This repo contains code to generate learning curves. The primary application is the supervised problem of predicting response of cancer cell lines to anti-cancer drugs.
There are three main scripts you need to run:
- Genrate topN dataset (build_topN.py)
- Genrate the data splits (gen_data_splits.py)
- Use data splits from (2) to generate learning curves (main_lrn_crv.py)
Script (1) requires to have a folder called "data" that contains a set of required files. You can just copy the folder /vol/ml/apartin/projects/candle/data to your parent dir.
- First, run the 1st script as follows:
python build_topN.py --top_n 6 --format parquet --labels
This will create dir called top6_data. The folder will contain a single parquet file. In addition, some plots are generated.
- Then, run the 2nd script to generate the data splits:
python gen_data_splits.py --dirpath top6_data
This will create dir called top6_data_splits that contains splits for various k-folds. In addition data files are generated xdata.parquet and meta.parquet.
- Finally, run the 3rd script to generate the learning curves:
python main_lrn_crv.py --dirpath top6_data --clr_mode trng1