Function reference • luz

Training

setup(): Set's up a nn_module to use with luz

fit(<luz_module_generator>): Fit a nn_module

predict(<luz_module_fitted>): Create predictions for a fitted model

evaluate(): Evaluates a fitted model on a dataset

set_hparams(): Set hyper-parameter of a module

set_opt_hparams(): Set optimizer hyper-parameters

get_metrics(): Get metrics from the object

ctx: Context object

context: Context object

lr_finder(): Learning Rate Finder

as_dataloader(): Creates a dataloader from its input

Metrics

luz_metric(): Creates a new luz metric

luz_metric_accuracy(): Accuracy

luz_metric_binary_accuracy(): Binary accuracy

luz_metric_binary_accuracy_with_logits(): Binary accuracy with logits

luz_metric_binary_auroc(): Computes the area under the ROC

luz_metric_mae(): Mean absolute error

luz_metric_mse(): Mean squared error

luz_metric_multiclass_auroc(): Computes the multi-class AUROC

luz_metric_rmse(): Root mean squared error

luz_metric_set(): Creates a metric set

Misc

nn_mixup_loss(): Loss to be used with callbacks_mixup().

nnf_mixup(): Mixup logic

Callbacks

luz_callback(): Create a new callback

luz_callback_auto_resume(): Resume training callback

luz_callback_csv_logger(): CSV logger callback

luz_callback_early_stopping(): Early stopping callback

luz_callback_gradient_clip(): Gradient clipping callback

luz_callback_interrupt(): Interrupt callback

luz_callback_keep_best_model(): Keep the best model

luz_callback_lr_scheduler(): Learning rate scheduler callback

luz_callback_metrics(): Metrics callback

luz_callback_mixed_precision(): Automatic Mixed Precision callback

luz_callback_mixup(): Mixup callback

luz_callback_model_checkpoint(): Checkpoints model weights

luz_callback_profile(): Profile callback

luz_callback_progress(): Progress callback

luz_callback_resume_from_checkpoint(): Allow resume model training from a specific checkpoint

luz_callback_tfevents(): tfevents callback

luz_callback_train_valid(): Train-eval callback

Accelerator

accelerator(): Create an accelerator

Serialization

luz_save(): Saves luz objects to disk

luz_load(): Load trained model

luz_load_model_weights() luz_save_model_weights(): Loads model weights into a fitted object.

luz_load_checkpoint(): Loads a checkpoint