Steps

ACE’s code uses a “step” registry system to allow various emulation configuration step objects to be specified. (A step object consists of a specific configuration of NN module calls and other operations such as normalization, denormalization, correction, etc.). In ACE’s hierarchy, a stepper contains the step object, which in turn may contain one or more NN modules. Step registry is managed by the fme.ace.StepSelector configuration class, which is used to select a step type and version.

class fme.ace.StepSelector(type, config)[source]

Bases: StepConfigABC

Parameters:
config: dict[str, Any]
classmethod get_available_types()[source]

This class method is used to expose all available types of Steps.

Return type:

set[str]

get_loss_normalizer(extra_names=None, extra_residual_scaled_names=None)[source]
Parameters:

  • extra_names (Optional[list[str]], default: None) – Names of additional variables to include in the loss normalizer.

  • extra_residual_scaled_names (Optional[list[str]], default: None) – extra_names which use residual scale factors, if enabled.

Return type:

StandardNormalizer

Returns:

The loss normalizer.

get_next_step_forcing_names()[source]
Return type:

list[str]

get_ocean()[source]
Return type:

Optional[OceanConfig]

get_step(dataset_info, init_weights=<function StepSelector.<lambda>>)[source]
Parameters:

  • dataset_info (DatasetInfo) – Information about the training dataset.

  • init_weights (Callable[[list[Module]], None], default: <function StepSelector.<lambda> at 0x76e045ce0ae0>) – Function to initialize the weights of the step before wrapping in DistributedDataParallel. This is particularly useful when freezing parameters, as the DistributedDataParallel will otherwise expect frozen weights to have gradients, and will raise an exception.

Return type:

StepABC

Returns:

The state of the stepper.

property input_names: list[str]
load()[source]

Update configuration in-place so it does not depend on external files.

property loss_names: list[str]

Names of variables to be included in the loss function.

property n_ic_timesteps: int
property next_step_input_names: list[str]

Names of variables required in next_step_input_data for .step.

property output_names: list[str]

Names of variables output by the step.

classmethod register(name)[source]

Register a virtual subclass of an ABC.

Returns the subclass, to allow usage as a class decorator.

Return type:

Callable[[Type[TypeVar(T, bound= StepConfigABC)]], Type[TypeVar(T, bound= StepConfigABC)]]

Parameters:

name (str) –

registry: ClassVar[Registry] = <fme.core.registry.registry.Registry object>
replace_ocean(ocean)[source]
Parameters:

ocean (OceanConfig | None) –

type: str

The following step types are available:

{'FCN3', 'multi_call', 'separate_radiation', 'single_module'}
fme.core.step.StepSelector.get_available_types()

This class method is used to expose all available types of Steps.

Return type:

set[str]

The following step builders are available:

class fme.core.step.SingleModuleStepConfig(builder, in_names, out_names, normalization, ocean=None, corrector=<factory>, next_step_forcing_names=<factory>, residual_prediction=False)[source]

Bases: StepConfigABC

Configuration for a single module stepper.

Parameters:

  • builder (ModuleSelector) – The module builder.

  • in_names (list[str]) – Names of input variables.

  • out_names (list[str]) – Names of output variables.

  • normalization (NetworkAndLossNormalizationConfig) – The normalization configuration.

  • ocean (Optional[OceanConfig], default: None) – The ocean configuration.

  • corrector (AtmosphereCorrectorConfig | CorrectorSelector, default: <factory>) – The corrector configuration.

  • next_step_forcing_names (list[str], default: <factory>) – Names of forcing variables for the next timestep.

  • residual_prediction (bool, default: False) – Whether to use residual prediction.

builder: ModuleSelector
corrector: AtmosphereCorrectorConfig | CorrectorSelector
property diagnostic_names: list[str]

Names of variables which are outputs only.

classmethod from_state(state)[source]
Return type:

SingleModuleStepConfig

get_loss_normalizer(extra_names=None, extra_residual_scaled_names=None)[source]
Parameters:

  • extra_names (Optional[list[str]], default: None) – Names of additional variables to include in the loss normalizer.

  • extra_residual_scaled_names (Optional[list[str]], default: None) – extra_names which use residual scale factors, if enabled.

Return type:

StandardNormalizer

Returns:

The loss normalizer.

get_next_step_forcing_names()[source]

Names of input-only variables which come from the output timestep.

Return type:

list[str]

get_ocean()[source]
Return type:

Optional[OceanConfig]

get_state()[source]
get_step(dataset_info, init_weights)[source]
Parameters:

  • dataset_info (DatasetInfo) – Information about the training dataset.

  • init_weights (Callable[[list[Module]], None]) – Function to initialize the weights of the step before wrapping in DistributedDataParallel. This is particularly useful when freezing parameters, as the DistributedDataParallel will otherwise expect frozen weights to have gradients, and will raise an exception.

Return type:

SingleModuleStep

Returns:

The state of the stepper.

in_names: list[str]
property input_names: list[str]

Names of variables required as inputs to step, either in input or next_step_input_data.

load()[source]

Update configuration in-place so it does not depend on external files.

property loss_names: list[str]

Names of variables to be included in the loss function.

property n_ic_timesteps: int
next_step_forcing_names: list[str]
property next_step_input_names: list[str]

Names of variables provided in next_step_input_data.

normalization: NetworkAndLossNormalizationConfig
ocean: Optional[OceanConfig] = None
out_names: list[str]
property output_names: list[str]

Names of variables output by the step.

replace_ocean(ocean)[source]

Replace the ocean model with a new one.

Parameters:

ocean (Optional[OceanConfig]) – The new ocean model configuration or None.

residual_prediction: bool = False
class fme.core.step.MultiCallStepConfig(wrapped_step, config=None, include_multi_call_in_loss=True)[source]

Bases: StepConfigABC

Configuration for a multi-call step.

Parameters:

  • wrapped_step (StepSelector) – The step to wrap.

  • config (Optional[MultiCallConfig], default: None) – The multi-call configuration.

  • include_multi_call_in_loss (bool, default: True) – Whether to include multi-call diagnostics in the loss.

build(step_method)[source]
Return type:

Optional[MultiCall]

Parameters:

step_method (Callable[[StepArgs, Callable[[Module], Module]], dict[str, torch.Tensor]]) –

config: Optional[MultiCallConfig] = None
extend_normalizer_with_multi_call_outputs(normalizer)[source]

Extend the normalizer by setting multi-call output names to use the same normalization as their base counterparts.

Return type:

StandardNormalizer

Parameters:

normalizer (StandardNormalizer) –

get_loss_normalizer(extra_names=None, extra_residual_scaled_names=None)[source]

Get the loss normalizer for the multi-call step.

Normalizer will use statistics from multi-call variables in the stats dataset, meaning the normalization for multi-call output versions will be different from the normalization for the base variables.

Parameters:

  • extra_names (Optional[list[str]], default: None) – Names of additional variables to include in the loss normalizer.

  • extra_residual_scaled_names (Optional[list[str]], default: None) – extra_names which use residual scale factors, if enabled.

Return type:

StandardNormalizer

get_next_step_forcing_names()[source]
Return type:

list[str]

get_ocean()[source]
Return type:

Optional[OceanConfig]

get_step(dataset_info, init_weights)[source]
Parameters:

  • dataset_info (DatasetInfo) – Information about the training dataset.

  • init_weights (Callable[[list[Module]], None]) – Function to initialize the weights of the step before wrapping in DistributedDataParallel. This is particularly useful when freezing parameters, as the DistributedDataParallel will otherwise expect frozen weights to have gradients, and will raise an exception.

Return type:

MultiCallStep

Returns:

The state of the stepper.

include_multi_call_in_loss: bool = True
property input_names: list[str]
load()[source]

Update configuration in-place so it does not depend on external files.

property loss_names: list[str]

Names of variables to be included in the loss function.

property n_ic_timesteps: int
property next_step_input_names: list[str]

Names of variables required in next_step_input_data for .step.

property output_names: list[str]

Names of variables output by the step.

replace_multi_call(multi_call)[source]
Parameters:

multi_call (MultiCallConfig | None) –

replace_ocean(ocean)[source]
Parameters:

ocean (OceanConfig | None) –

wrapped_step: StepSelector
class fme.core.step.SeparateRadiationStepConfig(builder, radiation_builder, main_prognostic_names, shared_forcing_names, radiation_only_forcing_names, radiation_diagnostic_names, main_diagnostic_names, normalization, next_step_forcing_names=<factory>, ocean=None, corrector=<factory>, detach_radiation=False, residual_prediction=False)[source]

Bases: StepConfigABC

Configuration for a separate radiation stepper.

Parameters:

  • builder (ModuleSelector) – The module builder.

  • radiation_builder (ModuleSelector) – The radiation module builder.

  • main_prognostic_names (list[str]) – Names of prognostic variables. These are provided as input to both the main and radiation models, and output by the main model.

  • shared_forcing_names (list[str]) – Names of forcing variables.

  • radiation_only_forcing_names (list[str]) – Names of forcing variables for the radiation model, in addition to the ones specified in shared_forcing_names.

  • radiation_diagnostic_names (list[str]) – Names of diagnostic variables for the radiation model.

  • main_diagnostic_names (list[str]) – Names of diagnostic variables for the main model.

  • normalization (NetworkAndLossNormalizationConfig) – The normalization configuration.

  • next_step_forcing_names (list[str], default: <factory>) – Names of forcing variables which come from the output timestep.

  • ocean (Optional[OceanConfig], default: None) – The ocean configuration.

  • corrector (AtmosphereCorrectorConfig | CorrectorSelector, default: <factory>) – The corrector configuration.

  • detach_radiation (bool, default: False) – Whether to detach the output of the radiation model before passing it to the main model. The radiation outputs returned by .step() will not be detached.

  • residual_prediction (bool, default: False) – Whether to use residual prediction.

builder: ModuleSelector
corrector: AtmosphereCorrectorConfig | CorrectorSelector
detach_radiation: bool = False
property diagnostic_names: list[str]
classmethod from_state(state)[source]
Return type:

SeparateRadiationStepConfig

get_loss_normalizer(extra_names=None, extra_residual_scaled_names=None)[source]
Parameters:

  • extra_names (Optional[list[str]], default: None) – Names of additional variables to include in the loss normalizer.

  • extra_residual_scaled_names (Optional[list[str]], default: None) – extra_names which use residual scale factors, if enabled.

Return type:

StandardNormalizer

Returns:

The loss normalizer.

get_next_step_forcing_names()[source]
Return type:

list[str]

get_ocean()[source]
Return type:

Optional[OceanConfig]

get_state()[source]
get_step(dataset_info, init_weights)[source]
Parameters:

  • dataset_info (DatasetInfo) – Information about the training dataset.

  • init_weights (Callable[[list[Module]], None]) – Function to initialize the weights of the step before wrapping in DistributedDataParallel. This is particularly useful when freezing parameters, as the DistributedDataParallel will otherwise expect frozen weights to have gradients, and will raise an exception.

Return type:

SeparateRadiationStep

Returns:

The state of the stepper.

property input_names: list[str]
load()[source]

Update configuration in-place so it does not depend on external files.

property loss_names: list[str]

Names of variables to be included in the loss function.

main_diagnostic_names: list[str]
property main_in_names: list[str]
property main_out_names: list[str]
main_prognostic_names: list[str]
property n_ic_timesteps: int
next_step_forcing_names: list[str]
property next_step_input_names: list[str]

Names of variables provided in next_step_input_data.

normalization: NetworkAndLossNormalizationConfig
ocean: Optional[OceanConfig] = None
property output_names: list[str]

Names of variables output by the step.

radiation_builder: ModuleSelector
radiation_diagnostic_names: list[str]
property radiation_in_names: list[str]
radiation_only_forcing_names: list[str]
property radiation_out_names: list[str]
replace_ocean(ocean)[source]
Parameters:

ocean (OceanConfig | None) –

residual_prediction: bool = False
shared_forcing_names: list[str]