之前写过几篇文章来进行文本数据的预处理，包括分词，去停用词，构造词向量。
这里调用前文方法获得词向量，利用pytorch编写cnn程序进行情感识别。
import torch
import torch.nn as nn
import torch.utils.data as Data
from data_helper import do_data_helper
import matplotlib.pyplot as plt
torch.manual_seed(1) # 用来固定随机数
# 超参数
EPOCH = 1 # 训练迭代轮数
BATCH_SIZE = 100
LR = 0.001 # 学习率
DROP = 0.3 # 放置过拟合扔掉一些
TRAIN_PENCENT = 0.8 # 训练数据占的比重
# 获得x,y
x, y = do_data_helper()
# 切分出训练数据和测试数据
train_num = int(len(y) * TRAIN_PENCENT)
train_x = torch.tensor(x[:train_num]).unsqueeze(1) # 增加一维channel（cnn中的厚度/层数）
# print(train_x.size()) # 观察x的形状维度
# print(train_x)
train_y = torch.tensor(y[:train_num])
test_x = torch.tensor(x[train_num:]).unsqueeze(1)
test_y = torch.tensor(y[train_num:])
# 封装用来拆分batch的数据
torch_dataset = Data.TensorDataset(train_x, train_y)
train_loader = Data.DataLoader(
dataset=torch_dataset,
batch_size=BATCH_SIZE,
shuffle=False,
num_workers=2
)
# 卷积网络
class CNN(nn.Module):
def __init__(self):
super(CNN, self).__init__()
self.conv1 = nn.Sequential( # input shape (1, 10, 100)
nn.Conv2d(
in_channels=1, # input height，厚度，黑白的，一层厚
out_channels=16, # 过滤器/卷积核的个数
kernel_size=5, # 过滤器/卷积核的规格、尺寸 5*5*channel
stride=1, # 步长
padding=2, # 如果想要 con2d 出来的图片长宽没有变化, padding=(kernel_size-1)/2 当 stride=1
), # output shape (16, 10, 100)
nn.Dropout2d(DROP),
nn.ReLU(), # activation
nn.MaxPool2d(kernel_size=2), # 在 2x2 空间里向下采样, output shape (16, 5, 50)
# nn.MaxPool2d()的stride默认值是kernel_size
)
self.conv2 = nn.Sequential( # input shape (16, 5, 50)
nn.Conv2d(16, 32, 5, 1, 2), # output shape (32, 5, 50)
nn.Dropout2d(DROP),
nn.ReLU(), # activation
nn.MaxPool2d(1), # output shape (32, 5, 50)
)
# 输出层
self.out = nn.Sequential(
nn.Linear(32 * 5 * 50, 6), # fully connected layer, output 10 classes
nn.Dropout2d(DROP),
nn.Softmax(dim=-1) # 在行上进行softmax
)
def forward(self, x):
x = self.conv1(x)
x = self.conv2(x)
x = x.view(x.size(0), -1) # 把conv2的宽7*高7*厚32的数据拉成1维 (batch_size, 32 * 7 * 7)
output = self.out(x)
return output
cnn = CNN()
losslist = [] # 用来记录么一个batch上的损失，最后用来出图观察
optimizer = torch.optim.Adam(cnn.parameters(), lr=LR) # optimize all cnn parameters
loss_func = nn.CrossEntropyLoss() # the target label is not one-hotted
if __name__ == '__main__':
# training and testing
for epoch in range(EPOCH):
for step, (b_x, b_y) in enumerate(train_loader): # 分配 batch data, normalize x when iterate train_loader
output = cnn(b_x) # cnn output
loss = loss_func(output, b_y) # cross entropy loss
optimizer.zero_grad() # clear gradients for this training step
loss.backward() # backpropagation, compute gradients
optimizer.step() # apply gradients
losslist.append(loss)
if step % 10 == 0:
test_output = cnn(test_x)
pred_y = torch.max(test_output, 1)[1]
accurary = sum((test_y == pred_y).numpy()) / len(test_y)
print('Epoch: {} | step: {:0>4d} | train loss: {:.5f} | test accurary: {:.3f}'.format(epoch, step, loss, accurary))
# Epoch: 0 | step: 0000 | train loss: 1.79165 | test accurary: 0.180
# Epoch: 0 | step: 0010 | train loss: 1.77615 | test accurary: 0.237
# Epoch: 0 | step: 0020 | train loss: 1.78648 | test accurary: 0.227
# Epoch: 0 | step: 0030 | train loss: 1.78640 | test accurary: 0.214
# Epoch: 0 | step: 0040 | train loss: 1.71782 | test accurary: 0.237
# Epoch: 0 | step: 0050 | train loss: 1.76866 | test accurary: 0.209
# Epoch: 0 | step: 0060 | train loss: 1.77528 | test accurary: 0.237
# Epoch: 0 | step: 0070 | train loss: 1.78668 | test accurary: 0.218
# 保存模型
torch.save(cnn, r'data/models/classmodel.pkl')
test_output = cnn(test_x)
pred_y = torch.max(test_output, 1)[1]
print(pred_y, 'prediction number')
print(test_y, 'real number')
print('accuracy:', sum((test_y == pred_y).numpy()) / len(test_y))
# tensor([0, 3, 3, ..., 0, 0, 0]) prediction number
# tensor([0, 3, 2, ..., 1, 5, 0]) real number
# accuracy: 0.2125
plt.plot(range(len(losslist)), losslist)
plt.show()
做loss变化图如下图所示
相关文献
文本预处理课参见 分类目录——情感识别
Pytorch框架可参见 分类目录——Pytorch
作图相关可参见 分类目录——Matplotlib
