ML FW training - jniedzie/SVJanalysis_wiki GitHub Wiki

Training is done in train.py script. You should run it providing path to the config, for instance:

python train.py -c config/ae_default.py

Several other flags can be used, see usage with python train.py -h.
Here is a short description of other important flags:

• -gs/--global_seed: Fix the ML model seed to the provided value
• -v/--first_model_number: The training number of the first model that will be trained. The number of trainings to run from there is written in the config file. Will be inferred automatically if not provided. If sending several trainings in batch, we recommend doing one training per job and specifying the training number via this flag.

Important note:
A script tailored for the NAE is set up to perform training, evaluation and plotting! See run_nae_training_evaluation_and_plots_example.sh. You can find example usage in this script.