round1

ECG Multi-Label Classfication - Round 1

1. Structure

• code: scripts for preprocessing, training and testing;
• data: Round 1 raw data;
• user_data: data generated by user, including models;
• prediction_result: prediction output.

2. Run

Only test scripts in Ubuntu 18.04 and CentOS 7.6.

cd code
./main.sh

Three commands in main.sh:

# Preprocessing
python ./prep/ecg_prep.py \
    -a ../data/hf_round1_arrythmia.txt \
    -tt ../data/hf_round1_label.txt \
    -at ../data/hefei_round1_ansA_20191008.txt \
    -td ../data/hf_round1_train/train \
    -ad ../data/hf_round1_testA/testA \
    -o ../user_data

# Train model
python ./train/ecg_train.py \
    -l ../user_data/train_testA_noDup.csv \
    -m ../user_data/models_large \
    -b large -nt 6 -c 5 -g 0

# Test model
# Ensemble 5 predictions
python ./test/ecg_test.py \
    -s ../data/hf_round1_testB_noDup_rename/testB_noDup_rename \
    -m ../user_data/models_large \
    -t ../data/hf_round1_subB_noDup_rename.txt \
    -a ../data/hf_round1_arrythmia.txt \
    -o ../prediction_result \
    -b large -g 0

You can also run these three commands one by one.