Are you sure you want to delete this task? Once this task is deleted, it cannot be recovered.
SCU-L JZ d09283a778 | 1 year ago | |
---|---|---|
model_files | 1 year ago | |
models | 1 year ago | |
submit | 1 year ago | |
utils | 1 year ago | |
xydata/raw | 1 year ago | |
.gitignore | 1 year ago | |
README.md | 1 year ago | |
inference_mini_batch.py | 1 year ago | |
requirements.txt | 1 year ago | |
train_mini_batch.py | 1 year ago |
这是第七届信也科技杯-欺诈用户风险识别RobertAckley的初赛代码。 测评AUC最终为0.83631,排名33。本代码主要参照比赛baseline代码 https://github.com/DGraphXinye/2022_finvcup_baseline
请在比赛网站上下载"初赛数据集.zip"文件,将zip文件中的"phase1_gdata.npz"放到路径'./xydata/raw'中。
Implementing environment:
详细见requirements.txt
python train_mini_batch.py --model sage_neighsampler --epochs 200 --device 0
python inference_mini_batch.py --model sage_neighsampler --device 0
Methods | Valid AUC | Test AUC |
---|---|---|
GraphSAGE (NeighborSampler) embedding + features + LightGBM | 0.8518 | 0.8363 |
本次比赛团队的解决方案主要包括三步:基于GraphSAGE的节点Embedding(与baseline一致),手工加入时序等特征,通过LightGBM分类
基于GraphSAGE的节点Embedding(与baseline一致)
基于baseline代码中GraphSAGE(NeighborSampler)的AUC最高,团队使用该网络对数据集中的节点进行embedding。网络训练与baseline中一致,修改的点为将17-128-2的GraphSAGE原模型修改为12-128-64-2的新模型,最后一层为64-2的线性层作为分类器。网络训练完成后,只取GraphSAGE的前两层,如此便将所有节点转换成了64维的向量。
值得注意的是,得到embedding时不再采用验证集(能使测试AUC上升0.003左右),因此inference_mini_batch.py中使用的是XYGraphP1_no_valid作为数据集类。
手工加入时序等特征
最终每个节点为201维的特征向量(由以下特征向量直接拼接得到),其中:
LightGBM分类
最终通过LightGBM分类器分类,参数设置具体见代码,1200个epoch的设置是由之前包含验证集的相同设置的实验推测而来。
基于GraphSAGE,加入时序特征,通过lightGBM进行分类
Python Text
Dear OpenI User
Thank you for your continuous support to the Openl Qizhi Community AI Collaboration Platform. In order to protect your usage rights and ensure network security, we updated the Openl Qizhi Community AI Collaboration Platform Usage Agreement in January 2024. The updated agreement specifies that users are prohibited from using intranet penetration tools. After you click "Agree and continue", you can continue to use our services. Thank you for your cooperation and understanding.
For more agreement content, please refer to the《Openl Qizhi Community AI Collaboration Platform Usage Agreement》