企业非法集资风险预测赛题-季军 四位靓仔 开源代码

# ===================== amount_feas 分箱特征 ===============
for fea in tqdm(amount_feas, desc="分箱特征"):
    # 通过除法映射到间隔均匀的分箱中，每个分箱的取值范围都是loanAmnt/1000
    data['{}_bin1'.format(fea)] = np.floor_divide(data[fea], 1000)
    ## 通过对数函数映射到指数宽度分箱
    data['{}_bin2'.format(fea)] = np.floor(np.log10(data[fea]))

# ===================== amount_feas 基本聚合特征 ===============
for f in tqdm(amount_feas, desc="amount_feas 基本聚合特征"):
    for cate in category_fea:
        if f != cate:
            data['{}_{}_medi'.format(cate, f)] = data.groupby(cate)[f].transform('median')
            data['{}_{}_mean'.format(cate, f)] = data.groupby(cate)[f].transform('mean')
            data['{}_{}_max'.format(cate, f)] = data.groupby(cate)[f].transform('max')
            data['{}_{}_min'.format(cate, f)] = data.groupby(cate)[f].transform('min')
            data['{}_{}_std'.format(cate, f)] = data.groupby(cate)[f].transform('std')

# =================== amount_feas 基本交叉特征  =============================
for f1 in tqdm(amount_feas, desc="amount_feas 基本交叉特征"):
    for f2 in amount_feas:
        if f1 != f2:
            data['{}_{}_ratio'.format(f1, f2)] = data[f1].values / data[f2].values
            data['{}_{}_multi'.format(f1, f2)] = data[f1].values * data[f2].values
            data['{}_{}_add'.format(f1, f2)] = data[f1].values + data[f2].values
            data['{}_{}_diff'.format(f1, f2)] = data[f1].values - data[f2].values
			

data['nmean'] = data[amount_feas].mean(1)
data['ntd'] = data[amount_feas].std(1)
data['nsum'] = data[amount_feas].sum(1)

for i in tqdm(category_fea, desc="类别特征nunique特征"):
    for j in category_fea:
        if i != j:
            data['nuni_{0}_{1}'.format(i, j)] = data[i].map(data.groupby(i)[j].nunique())

# ===================== 五折转化率特征 ====================


data['ID'] = data.index
data['fold'] = data['ID'] % 5
data.loc[data['label'].isnull(), 'fold'] = 5
target_feat = []
for i in tqdm(cat_list, desc="5折转化率特征"):
    target_feat.extend([i + '_mean_last_1'])
    data[i + '_mean_last_1'] = None
    for fold in range(6):
        data.loc[data['fold'] == fold, i + '_mean_last_1'] = data[data['fold'] == fold][i].map(
            data[(data['fold'] != fold) & (data['fold'] != 5)].groupby(i)['label'].mean()
        )
    data[i + '_mean_last_1'] = data[i + '_mean_last_1'].astype(float)
	

qiye/
├── data
│   ├── code 程序目录
│   │   ├── catboot.py
│   │   ├── ensemble.py
│   │   ├── gen_feas.py
│   │   ├── gen_feas_v2.py
│   │   ├── init.py
│   │   ├── lgb.py
│   │   ├── nohup.out
│   │   └── xgb.py
│   ├── prediction_result 运行结果文件
│   │   ├── cat.csv：catboost模型结果文件
│   │   ├── lgb.csv：lightgbm 模型结果文件
│   │   ├── result.csv 三个单模融合文件
│   │   └── xgb.csv xgboost模型结果文件
│   ├── raw_data 比赛数据目录
│   │   ├── entprise_evaluate.csv
│   │   ├── entprise_submit.csv
│   │   └── train
│   │       ├── annual_report_info.csv
│   │       ├── base_info.csv
│   │       ├── change_info.csv
│   │       ├── entprise_info.csv
│   │       ├── news_info.csv
│   │       ├── other_info.csv
│   │       └── tax_info.csv
│   ├── README.md
│   └── user_data

└── image
    ├── great.tar 打包完整的镜像
    ├── READEME.md
    └── run.sh  程序运行入口sh文件

Package         Version
--------------- -------
catboost        0.18.1
cycler          0.10.0
dill            0.3.3
graphviz        0.15
joblib          0.17.0
kiwisolver      1.3.1
lightgbm        2.3.1
matplotlib      3.3.3
numpy           1.17.4
pandarallel     1.5.1
pandas          1.1.4
Pillow          8.0.1
pip             20.3.1
plotly          4.14.1
pyparsing       2.4.7
python-dateutil 2.8.1
pytz            2020.4
retrying        1.3.3
scikit-learn    0.23.2
scipy           1.5.4
setuptools      41.0.1
six             1.15.0
threadpoolctl   2.1.0
tqdm            4.54.1
wheel           0.33.1
xgboost         0.90