逻辑回归-Kaggle-GiveMeSomeCredit问题-PyTorch实现

下文仅给出模型的PyTorch实现，具体问题分析过程以及数据下载可以去作者的GitHub

逻辑回归介绍：可以看这里 主要是区分它和线性回归的差异

导入相关库：

import numpy
import pandas as pd
import matplotlib.pyplot as plt
import torch
import torch.nn as nn

载入数据：

df = pd.read_csv('./data/GiveMeSomeCredit/cs-training.csv')
sampleEntry = pd.read_csv('./data/GiveMeSomeCredit/sampleEntry.csv')

训练数据是一个包含用户各种信息的Excel文件，这里仅列出几行

    Unnamed: 0    SeriousDlqin2yrs    RevolvingUtilizationOfUnsecuredLines    age    NumberOfTime30-59DaysPastDueNotWorse    DebtRatio    MonthlyIncome    NumberOfOpenCreditLinesAndLoans    NumberOfTimes90DaysLate    NumberRealEstateLoansOrLines    NumberOfTime60-89DaysPastDueNotWorse    NumberOfDependents
0    1    1    0.766127    45    2    0.802982    9120.0    13    0    6    0    2.0
1    2    0    0.957151    40    0    0.121876    2600.0    4    0    0    0    1.0
2    3    0    0.658180    38    1    0.085113    3042.0    2    1    0    0    0.0
3    4    0    0.233810    30    0    0.036050    3300.0    5    0    0    0    0.0
4    5    0    0.907239    49    1    0.024926    63588.0    7    0    1    0    0.0

数据初步处理：

# 将缺失值补全
df.rename(columns={'Unnamed: 0':'Id', 'SeriousDlqin2yrs':'Default'}, inplace=True)
df['MonthlyIncome'].fillna(df['MonthlyIncome'].mean(),inplace=True)
df['NumberOfDependents'].fillna(test['NumberOfDependents'].mode()[0], inplace=True)

数据的归一化，并将读取的所有数据按照7：3分为训练数据和测试数据：

# 获取训练集和测试集
features=['RevolvingUtilizationOfUnsecuredLines','age','NumberOfTime30-59DaysPastDueNotWorse','DebtRatio','MonthlyIncome','DebtRatio','MonthlyIncome','NumberOfOpenCreditLinesAndLoans','NumberRealEstateLoansOrLines','NumberOfDependents']
dep=['Default']
allfeatures = ['RevolvingUtilizationOfUnsecuredLines','age','NumberOfTime30-59DaysPastDueNotWorse','DebtRatio','MonthlyIncome','DebtRatio','MonthlyIncome','NumberOfOpenCreditLinesAndLoans','NumberRealEstateLoansOrLines','NumberOfDependents','Default']
alldata = df[allfeatures]

# 数据归一化
means, stds = dict(), dict()

for col in features:
    means[col] = alldata[col].mean()
    stds[col] = alldata[col].std()

for col in features:
    alldata[col] = (alldata[col] - means[col]) / stds[col]

x=alldata[features]
y=alldata[dep]

train_size = int(x.shape[0] * 0.7)
test_size = x.shape[0] - train_size

x_train = x[:train_size]
y_train = y[:train_size]
x_test = x[train_size - 1:-1]
y_test = y[train_size - 1:-1]

# 将数据转化为Tensor
x_train = torch.tensor(x_train.values, dtype=torch.float)
y_train = torch.tensor(y_train.values, dtype=torch.float)
x_test = torch.tensor(x_test.values, dtype=torch.float)
y_test = torch.tensor(y_test.values, dtype=torch.float)

具体模型，这里的输入维度14和输出维度2可以从上面提供的x,y数据看出：

class MulNet(nn.Module):
    def __init__(self):# 输入维度14 输出维度2
        super(MulNet, self).__init__()
        self.l1 = nn.Linear(14, 140)
        self.l2 = nn.Linear(140, 140)
        self.l3 = nn.Linear(140, 2)
        self.sigmod = nn.Sigmoid()
        
    def forward(self, x):
        y = self.l1(x)
        y = self.l2(y)
        y = self.l3(y)
        return self.sigmod(y)

超参数设定：

model3 = MulNet()
criterion = nn.MSELoss()
optimizer = torch.optim.SGD(model3.parameters(), lr=0.01)

训练模型：

for t in range(500):
    y_pred = model3(x_train)
    
    loss = criterion(y_pred, y_train)
    if (t + 1) % 10 == 0:
        print(t, loss.item())
    losses1.append(loss.item())
    
    if torch.isnan(loss):
        break
    
    optimizer.zero_grad()
    loss.backward()
    optimizer.step()

9 0.26426613330841064
19 0.24677835404872894
29 0.23098362982273102
39 0.21666239202022552
49 0.20363378524780273
59 0.19174979627132416
69 0.18088917434215546
79 0.17095163464546204
89 0.16185317933559418
99 0.15352196991443634
109 0.14589543640613556
119 0.1389177143573761
129 0.13253812491893768
139 0.12670989334583282
149 0.12138950824737549
159 0.11653617769479752
169 0.11211156845092773
179 0.1080796867609024
189 0.10440683364868164
199 0.10106151551008224
209 0.0980144515633583
219 0.09523852169513702
229 0.0927087813615799
239 0.09040225297212601
249 0.08829796314239502
259 0.08637678623199463
269 0.08462131768465042
279 0.0830157995223999
289 0.08154597878456116
299 0.08019895851612091
309 0.07896309345960617
319 0.07782793790102005
329 0.07678405195474625
339 0.07582291960716248
349 0.07493693381547928
359 0.0741191878914833
369 0.07336351275444031
379 0.07266434282064438
389 0.072016641497612
399 0.07141589373350143
409 0.07085802406072617
419 0.07033936679363251
429 0.06985656172037125
439 0.06940664350986481
449 0.06898687034845352
459 0.0685947984457016
469 0.06822817772626877
479 0.06788500398397446
489 0.06756342947483063
499 0.06726177781820297

训练结果：

preds = model3(x_test)
print(preds.shape)
_, ys = torch.max(preds, 1)
print(ys.shape)
right = 0.0
for i in range(len(ys)):
    if ys[i] == y_test[i]:
        right += 1
acc = right / len(ys)
print(acc)
# 之前试过学习率设为0.1时 正确率只有0.45

torch.Size([45000, 2])
torch.Size([45000])
0.8359111111111112