【神经网络-数据分类】利用深度学习神经网络（DNN）来对数据进行分类

对于数据量大，适合以一维数据作为输入且难以探索数据内部规律的数据集，可以尝试利用深度学习神经网络（DNN）来对数据进行分类。

由于是分类问题，使用Adam作为优化器，交叉熵作为损失函数，大概的网络结构代码如下：

import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.utils.data.dataset import Dataset
from torch.utils.data import DataLoader

class CipvDNN(nn.Module):
    def __init__(self, input_dim,output_dim):
        super(CipvDNN, self).__init__()
        self.fc1 = nn.Linear(input_dim, 64)
        self.fc2 = nn.Linear(64, 32)
        self.fc3 = nn.Linear(32, output_dim)

    def forward(self, x):
        x = F.relu(self.fc1(x))
        x = F.relu(self.fc2(x))
        x = self.fc3(x)
        #x = F.dropout(x, p=0.1)
        return x

    def predict(self, features):
        self.eval()
        features = torch.from_numpy(features).float()
        return self.forward(features).detach().numpy()

class CipvDataset(Dataset):
    def __init__(self, labels, features):
        super(CipvDataset, self).__init__()
        self.labels = labels
        self.features = features

    def __len__(self):
        return self.features.shape[0]

    def __getitem__(self, idx):
        feature = self.features[idx]
        label = self.labels[idx]
        return {'feature': feature, 'label': label}

class CipvTrain_DNN():
    def __init__(self):

        self.network = CipvDNN(100,20)
        self.learning_rate = 0.0001
        self.optimizer = torch.optim.Adam(self.network.parameters(), lr=self.learning_rate)
        self.criterion = nn.CrossEntropyLoss()
        self.num_epochs = 1000
        self.batchsize = 500
        #self.shuffle = True

    def train(self, features, labels):

        self.network.train()
        dataset = CipvDataset(labels, features)
        loader = DataLoader(dataset, batch_size=self.batchsize)
        #loader = DataLoader(dataset, shuffle=self.shuffle, batch_size = self.batchsize)

        for epoch in range(self.num_epochs):
            total_loss = 0.0
            for i, data in enumerate(loader):
                features = data['feature'].float()
                labels = data['label'].long()
                self.optimizer.zero_grad()
                predictions = self.network(features)
                loss = self.criterion(predictions, labels.squeeze(-1))
                loss.backward()
                total_loss += loss.item()
                self.optimizer.step()
            print('loss', total_loss / (i+1))

    def get_action(self, feature):
        self.network.eval()
        p = self.network.predict(feature)
        label = np.argmax(p)
        return label

在训练好之后，如何知道自己模型的预测效果呢？这里提供一个简单的预测函数。

def test_model(policy,features,labels):
    success_time=0
    for i in range(len(features)):
        label_got=policy.get_action(features[i])
        if label_got==labels[i]:
            success_time +=1
    success_rate = success_time/len(features)
    print('success rate',success_rate)
    return 0