The code provided implements W-Augment, alpha-trimmed Augment and Rand Augment and evaluates these methods on the UCR Archive dataset using the InceptionTime network described in section 4.3.1, https://arxiv.org/abs/2102.08310 Usage Summary ------------- 1) conda create --name <env> --file requirements.txt 2) Download UCR 2018 Archive data from http://www.timeseriesclassification.com/ and change Line 385 in the code to change the data path 3) python3 main.py --run_path <dest_dir> --augment w_augment This will create a folder <dest_dir> with the output of the run which consists of two folders: UCR_results with the individual results for each model and the resulting ensemble on each dataset, and a summary_results folder which includes the summary metrics of the full run. ------------------------------------------------------------------------------ usage: main.py [-h] [--run_path RUN_PATH] [--n_epochs N_EPOCHS] [--n_iters N_ITERS] [--datasets DATASETS] [--augment {baseline,rand_augment,w_augment,atrim_augment}] [--param_M PARAM_M] optional arguments: -h, --help show this help message and exit --run_path RUN_PATH (default: ./) --n_epochs N_EPOCHS (default: 1500) --n_iters N_ITERS (default: 5) --datasets DATASETS (default: all) --augment {baseline,rand_augment,w_augment,atrim_augment} (default: None) --param_M PARAM_M (default: 10)