前言

本文给出2020年MathorCup高校数学建模挑战赛——大数据竞赛中的赛道A移动通信基站流量预测的baseline，这个题目的具体描述和数据集请见链接。
整个程序是用python写的，步骤包括文件读取、数据处理、特征构造、模型训练和预测、输出文件保存。读者可以在本文的基础上进行模型的提升。

一、简单分析

本文的训练数据有9G左右的大小，且特征字段是中文的，panda读取的时候需要注意。另外，训练数据中含有重复项，程序中直接删除了重复项只保留一个。关于日期字段，训练数据中的格式有两种例如“2018/3/26”和“018-4-09”因此需要分别处理，baseline中的程序使用的是通用的处理方式。需要注意的是，baseline的程序是将两个流量当作关于时间的函数，没有考虑时间序列特性，这里有很大的改进空间。但是如果考虑时序特性的话，不能直接使用程序中的交叉验证方法，这样会存在数据泄露，并且需要考虑对缺失值进行填充。

二、具体程序

1.引入库

import pandas as pd 
import numpy as np
from tqdm import tqdm
import os
import lightgbm as lgb
import warnings
from sklearn.metrics import mean_squared_error
from sklearn.model_selection import StratifiedKFold, KFold
import gc
import time

2.读入数据

博主将训练数据和测试数据全都重命名了，其实可以不用，这里读取必须要用gbk编码，这样才能读取中文。另外移植代码的时候注意路径匹配。

train = pd.read_csv('D:/mathorup/traindata.csv',encoding="gbk")
test1 = pd.read_csv('D:/mathorup/test1.csv',encoding="gbk")#短期测试集
test2 = pd.read_csv('D:/mathorup/test2.csv',encoding="gbk")#长期测试集
train.info()

为了方便后续处理，我用英文字段重命名了列名：

new_col = ['DATE', 'HOUR','NAME' , 'LABEL1','LABEL2']
train.columns = new_col
test1.columns = new_col
new_col2 = ['DATE','NAME' , 'LABEL1','LABEL2']
test2.columns = new_col2

输出进行观察：

train.head()
test1.head()
test2.head()

3.数据处理

删除训练数据中的重复列，保留一个：

train = train.drop_duplicates(keep='first')

删除含有缺失值的数据

train = train.dropna()

处理小时、日、月：

train['HOUR'] = train['HOUR'].apply(lambda x: int(x.split(':')[0]))
train['DAY'] = train['DATE'].apply(lambda x: int(x.split('/')[-1][-2:]))
train['MON'] = train['DATE'].apply(lambda x: int(x[5]))

测试集1做相同处理：

test1['HOUR'] = test1['HOUR'].apply(lambda x: int(x.split(':')[0]))
test1['DAY'] = test1['DATE'].apply(lambda x: int(x.split('/')[-1]))
test1['MON'] = test1['DATE'].apply(lambda x: int(x[5]))

输出观察：

train.head(20)

结果如下图所示：
在这里插入图片描述

4.模型训练和预测

模型我使用的是LGB模型，参数没有进行过优化，大家可以稍微做一下参数优化。训练和预测我使用了两个随机种子和五折交叉验证：
上行流量预测：

used_feat = ['HOUR','DAY','NAME','MON']
train_x = train[used_feat]
train_y = train['LABEL1']
test_x = test1[used_feat]
print(train_x.shape, test_x.shape)# -----------------------------------------------
scores = []params = {'learning_rate': 0.1, 'boosting_type': 'gbdt', 'objective': 'regression','metric': 'rmse','min_child_samples': 46, 'min_child_weight': 0.01,'feature_fraction': 0.8, 'bagging_fraction': 0.8, 'bagging_freq': 2, 'num_leaves': 16, 'max_depth': 5, 'n_jobs': -1, 'seed': 2019, 'verbosity': -1, }oof_train = np.zeros(len(train_x))
preds = np.zeros(len(test_x))
folds = 5
seeds = [2048, 1997]
for seed in seeds: kfold = KFold(n_splits=folds, shuffle=True, random_state=seed)for fold, (trn_idx, val_idx) in enumerate(kfold.split(train_x, train_y)):print('fold ', fold + 1)x_trn, y_trn, x_val, y_val = train_x.iloc[trn_idx], train_y.iloc[trn_idx], train_x.iloc[val_idx], train_y.iloc[val_idx]train_set = lgb.Dataset(x_trn, y_trn)val_set = lgb.Dataset(x_val, y_val)model = lgb.train(params, train_set, num_boost_round=5000,valid_sets=(train_set, val_set), early_stopping_rounds=25,verbose_eval=50)oof_train[val_idx] += model.predict(x_val) / len(seeds)preds += model.predict(test_x) / folds / len(seeds)del x_trn, y_trn, x_val, y_val, model, train_set, val_setgc.collect()mse = (mean_squared_error(oof_train, train['LABEL1']))print('-'*120)print('rmse ', round(mse, 5))test1['LABEL1'] = preds

下行流量预测：

train_x = train[used_feat]
train_y = train['LABEL2']
test_x = test1[used_feat]
print(train_x.shape, test_x.shape)# -----------------------------------------------
scores = []params = {'learning_rate': 0.1, 'boosting_type': 'gbdt', 'objective': 'regression','metric': 'rmse','min_child_samples': 46, 'min_child_weight': 0.01,'feature_fraction': 0.8, 'bagging_fraction': 0.8, 'bagging_freq': 2, 'num_leaves': 16, 'max_depth': 5, 'n_jobs': -1, 'seed': 2019, 'verbosity': -1, }oof_train = np.zeros(len(train_x))
preds = np.zeros(len(test_x))
folds = 5
seeds = [2048, 1997]
for seed in seeds: kfold = KFold(n_splits=folds, shuffle=True, random_state=seed)for fold, (trn_idx, val_idx) in enumerate(kfold.split(train_x, train_y)):print('fold ', fold + 1)x_trn, y_trn, x_val, y_val = train_x.iloc[trn_idx], train_y.iloc[trn_idx], train_x.iloc[val_idx], train_y.iloc[val_idx]train_set = lgb.Dataset(x_trn, y_trn)val_set = lgb.Dataset(x_val, y_val)model = lgb.train(params, train_set, num_boost_round=5000,valid_sets=(train_set, val_set), early_stopping_rounds=25,verbose_eval=50)oof_train[val_idx] += model.predict(x_val) / len(seeds)preds += model.predict(test_x) / folds / len(seeds)del x_trn, y_trn, x_val, y_val, model, train_set, val_setgc.collect()mse = (mean_squared_error(oof_train, train['LABEL2']))print('-'*120)print('rmse ', round(mse, 5))test1['LABEL2'] = preds

5.结果文件输出

由于我们对原文件做了较大的改动，上交的文件要处理成原文件的格式：

test11 = pd.read_csv('D:/mathorup/test1.csv',encoding="gbk")#短期测试集
test11['上行业务量GB'] = test1['LABEL1']
test11['下行业务量GB'] = test1['LABEL2']
test11.to_csv('短期验证选择的小区数据集.csv', index = False)
test11.head()