deepblink.training module

Training functions.

deepblink.training.run_experiment(cfg: Dict[KT, VT], pre_model: keras.engine.training.Model = None)

Run a training experiment.

Configuration file can be generated using deepblink config.

Parameters:

• cfg – Dictionary configuration file.
• pre_model – Pre-trained model if not training from scratch.

deepblink.training.train_model(model: deepblink.models._models.Model, dataset: deepblink.datasets._datasets.Dataset, cfg: Dict[KT, VT], run_name: str = 'model', use_wandb: bool = True) → deepblink.models._models.Model

Model training loop with callbacks.

Parameters:

• model – Model class with the .fit method.
• dataset – Dataset class with access to train and validation images.
• cfg – Configuration file equivalent to the one used in pink.training.run_experiment.
• run_name – Name given to the model.h5 file saved.
• use_wandb – If Wandb should be used.