AutoDA-Timeseries

This repository contains the cleaned AutoDA-Timeseries implementation for time-series data augmentation experiments.

Project website: https://netmanaiops.github.io/AutoDA-Timeseries/

Included

• autoaugment/AutoDA_Timeseries.py: AutoDA-Timeseries augmentation policy.
• autoaugment/augments/basic_transforms.py: time-series augmentation operators used by AutoDA.
• exp/: training and evaluation loops for classification, regression, forecasting, and anomaly detection.
• dataloader/: dataset adapters and feature extraction.
• downstream/: downstream models.
• examples/AutoDA-Timeseries.classification.json: example AutoDA configuration.

Historical augmentation baselines and backup files are intentionally not included.

Example

source /workspace/douzj/AutoTSA-master/douzj_AutoTSA/bin/activate
python -u run.py \
  --task classification \
  --tsa AutoDA-Timeseries \
  --downstream ROCKET \
  --feature_extractor Catch22 \
  --tsa_config_path examples/AutoDA-Timeseries.classification.json \
  --dataset_root /workspace/douzj/AutoTSA-master/dataset \
  --dataset ArticularyWordRecognition \
  --dataset_type Default \
  --use_gpu 1 \
  --batch_size 16 \
  --learning_rate 0.005 \
  --train_epochs 35 \
  --patience 17

The classification helper script is available at:

bash scripts/classification/AutoDA_Timeseries.classification_distri.sh