Skip to content

“合肥高新杯”心电人机智能大赛——心电异常事件预测 rank 17th solution

Notifications You must be signed in to change notification settings

yshanyes/hefei-prediction-of-abnormal-ecg-events

Repository files navigation

“合肥高新杯”心电人机智能大赛——心电异常事件预测 rank 17th solution

来源:"合肥高新杯"心电人机智能大赛

round2 Score=0.92696

系统环境:ubuntu python3.6 pytorch:latest

大致思路:每个导联作为一个通道,使用1维卷积进行训练

1、数据预处理

采用wfdb库QRS波提取每个导联心拍,然后对每个样本每个导联计算所有心拍median。data_process.py 对训练和验证集进行划分,beat_process.py 对心拍进行提取,

python round1_data_process.py 

python round1_beat_process.py

python round2_data_process.py

python round2_beat_process.py

2、模型训练

采用round1数据进行模型预训练,然后在round2数据上进行fine-tune.

python main.py train --model_name=mixnet_sm_pretrain

python main.py train_cv --model_name=mixnet_sm

python main.py train --model_name=mixnet_mm_pretrain

python main.py train_cv --model_name=mixnet_mm

python main.py val_cv --ckpt=ckpt/mixnet_sm_cv --model_name=mixnet_sm_predict

python main.py val_cv --ckpt=ckpt/mixnet_mm_cv --model_name=mixnet_mm_predict

3、模型测试

模型测试,在submit文件夹下生成提交结果

python main.py test_cv_ensemble --ckpt=ckpt #加载训练权重进行测试

一些细节

  1. 本次测试模型为1d mixnet;
  2. 训练数据没有只进行了简单的数据增强.

参考: baseline 答辩视频 top solution

About

“合肥高新杯”心电人机智能大赛——心电异常事件预测 rank 17th solution

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages