kinoml.ml.torch_loops

WIP

Module Contents

kinoml.ml.torch_loops.multi_measurement_training_loop(dataloaders, observation_models, model, optimizer, loss_function, epochs=100)

Standard training loop with multiple dataloaders and observation models.

Parameters

  • dataloaders (dict of str -> torch.utils.data.DataLoader) – key must refer to the measurement type present in the dataloader

  • observation_models (dict of str -> callable) – keys must be the same as in dataloaders, and point to pytorch-compatible callables that convert delta_over_kt to the corresponding measurement_type

  • model (torch.nn.Model) – instance of the model to train

  • optimizer (torch.optim.Optimizer) – instance of the optimization engine

  • loss_function (torch.nn.modules._Loss) – instance of the loss function to apply (e.g. MSELoss())

  • epochs (int) – number of iterations the loop will run

Returns

  • model (torch.nn.Model) – The trained model (same instance as provided in parameters)

  • loss_timeseries (list of float) – Cumulative loss per epoch

class kinoml.ml.torch_loops.EarlyStopping(patience=5, min_delta=0)

Early stopping to stop the training when the loss does not improve after certain epochs.

Taken from https://debuggercafe.com/using-learning-rate-scheduler-and-early-stopping-with-pytorch/

__call__(val_loss)
class kinoml.ml.torch_loops.LRScheduler(optimizer, patience=5, min_lr=1e-06, factor=0.5)

Learning rate scheduler. If the validation loss does not decrease for the given number of patience epochs, then the learning rate will decrease by by given factor.

Taken from https://debuggercafe.com/using-learning-rate-scheduler-and-early-stopping-with-pytorch/

__call__(val_loss)
kinoml.ml.torch_loops._old_training_loop()

Deprecated