nh.training

neuralhydrology.training.get_loss_obj(cfg: Config) → BaseLoss

Get loss object, depending on the run configuration.

Currently supported are ‘MSE’, ‘NSE’, ‘RMSE’, ‘GMMLoss’, ‘CMALLoss’, and ‘UMALLoss’.

Parameters:

cfg (Config) – The run configuration.

Returns:

A new loss instance that implements the loss specified in the config or, if different, the loss required by the head.

Return type:

loss.BaseLoss

neuralhydrology.training.get_optimizer(model: Module, cfg: Config) → Optimizer

Get specific optimizer object, depending on the run configuration.

Currently only ‘Adam’ and ‘AdamW’ are supported.

Parameters:

• model (torch.nn.Module) – The model to be optimized.

• cfg (Config) – The run configuration.

Returns:

Optimizer object that can be used for model training.

Return type:

torch.optim.Optimizer

neuralhydrology.training.get_regularization_obj(cfg: Config) → List[BaseRegularization]

Get list of regularization objects.

Currently, only the ‘tie_frequencies’ regularization is implemented.

Parameters:

cfg (Config) – The run configuration.

Returns:

List of regularization objects that will be added to the loss during training.

Return type:

List[regularization.BaseRegularization]

• BaseTrainer
  • BaseTrainer
    • BaseTrainer.initialize_training()
    • BaseTrainer.train_and_validate()
• Logger
  • Logger
    • Logger.log_figures()
    • Logger.log_step()
    • Logger.start_tb()
    • Logger.stop_tb()
    • Logger.summarise()
    • Logger.tag
    • Logger.train()
    • Logger.valid()
• loss
  • BaseLoss
    • BaseLoss.forward()
    • BaseLoss.set_regularization_terms()
  • MaskedCMALLoss
  • MaskedGMMLoss
  • MaskedMSELoss
  • MaskedNSELoss
  • MaskedRMSELoss
  • MaskedUMALLoss
• regularization
  • BaseRegularization
    • BaseRegularization.forward()
  • ForecastOverlapMSERegularization
    • ForecastOverlapMSERegularization.forward()
  • TiedFrequencyMSERegularization
    • TiedFrequencyMSERegularization.forward()
• train
  • start_training()