Kaggle新手题： 基于神经网络的Titanic disaster存活预测

第一次尝试搭建神经网络，一共8层，最终提交准确率在70%左右，电脑硬件较好可以增加迭代次数继续降低损失函数。

import torch
import numpy
import matplotlib.pyplot as plt

#从文件读取数据
xy=numpy.loadtxt('D:Titanic\train_processed.csv',delimiter=',',dtype=numpy.float32)
xz=numpy.loadtxt('D:Titanic\test_processed.csv',delimiter=',',dtype=numpy.float32)
x_data=torch.from_numpy(xy[:,:-1])
y_data=torch.from_numpy(xy[:,[-1]])
test_data=torch.from_numpy(xz[:,:])
#检查维度
print(test_data.shape)
print(x_data.shape,y_data.shape)

loss_list=[]
#创建模型
class LogisticRegressionModel(torch.nn.Module):
    def __init__(self):
        super().__init__()
        self.linear1=torch.nn.Linear(7,15)
        self.linear2=torch.nn.Linear(15,30)
        self.linear3=torch.nn.Linear(30,100)
        self.linear4=torch.nn.Linear(100,70)
        self.linear5=torch.nn.Linear(70,20)
        self.linear6=torch.nn.Linear(20,10)
        self.linear7=torch.nn.Linear(10,1)
        self.relu=torch.nn.ReLU()
        self.sigmoid=torch.nn.Sigmoid()
    def forward(self,x):
        x=self.relu(self.linear1(x))
        x=self.relu(self.linear2(x))
        x=self.relu(self.linear3(x))
        x=self.relu(self.linear4(x))
        x=self.relu(self.linear5(x))
        x=self.relu(self.linear6(x))
        x=self.sigmoid(self.linear7(x))
        return x
#损失函数和优化器
model=LogisticRegressionModel()
criterion=torch.nn.BCELoss(reduction='sum')
optimizer=torch.optim.Adam(model.parameters(),lr=0.01)
scheduler = torch.optim.lr_scheduler.ExponentialLR(optimizer, gamma=0.999) #学习率衰减
#使用GPU运算
#device=torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
#x_data,y_data=x_data.to(device),y_data.to(device)

for epoch in range(5000):
    y_pre=model(x_data)
    loss=criterion(y_pre,y_data)
    loss_list.append(loss.item())
    optimizer.zero_grad()
    loss.backward()
    optimizer.step()
    scheduler.step() #学习率衰减
epoch_list=numpy.arange(5000)
y_pred=model(test_data)
y_pred_label=torch.where(y_pred>=0.5,torch.tensor([1]),torch.tensor([0]))
#结果文件输出
y_end=y_pred_label.numpy()
numpy.savetxt('D:Titanic\predict.csv',y_end,delimiter=',')
#可视化训练过程
plt.plot(epoch_list,loss_list,label='loss',color='blue')
plt.show()

损失函数的下降情况：

在Kaggle平台的提交结果：

在使用mini-batch进行小批量梯度下降和学习率指数衰减后可以进一步优化，准确率在74%左右，但由于电脑性能原因只迭代了1500次，尚有进步空间，代码改进如下：

import torch
import numpy
import matplotlib.pyplot as plt
from torch.utils.data import Dataset
from torch.utils.data import DataLoader

loss_list=[]#记录损失函数
#构建数据集类
class TitanicDataset(Dataset):
   def __init__(self):
       super(TitanicDataset, self).__init__()
       xy = numpy.loadtxt('D:Titanic\train_processed.csv', delimiter=',', dtype=numpy.float32)
       xz = numpy.loadtxt('D:Titanic\test_processed.csv', delimiter=',', dtype=numpy.float32)
       self.len=xy.shape[0]
       self.x_data=torch.from_numpy(xy[:,:-1])
       self.y_data=torch.from_numpy(xy[:,[-1]])
       self.test_data=torch.from_numpy(xz[:,:])

   def __getitem__(self, item):
       return self.x_data[item],self.y_data[item]

   def __len__(self):
       return self.len

#创建模型
class LogisticRegressionModel(torch.nn.Module):
    def __init__(self):
        super().__init__()
        self.linear1=torch.nn.Linear(7,15)
        self.linear2=torch.nn.Linear(15,30)
        self.linear3=torch.nn.Linear(30,100)
        self.linear4=torch.nn.Linear(100,70)
        self.linear5=torch.nn.Linear(70,20)
        self.linear6=torch.nn.Linear(20,10)
        self.linear7=torch.nn.Linear(10,1)
        self.relu=torch.nn.ReLU()
        self.sigmoid=torch.nn.Sigmoid()
    def forward(self,x):
        x=self.relu(self.linear1(x))
        x=self.relu(self.linear2(x))
        x=self.relu(self.linear3(x))
        x=self.relu(self.linear4(x))
        x=self.relu(self.linear5(x))
        x=self.relu(self.linear6(x))
        x=self.sigmoid(self.linear7(x))
        return x

data_set=TitanicDataset()
train_loader=DataLoader(dataset=data_set,batch_size=30,shuffle=True)
#损失函数和优化器
model=LogisticRegressionModel()
criterion=torch.nn.BCELoss(reduction='sum')
optimizer=torch.optim.Adam(model.parameters(),lr=0.005)
scheduler = torch.optim.lr_scheduler.ExponentialLR(optimizer, gamma=0.9999) #学习率衰减
#使用GPU运算
#device=torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
#x_data,y_data=x_data.to(device),y_data.to(device)

def train():
    for epoch in range(1500):
        for i,data in enumerate(train_loader,0):
            input,label=data
            y_pre=model(input)
            loss=criterion(y_pre,label)
            if(i==20):
                loss_list.append(loss.item())
            if(loss<0.2):
                return
            optimizer.zero_grad()
            loss.backward()
            optimizer.step()
            scheduler.step() #学习率衰减

train()
epoch_list=numpy.arange(len(loss_list))
y_pred=model(data_set.test_data)
y_pred_label=torch.where(y_pred>=0.5,torch.tensor([1]),torch.tensor([0]))
#结果文件输出
y_end=y_pred_label.numpy()
numpy.savetxt('D:Titanic\predict.csv',y_end,delimiter=',')
#可视化训练过程
plt.plot(epoch_list,loss_list,label='loss',color='blue')
plt.show()

损失函数如下，每个batch内周期性起伏，整体呈下降趋势

提交结果：

本文章作为记录，希望进一步学习后可以搭建效果更好的网络。