SimPSI: A Simple Strategy to Preserve Spectral Information in Time Series Data Augmentation

Paper

Official source code for "SimPSI: A Simple Strategy to Preserve Spectral Information in Time Series Data Augmentation", AAAI 2024 [1].

We propose a simple strategy to preserve spectral information (SimPSI) in time series data augmentation. SimPSI preserves the spectral information by mixing the original and augmented input spectrum weighted by a preservation map, which indicates the importance score of each frequency. Specifically, our experimental contributions are to build three distinct preservation maps: magnitude spectrum, saliency map, and spectrum-preservative map.

Prerequisites

• Linux or macOS
• Python 3
• CPU or NVIDIA GPU + CUDA CuDNN

Signal Demodulation (Simulation)

• Simulation dataset can be generated by running the code on MATLAB.

• Training with SimPSI (Magnitude spectrum and Saliency map)

cd simulation
python train_nonctr.py --data_name matlab_awgn_32fsk_snr_p10p28 --exp_name ${EXP_NAME} --n_class 32 --aug_list ${AUG_LIST} --prior ${PRSRVN_MAP} --seed ${SEED}

where EXP_NAME is the name of the experiment, AUG_LIST is a sequence of random augmentations (e.g., scale shift jitter), PRSRVN_MAP is a type of preservation map (i.e., mag or slc), and SEED is a random seed value.

• Training with SimPSI (Spectrum-preservative map)

cd simulation
python train_ctr.py --data_name matlab_awgn_32fsk_snr_p10p28 --exp_name ${EXP_NAME} --n_class 32 --aug_list ${AUG_LIST} --prior self --equalizer transformer --seed 

where EXP_NAME is the name of the experiment, AUG_LIST is a sequence of random augmentations (e.g., scale shift jitter), and SEED is a random seed value.

• Testing

cd simulation
python test.py --data_name matlab_awgn_32fsk_snr_p10p28 --exp_name ${EXP_NAME} --n_class 32

where EXP_NAME is the name of the experiment.

Human Activity Recognition (HAR)

• Training and Testing with SimPSI (Magnitude spectrum and Saliency map)

cd har_sleepedf
python main.py --experiment_description har_${PRSRVN_MAP} --run_description ${EXP_NAME} --selected_dataset HAR --aug_list ${AUG_LIST} --prior ${PRSRVN_MAP} --equalizer conv --eq_kernel_size 9 --mode ce --seed ${SEED}

where PRSRVN_MAP is a type of preservation map (i.e., mag or slc), EXP_NAME is the name of the experiment, AUG_LIST is a sequence of random augmentations (e.g., scale shift jitter), and SEED is a random seed value.

• Training and Testing with SimPSI (Spectrum-preservative map)

cd har_sleepedf
python main.py --experiment_description har_self --run_description ${EXP_NAME} --selected_dataset HAR --aug_list ${AUG_LIST} --prior self --equalizer transformer --mode ctr --seed ${SEED}

where EXP_NAME is the name of the experiment, AUG_LIST is a sequence of random augmentations (e.g., scale shift jitter), and SEED is a random seed value.

Sleep Stage Detection (SleepEDF)

• Training and Testing commands are the same as HAR, by replacing har and HAR to sleepedf and SleepEDF, respectively.

Atrial Fibrillation Classification (Waveform)

• Training and Testing with SimPSI (Magnitude spectrum and Saliency map)

cd waveform
python -m evaluations.main_nonctr --experiment_description waveform_${PRSRVN_MAP} --run_description ${EXP_NAME} --aug_list ${AUG_LIST} --prior ${PRSRVN_MAP} --seed ${SEED}

where PRSRVN_MAP is a type of preservation map (i.e., mag or slc), EXP_NAME is the name of the experiment, AUG_LIST is a sequence of random augmentations (e.g., scale shift jitter), and SEED is a random seed value.

• Training and Testing with SimPSI (Spectrum-preservative map)

cd waveform
python -m evaluations.main_ctr --experiment_description waveform_self --run_description ${EXP_NAME} --aug_list ${AUG_LIST} --prior self --seed ${SEED}

where EXP_NAME is the name of the experiment, AUG_LIST is a sequence of random augmentations (e.g., scale shift jitter), and SEED is a random seed value.

Citation

@article{ryu2023simpsi,
  title={SimPSI: A Simple Strategy to Preserve Spectral Information in Time Series Data Augmentation},
  author={Ryu, Hyun and Yoon, Sunjae and Yoon, Hee Suk and Yoon, Eunseop and Yoo, Chang D.},
  journal={arXiv preprint arXiv:2312.05790},
  year={2023}
}