atlas_dl

Running the code on Cori

Training on N examples

./run_cori.sh --num_tr N

Training on all examples

./run_cori.sh

or

./run_cori.sh --num_tr -1

Testing with the Weights from the Run from 1/19/2017

./run_cori.sh --test --load_path ./results/run111/models/net_best_val_loss.pkl

If you moved the data

./run_cori.sh --tr_file path/to/your/tr_file --val_file path/to/your/val_file

For test

./run_cori.sh --test --test_file path/to/your/test_file --load_path path/to/your/weights

Running as batch script on Cori

• same commands as above but replace "./run_cori.sh" with "sbatch cori_batch.sl"

Checking out the Other Command Line Args

./run_cori.sh --help

Running the code on a NERSC notebook:

module load deeplearning

• go to jupyter.nersc.gov
• open atlas_main.ipynb
• run the cells

Merging all events, then splitting them into train, val and test

./preproc_files.sh --source_path path/where/initial/input/files/are --dest_path path/where/you/want/to/put/trvaltest/files --suffix "string_to_append_to_files_for_your_own_benefit"

• if you pick suffix to be "_run5", then in your dest_path will be four files:

  • all_data_merged_run5.h5
  • train_run5.h5
  • test_run5.h5
  • val_run5.h5

• you can delete all_data_merged_run5.h5, which is all your data in one file, to save space or you can keep it for maybe a new tr,val,test split later

running the code on GPU number K for N examples

./run_maeve.sh K --num_tr N