dair_pll.data_config

class dair_pll.data_config.TrajectorySliceConfig(t_skip=0, t_history=1, t_prediction=1)

Bases: object

dataclass() for configuring a trajectory slicing for training process.

t_skip: int = 0

Index of first time to predict from >= t_history - 1.

t_history: int = 1

Number of steps in initial condition for prediction, >= 1.

t_prediction: int = 1

Number of future steps to use during training/evaluation, >= 1.

class dair_pll.data_config.DataConfig(dt=0.001, train_fraction=0.5, valid_fraction=0.25, test_fraction=0.25, slice_config=<factory>, update_dynamically=False)

Bases: object

dataclass() for configuring a trajectory dataset.

dt: float = 0.001

Time step, > 0.

train_fraction: float = 0.5

Fraction of training trajectories to select, <= 1, >= 0.

valid_fraction: float = 0.25

Fraction of validation trajectories to select, <= 1, >= 0.

test_fraction: float = 0.25

Fraction of testing trajectories to select, <= 1, >= 0.

slice_config: dair_pll.data_config.TrajectorySliceConfig

Config for arranging trajectories into times slices for training.

update_dynamically: bool = False

Whether to check for new trajectories after each epoch.