ML_Training module

MLTrainingTask

class esteem.tasks.ml_training.MLTrainingTask(**kwargs)

run()

Main routine for the ML_Training task

Task Arguments

usage: __main__.py [-h] [--seed SEED] [--calc_suffix CALC_SUFFIX]
                   [--calc_dir_suffix CALC_DIR_SUFFIX]
                   [--calc_prefix CALC_PREFIX] [--target TARGET]
                   [--traj_prefix TRAJ_PREFIX] [--traj_suffix TRAJ_SUFFIX]
                   [--geom_prefix [GEOM_PREFIX]] [--ntraj NTRAJ]
                   [--restart [RESTART]] [--reset_loss [RESET_LOSS]]
                   [--which_trajs WHICH_TRAJS]
                   [--which_trajs_valid WHICH_TRAJS_VALID]
                   [--which_trajs_test WHICH_TRAJS_TEST]
                   [--traj_links TRAJ_LINKS]
                   [--traj_links_valid TRAJ_LINKS_VALID]
                   [--traj_links_test TRAJ_LINKS_TEST] [--cutoff CUTOFF]

Named Arguments

--seed, -s	Base name stem for the calculation (often the name of the molecule)
--calc_suffix, -S
	Suffix for the calculator Default: “”
--calc_dir_suffix, -D
	Suffix for the calculator directory
--calc_prefix, -P
	Prefix for the calculator (often specifies directory) Default: “”
--target, -t	Excited state index, zero for ground state Default: 0
--traj_prefix, -Q
	Prefix for the trajectory on which to train the calculator Default: “training”
--traj_suffix, -T
	Suffix for the trajectory on which to train the calculator Default: “training”
--geom_prefix	Prefix for the path at which to find the input geometry Default: “gs_PBE0/is_opt_{solv}”
--ntraj, -n	How many total trajectories (A,B,C…) with this naming are present for training Default: 1
--restart, -r	Whether to load a pre-existing calculator and resume training Default: False
--reset_loss, -R
	Whether to reset the loss function due to new training data being added Default: False
--which_trajs, -w
	Which trajectories (A,B,C…) with this naming are to be trained against
--which_trajs_valid, -v
	Which trajectories (A,B,C…) with this naming are to be validated against
--which_trajs_test, -u
	Which trajectories (A,B,C…) with this naming are to be tested against
--traj_links, -L
	Targets for links to create for training trajectories
--traj_links_valid, -V
	Targets for links to create for validation trajectories
--traj_links_test, -U
	Targets for links to create for testing trajectories
--cutoff, -d	Gaussian descriptor cutoff Default: 6.5

Standalone module routines

Defines a task to train a Machine Learning calculator on a trajectory of snapshots by calling the train() function of the MLWrapper