pandas数据处理之构建联邦学习数据

news2024/11/26 22:31:47

本文以天池比赛《车辆贷款违约预测》的数据为例，通过pandas处理数据，构建联邦学习数据，用于FATE框架联邦学习。

通过pandas处理数据

1. 读取数据

下载car_loan_train.csv数据后，用pandas读取数据。

import pandas as pd

data=pd.read_csv("car_loan_train.csv")

2. 查看所有列名及数据行数

print(data.columns)
print(data.shape[0])

3. 查看每列数据最大值及最大值所在行

print(data.max())
print(data.idxmax())

4. 查看data的统计信息

包括count、mean、std、max、min等

data.describe()

5. 按规则删除特定行

遍历数据的每一列，删除存在>=1e9数据的行。这里先获取值>=1e9所在行的索引，然后根据索引删除行。

for column in data.columns:
    print(data.shape[0])
    data=data.drop(index=data[str(column)][data[str(column)]>=1e9].index)

6. 删除部分无用的特征

data.drop(labels=["customer_id","disbursed_date","loan_default","mobileno_flag","idcard_flag","outstanding_disburse_ratio"],axis=1)

7. 数据归一化

features = ['main_account_loan_no', 'main_account_active_loan_no',
       'main_account_overdue_no', 'main_account_outstanding_loan',
       'main_account_sanction_loan', 'main_account_disbursed_loan',
       'sub_account_loan_no', 'sub_account_active_loan_no',
       'sub_account_overdue_no', 'sub_account_outstanding_loan',
       'sub_account_sanction_loan', 'sub_account_disbursed_loan',
       'disbursed_amount', 'asset_cost', 'branch_id', 'supplier_id',
       'manufacturer_id', 'area_id', 'employee_code_id', 'Driving_flag', 'passport_flag', 'credit_score',
       'main_account_monthly_payment', 'sub_account_monthly_payment',
       'last_six_month_new_loan_no', 'last_six_month_defaulted_no',
       'average_age', 'credit_history', 'enquirie_no', 'loan_to_asset_ratio',
       'total_account_loan_no', 'sub_account_inactive_loan_no',
       'total_inactive_loan_no', 'main_account_inactive_loan_no',
       'total_overdue_no', 'total_outstanding_loan', 'total_sanction_loan',
       'total_disbursed_loan', 'total_monthly_payment', 'main_account_tenure',
       'sub_account_tenure', 'disburse_to_sactioned_ratio',
       'active_to_inactive_act_ratio', 'year_of_birth',
       'Credit_level', 'employment_type', 'age']

data[features] = (data[features]-data[features].mean())/(data[features].std())

8. 创建索引

data['idx']=range(data.shape[0])

9. 打乱数据

data=data.sample(frac=1)

10.划分训练集、测试集

因为用于联邦学习，将训练数据分为两部分。

train=data.iloc[:int(data.shape[0]*0.2)]
loan_eval=data.iloc[int(data.shape[0]*0.8):]
loan_1_train=train.iloc[:int(train.shape[0]*0.5)]
loan_2_train=train.iloc[int(train.shape[0]*0.5):]

print(train.shape)
print(loan_eval.shape)
print(loan_1_train.shape)
print(loan_2_train.shape)

11. 保存数据

loan_1_train.to_csv("loan_1_train.csv",index=False,header=True)
loan_2_train.to_csv("loan_2_train.csv",index=False,header=True)
loan_eval.to_csv("loan_eval.csv",index=False,header=True)