label[idx]def __len__(self):return len(self.data)

model.py

  定义模型。

import torch
from torch import nn
class LSTM_Net(nn.Module):def __init__(self, embedding, embedding_dim, hidden_dim, num_layers, dropout=0.5, fix_embedding=True):super(LSTM_Net, self).__init__()# 制作 embedding layerself.embedding = torch.nn.Embedding(embedding.size(0),embedding.size(1))self.embedding.weight = torch.nn.Parameter(embedding)# 是否将 embedding fix 住，如果 fix_embedding 为 False，在训练过程中，embedding 也会跟着被训练self.embedding.weight.requires_grad = False if fix_embedding else Trueself.embedding_dim = embedding.size(1)self.hidden_dim = hidden_dimself.num_layers = num_layersself.dropout = dropoutself.lstm = nn.LSTM(embedding_dim, hidden_dim, num_layers=num_layers, batch_first=True)self.classifier = nn.Sequential( nn.Dropout(dropout),nn.Linear(hidden_dim, 1),nn.Sigmoid() )def forward(self, inputs):inputs = self.embedding(inputs)x, _ = self.lstm(inputs, None)# x 的 dimension (batch, seq_len, hidden_size)# 取用 LSTM 最后一层的 hidden statex = x[:, -1, :]x = self.classifier(x)return x

train.py

  定义训练过程如何进行参数更新及模型保存。

import torch
from torch import nn
import torch.optim as optim
import torch.nn.functional as F
from utils import *def training(batch_size, n_epoch, lr, model_dir, train, valid, model, device):total = sum(p.numel() for p in model.parameters())trainable = sum(p.numel() for p in model.parameters() if p.requires_grad)print('\nstart training, parameter total:{}, trainable:{}\n'.format(total, trainable))model.train()  # 将 model 的模式设为 train，这样 optimizer 就可以更新 model 的参数criterion = nn.BCELoss()  # 定义损失函数，binary cross entropy losst_batch = len(train)v_batch = len(valid)optimizer = optim.Adam(model.parameters(), lr=lr)  # 将模型的参数給 optimizer，并给予适当的 learning ratetotal_loss, total_acc, best_acc = 0, 0, 0for epoch in range(n_epoch):total_loss, total_acc = 0, 0# 开始 trainingfor i, (inputs, labels) in enumerate(train):inputs = inputs.to(device, dtype=torch.long)  # device 为 "cuda"，将 inputs 转成 torch.cuda.LongTensorlabels = labels.to(device, dtype=torch.float)  # device为 "cuda"，将 labels 转成 torch.cuda.FloatTensor，因为等下要喂进 criterion，所以类型要是 floatoptimizer.zero_grad()  # 由于 loss.backward() 的 gradient 会累加，所以每次喂完一個 batch 后需要归零outputs = model(inputs)  # 将 input 喂給模型outputs = outputs.squeeze()  # 去掉最外面的 dimension，好让 outputs 可以喂进 criterion()loss = criterion(outputs, labels)  # 计算此时模型的 training lossloss.backward()  # 算 loss 的 gradientoptimizer.step()  # 更新correct = evaluation(outputs, labels)  # 计算此时模型的 training accuracytotal_acc += (correct / batch_size)total_loss += loss.item()print('[ Epoch{}: {}/{} ] loss:{:.3f} acc:{:.3f} '.format(epoch+1, i+1, t_batch, loss.item(), correct*100/batch_size), end='\r')print('\nTrain | Loss:{:.5f} Acc: {:.3f}'.format(total_loss/t_batch, total_acc/t_batch*100))# 这段做 validationmodel.eval()  # 将 model 的模式设为 eval，这样 model 的参数就会被固定with torch.no_grad():total_loss, total_acc = 0, 0for i, (inputs, labels) in enumerate(valid):inputs = inputs.to(device, dtype=torch.long)  # device 为 "cuda"，将 inputs 转成 torch.cuda.LongTensorlabels = labels.to(device, dtype=torch.float)  # device 为 "cuda"，将 labels 转成 torch.cuda.FloatTensor，因为等下要喂进 criterion，所以类型要是 floatoutputs = model(inputs)  # 将 input 喂給模型outputs = outputs.squeeze()  # 去掉最外面的 dimension，好让 outputs 可以喂进 criterion()loss = criterion(outputs, labels)  # 计算此时模型的 validation losscorrect = evaluation(outputs, labels)  # 计算此时模型的 validation accuracytotal_acc += (correct / batch_size)total_loss += loss.item()print("Valid | Loss:{:.5f} Acc: {:.3f} ".format(total_loss/v_batch, total_acc/v_batch*100))if total_acc > best_acc:# 如果 validation 的结果大于之前所有的結果，就把现在的模型存下來以备之后做预测时使用best_acc = total_acctorch.save(model, "{}/ckpt.model".format(model_dir))print('saving model with acc {:.3f}'.format(total_acc/v_batch*100))print('-----------------------------------------------')model.train()

train_main.py

  正式开始训练，模型训练的完整流程，包括数据预处理等。

import os
import torch
import argparse
import numpy as np
from torch import nn
from gensim.models import word2vec
from sklearn.model_selection import train_test_split
from utils import *
from data_preprocess import *
from model import *
from dataset import *
from train import *if __name__ == "__main__":# 判断是否可以使用GPU, 如果可以的话 device 就设为 "cuda"，不行的话设为 "cpu"device = torch.device("cuda" if torch.cuda.is_available() else "cpu")print(device)# 处理好各个 data 的路径train_with_label = 'training_label.txt'train_no_label = 'training_nolabel.txt'testing_data = 'testing_data.txt'w2v_path = 'w2v_all.model'# 定义句子长度、要不要固定 embedding、batch 大小、要训练几个 epoch、learning rate 的值、model 的路径sen_len = 20fix_embedding = Truebatch_size = 32epoch = 5lr = 0.001# model_dir = os.path.join(path_prefix, 'model/') # model directory for checkpoint modelprint("loading data ...") # 把 'training_label.txt' 跟 'training_nolabel.txt' 讀進來train_x, y = load_training_data(train_with_label)train_x_no_label = load_training_data(train_no_label)# 对 input 跟 labels 做预处理preprocess = Preprocess(train_x, sen_len, w2v_path=w2v_path)embedding = preprocess.make_embedding(load=True)train_x = preprocess.sentence_word2idx()y = preprocess.labels_to_tensor(y)# 制作一个 model 的对象model = LSTM_Net(embedding, embedding_dim=250, hidden_dim=150, num_layers=1, dropout=0.5, fix_embedding=fix_embedding)model = model.to(device)# 把 data 分为 training data 跟 validation dataX_train, X_val, y_train, y_val = train_x[:180000], train_x[180000:], y[:180000], y[180000:]# 制作成dataset格式train_dataset = TwitterDataset(X=X_train, y=y_train)val_dataset = TwitterDataset(X=X_val, y=y_val)# 把 data 转成 batch of tensorstrain_loader = torch.utils.data.DataLoader(dataset=train_dataset,batch_size=batch_size,shuffle=True,num_workers=2)val_loader = torch.utils.data.DataLoader(dataset=val_dataset,batch_size=batch_size,shuffle=False,num_workers=2)# 开始训练training(batch_size, epoch, lr, 'model_dir', train_loader, val_loader, model, device)

test.py

  定义测试过程。

import torch
from torch import nn
import torch.optim as optim
import torch.nn.functional as Fdef testing(batch_size, test_loader, model, device):model.eval()ret_output = []with torch.no_grad():for i, inputs in enumerate(test_loader):inputs = inputs.to(device, dtype=torch.long)outputs = model(inputs)outputs = outputs.squeeze()outputs[outputs >= 0.5] = 1  # 大于等于 0.5 为正outputs[outputs < 0.5] = 0  # 小于 0.5 为负ret_output += outputs.int().tolist()return ret_output

predict.py

  对测试集进行预测并将结果保存到predict.csv中。

import os
import pandas as pd
from data_preprocess import *
from dataset import *
from utils import *
from test import *if __name__ == "__main__":device = torch.device("cuda" if torch.cuda.is_available() else "cpu")testing_data = 'testing_data.txt'w2v_path = 'w2v_all.model'sen_len = 20batch_size = 32model_dir = 'model_dir'print("loading testing data ...")test_x = load_testing_data(testing_data)preprocess = Preprocess(test_x, sen_len, w2v_path=w2v_path)embedding = preprocess.make_embedding(load=True)test_x = preprocess.sentence_word2idx()test_dataset = TwitterDataset(X=test_x, y=None)test_loader = torch.utils.data.DataLoader(dataset=test_dataset,batch_size=batch_size,shuffle=False,num_workers=3)print('\nload model ...')model = torch.load(os.path.join(model_dir, 'ckpt.model'))outputs = testing(batch_size, test_loader, model, device)tmp = pd.DataFrame({"id": [str(i) for i in range(len(test_x))], "label":outputs})print("save csv ...")tmp.to_csv('predict.csv', index=False)print("Finish Predicting")

  predict.csv的样式如下：

  predict.csv分为两列，第一列是用来判断倾向的句子id，第二列是预测的句子正负。