写在前面
该文章主要内容是全球城市计算AI挑战赛的基本分析和Baseline.
基本介绍
大赛以“地铁乘客流量预测”为赛题，参赛者可通过分析地铁站的历史刷卡数据，预测站点未来的客流量变化，帮助实现更合理的出行路线选择，规避交通堵塞，提前部署站点安保措施等，最终实现用大数据和人工智能等技术助力未来城市安全出行。
完整代码
代码语言：javascript复制import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
import lightgbm as lgb
import xgboost as xgb
from sklearn.model_selection import KFold, StratifiedKFold
from sklearn.metrics import accuracy_score, roc_auc_score
from scipy.stats import norm, rankdata
import warnings
import gc
import os
import time
import sys
import datetime
import matplotlib.pyplot as plt
import seaborn as sns
from sklearn.metrics import mean_squared_error
warnings.simplefilter(action='ignore', category=FutureWarning)
warnings.filterwarnings('ignore')
from sklearn import metrics
plt.style.use('seaborn')
sns.set(font_scale=1)
pd.set_option('display.max_columns', 500)此处直接粘贴复制，很多库都用不上。
代码语言：javascript复制path = 'input'
test = pd.read_csv(path '/Metro_testA/testA_submit_2019-01-29.csv')
test_28 = pd.read_csv(path '/Metro_testA/testA_record_2019-01-28.csv')构造基本的特征，主要为时间特征，count，sum等。
代码语言：javascript复制def get_base_features(df_):
df = df_.copy()
# base time
df['day'] = df['time'].apply(lambda x: int(x[8:10]))
df['week'] = pd.to_datetime(df['time']).dt.dayofweek 1
df['weekend'] = (pd.to_datetime(df.time).dt.weekday >=5).astype(int)
df['hour'] = df['time'].apply(lambda x: int(x[11:13]))
df['minute'] = df['time'].apply(lambda x: int(x[14:15] '0'))
# count,sum
result = df.groupby(['stationID', 'week', 'weekend', 'day', 'hour', 'minute']).status.agg(['count', 'sum']).reset_index()
# nunique
tmp = df.groupby(['stationID'])['deviceID'].nunique().reset_index(name='nuni_deviceID_of_stationID')
result = result.merge(tmp, on=['stationID'], how='left')
tmp = df.groupby(['stationID','hour'])['deviceID'].nunique().reset_index(name='nuni_deviceID_of_stationID_hour')
result = result.merge(tmp, on=['stationID','hour'], how='left')
tmp = df.groupby(['stationID','hour','minute'])['deviceID'].nunique().
reset_index(name='nuni_deviceID_of_stationID_hour_minute')
result = result.merge(tmp, on=['stationID','hour','minute'], how='left')
# in,out
result['inNums'] = result['sum']
result['outNums'] = result['count'] - result['sum']
#
result['day_since_first'] = result['day'] - 1
result.fillna(0, inplace=True)
del result['sum'],result['count']
return result
data = get_base_features(test_28)加载所有文件数据
代码语言：javascript复制data_list = os.listdir(path '/Metro_train/')
for i in range(0, len(data_list)):
if data_list[i].split('.')[-1] == 'csv':
print(data_list[i], i)
df = pd.read_csv(path '/Metro_train/' data_list[i])
df = get_base_features(df)
data = pd.concat([data, df], axis=0, ignore_index=True)
else:
continue由于26和27号数据不存在，故剔除所有周末的数据。
代码语言：javascript复制# 剔除周末,并修改为连续时间
data = data[(data.day!=5)&(data.day!=6)]
data = data[(data.day!=12)&(data.day!=13)]
data = data[(data.day!=19)&(data.day!=20)]
data = data[(data.day!=26)&(data.day!=27)]
def fix_day(d):
if d in [1,2,3,4]:
return d
elif d in [7,8,9,10,11]:
return d - 2
elif d in [14,15,16,17,18]:
return d - 4
elif d in [21,22,23,24,25]:
return d - 6
elif d in [28]:
return d - 8
data['day'] = data['day'].apply(fix_day)拼接测试集
代码语言：javascript复制test['week'] = pd.to_datetime(test['startTime']).dt.dayofweek 1
test['weekend'] = (pd.to_datetime(test.startTime).dt.weekday >=5).astype(int)
test['day'] = test['startTime'].apply(lambda x: int(x[8:10]))
test['hour'] = test['startTime'].apply(lambda x: int(x[11:13]))
test['minute'] = test['startTime'].apply(lambda x: int(x[14:15] '0'))
test['day_since_first'] = test['day'] - 1
test = test.drop(['startTime','endTime'], axis=1)
data = pd.concat([data,test], axis=0, ignore_index=True)
stat_columns = ['inNums','outNums']提取前一天的记录作为特征
代码语言：javascript复制def get_refer_day(d):
if d == 20:
return 29
else:
return d 1
tmp = data.copy()
tmp_df = tmp[tmp.day==1]
tmp_df['day'] = tmp_df['day'] - 1
tmp = pd.concat([tmp, tmp_df], axis=0, ignore_index=True)
tmp['day'] = tmp['day'].apply(get_refer_day)
for f in stat_columns:
tmp.rename(columns={f: f '_last'}, inplace=True)
tmp = tmp[['stationID','day','hour','minute','inNums_last','outNums_last']]
data = data.merge(tmp, on=['stationID','day','hour','minute'], how='left')
data.fillna(0, inplace=True)按week,hour,minute分别对inNums和outNums构造统计特征
代码语言：javascript复制tmp = data.groupby(['stationID','week','hour','minute'], as_index=False)['inNums'].agg({
'inNums_whm_max' : 'max',
'inNums_whm_min' : 'min',
'inNums_whm_mean' : 'mean'
})
data = data.merge(tmp, on=['stationID','week','hour','minute'], how='left')
tmp = data.groupby(['stationID','week','hour','minute'], as_index=False)['outNums'].agg({
'outNums_whm_max' : 'max',
'outNums_whm_min' : 'min',
'outNums_whm_mean' : 'mean'
})
data = data.merge(tmp, on=['stationID','week','hour','minute'], how='left')
tmp = data.groupby(['stationID','week','hour'], as_index=False)['inNums'].agg({
'inNums_wh_max' : 'max',
'inNums_wh_min' : 'min',
'inNums_wh_mean' : 'mean'
})
data = data.merge(tmp, on=['stationID','week','hour'], how='left')
tmp = data.groupby(['stationID','week','hour'], as_index=False)['outNums'].agg({
#'outNums_wh_max' : 'max',
#'outNums_wh_min' : 'min',
'outNums_wh_mean' : 'mean'
})
data = data.merge(tmp, on=['stationID','week','hour'], how='left')恢复初始时间
代码语言：javascript复制def recover_day(d):
if d in [1,2,3,4]:
return d
elif d in [5,6,7,8,9]:
return d 2
elif d in [10,11,12,13,14]:
return d 4
elif d in [15,16,17,18,19]:
return d 6
elif d == 20:
return d 8
else:
return d
all_columns = [f for f in data.columns if f not in ['weekend','inNums','outNums']]
### all data
all_data = data[data.day!=29]
all_data['day'] = all_data['day'].apply(recover_day)
X_data = all_data[all_columns].values
train = data[data.day <20]
train['day'] = train['day'].apply(recover_day)
X_train = train[all_columns].values
valid = data[data.day==20]
valid['day'] = valid['day'].apply(recover_day)
X_valid = valid[all_columns].values
test = data[data.day==29]
X_test = test[all_columns].values构建模型并训练
代码语言：javascript复制params = {
'boosting_type': 'gbdt',
'objective': 'regression',
'metric': 'mae',
'num_leaves': 63,
'learning_rate': 0.01,
'feature_fraction': 0.9,
'bagging_fraction': 0.9,
'bagging_seed':0,
'bagging_freq': 1,
'verbose': 1,
'reg_alpha':1,
'reg_lambda':2
}
######################################################inNums
y_train = train['inNums']
y_valid = valid['inNums']
y_data = all_data['inNums']
lgb_train = lgb.Dataset(X_train, y_train)
lgb_evals = lgb.Dataset(X_valid, y_valid , reference=lgb_train)
gbm = lgb.train(params,
lgb_train,
num_boost_round=10000,
valid_sets=[lgb_train,lgb_evals],
valid_names=['train','valid'],
early_stopping_rounds=200,
verbose_eval=1000,
)
### all_data
lgb_train = lgb.Dataset(X_data, y_data)
gbm = lgb.train(params,
lgb_train,
num_boost_round=gbm.best_iteration,
valid_sets=[lgb_train],
valid_names=['train'],
verbose_eval=1000,
)
test['inNums'] = gbm.predict(X_test)
######################################################outNums
y_train = train['outNums']
y_valid = valid['outNums']
y_data = all_data['outNums']
lgb_train = lgb.Dataset(X_train, y_train)
lgb_evals = lgb.Dataset(X_valid, y_valid , reference=lgb_train)
gbm = lgb.train(params,
lgb_train,
num_boost_round=10000,
valid_sets=[lgb_train,lgb_evals],
valid_names=['train','valid'],
early_stopping_rounds=200,
verbose_eval=1000,
)
### all_data
lgb_train = lgb.Dataset(X_data, y_data)
gbm = lgb.train(params,
lgb_train,
num_boost_round=gbm.best_iteration,
valid_sets=[lgb_train],
valid_names=['train'],
verbose_eval=1000,
)
test['outNums'] = gbm.predict(X_test)
sub = pd.read_csv(path '/Metro_testA/testA_submit_2019-01-29.csv')
sub['inNums'] = test['inNums'].values
sub['outNums'] = test['outNums'].values
# 结果修正
sub.loc[sub.inNums<0 , 'inNums'] = 0
sub.loc[sub.outNums<0, 'outNums'] = 0
sub[['stationID', 'startTime', 'endTime', 'inNums', 'outNums']].to_csv('output/sub_model.csv', index=False)
