torchfilter.train._train_kalman_filter_measurement

Private module; avoid importing from directly.

Module Contents

Functions

train_kalman_filter_measurement(buddy: fannypack.utils.Buddy, measurement_model: torchfilter.base.KalmanFilterMeasurementModel, dataloader: DataLoader, *, loss_function: Callable[[torch.Tensor, torch.Tensor], torch.Tensor] = F.mse_loss, log_interval: int = 10, optimizer_name='train_kalman_filter_measurement') → None

Optimizes a Kalman filter measurement model's prediction accuracy. Minimizes

torchfilter.train._train_kalman_filter_measurement.train_kalman_filter_measurement(buddy: fannypack.utils.Buddy, measurement_model: torchfilter.base.KalmanFilterMeasurementModel, dataloader: DataLoader, *, loss_function: Callable[[torch.Tensor, torch.Tensor], torch.Tensor] = F.mse_loss, log_interval: int = 10, optimizer_name='train_kalman_filter_measurement') → None

Optimizes a Kalman filter measurement model’s prediction accuracy. Minimizes output mean error only; does not define a loss on uncertainty.

Parameters:

• buddy (fannypack.utils.Buddy) – Training helper.

• measurement_model (torchfilter.base.KalmanFilterMeasurementModel) – Model to train.

• dataloader (DataLoader) – Loader for a SingleStepDataset.

Keyword Arguments:

• loss_function (callable, optional) – Loss function, from torch.nn.functional. Defaults to MSE.

• log_interval (int, optional) – Minibatches between each Tensorboard log.