AUTOMATA is a Gradient Based Data Subset Selection approach for Compute-Efficient Hyper-parameter Tuning. This repository incorporates various existing subset selection strategies for efficient hyper-parameter tuning.
This repository uses the code from CORDS and RayTune libraries.
- change the hyper-parametertuning (check configs/SL/config_hp_*.py) and model config files in AUTOMATA/configs/SL/ to your desired configurations
- Run text_run_sl.py for text datasets (or vision_run_sl.py for vision datasets) with appropriate arguments.
One example with some of the arguments passed ispython3 text_run_sl.py --config_file configs/SL/config_<dataset>.py --config_hp configs/SL/config_hp_hb.py --fraction 0.1 > output.log 2> error.log