discotime.utils.misc module

class discotime.utils.misc.BaseFabricTrainer(module: LitSurvModule, dset_train: SurvDataset, dset_valid: SurvDataset, batch_size: int)

Bases: object

Bare-bones trainer class using Fabric.

This trainer class supports only a very limited set of the full lightning functionality. It can be useful for quick prototyping or for k-fold cross-validation as exemplified in the scripts that can be found in the /experiments folder on github.

class discotime.utils.misc.EarlyStopping(patience: int, min_delta: float, direction: str = 'minimize')

Bases: object

Lightning-free implementation of early stopping.

To be used with the bare-bones simple trainer that have no support for lightning callbacks.

should_stop(metric) → bool

class discotime.utils.misc.OptunaCallback(trial: Trial, monitor: str, minvalue: float = -inf, maxvalue: float = inf)

Bases: Callback

Lightning callback for hyperparameter optimization with Optuna.

Parameters:

• trial – A Trial corresponding to the current evaluation of the objective function.

• monitor – An evaluation metric for pruning, e.g., val_loss or val_acc.

on_validation_end(trainer: Trainer, module: LightningModule) → None

discotime.utils.misc.clamp(value: float, minvalue: float, maxvalue: float)

discotime.utils.misc.get_last_intermediate_value(trial: Trial)

discotime.utils.misc.recursive_defaultdict()

discotime.utils.misc.split_dataset(dataset: Dataset[T], k: int = 5) → Generator[tuple[torch.utils.data.dataset.ConcatDataset[T], torch.utils.data.dataset.Subset[T]], None, None]

Split dataset into k train/validation splits

discotime.utils.misc.update_mapping(d: Mapping, u: Mapping) → Mapping

Update nested mappings recursively.

Parameters:

• d (Mapping) – dictionary/mapping of things.

• u (Mapping) – other dictionary/mapping of things to update d with.

Example

>>> foo = {"dog" : {"color" : "black", "age" : 10}}
>>> bar = {"dog" : {"age" : 11}, "cat" : None}
>>> update_mapping(foo, bar)
{'dog': {'color': 'black', 'age': 11}, 'cat': None}