Skip to main content

OptimizationSpec

class OptimizationSpec(BaseModel)  # adk.models.optimization.optimization_spec
Internal / read-only

The ADK builds this from the platform payload when an optimization starts. Agent authors should use OptimizationContext (or EnvData / AgentData on the RL path), not construct or parse OptimizationSpec directly. step_system is implemented by the domain layer, not by user code.

Wire-format description of an optimization job: nodes, parameters, targets, and the batched simulator callback.

Definition​

class OptimizationSpec(BaseModel):
    inference: bool
    genie_model: str

    optimization_nodes: dict[str, OptimizationNodeSpec]

    static_parameters: dict[str, DesignParamSpec]
    optimized_parameters: dict[str, DesignParamSpec]
    randomized_parameters: dict[str, DesignParamSpec]

    targets: dict[str, TargetSpec]

    step_system: BatchedStepWorldCB

Members​

inference​

inference: bool

Indicates whether the optimization process is running in inference mode. When True, the system performs evaluation without learning or modifying parameters.

genie_model​

genie_model: str

Identifier or name of the Genie model used in the optimization process. Defines the model architecture or behavior under optimization.

optimization_nodes​

optimization_nodes: dict[str, OptimizationNodeSpec]

Dictionary containing the specification of all optimization nodes involved in the process. Each key is a node name, and each value is an OptimizationNodeSpec describing that node's configuration, parameters, and targets.

static_parameters​

static_parameters: dict[str, DesignParamSpec]

Dictionary of parameters with fixed values that remain constant during optimization. Typically represent system constants or non-tunable configurations.

optimized_parameters​

optimized_parameters: dict[str, DesignParamSpec]

Dictionary of parameters that are actively optimized by the system to achieve the defined objectives.

randomized_parameters​

randomized_parameters: dict[str, DesignParamSpec]

Dictionary of parameters that are randomly varied to introduce diversity or stochastic behavior in the optimization process.

targets​

targets: dict[str, TargetSpec]

Dictionary defining the optimization targets. Each entry is a runtime TargetSpec.

step_system​

step_system: BatchedStepWorldCB

A callable function that executes one batched optimization step in the system. It takes as input:

  • A dictionary mapping node and parameter names to their respective values (dict[str, dict[str, DesignParamValue]]).
  • Two boolean flags controlling step behavior (e.g., inference and reset conditions).

It returns a tuple containing:

  1. A dictionary of system observations.
  2. A dictionary of measured outcomes.
  3. A dictionary of additional metadata or runtime outputs.

DesignParamValue​

DesignParamValue = float | str

Represents a single design parameter value, which can be either numeric or string-based.

ObservationValue​

ObservationValue = float | list[float] | list[list[float]]

Represents the observed value from the system, supporting scalar, vector, and matrix-like data.

StepWorldCB​

StepWorldCB

A callable for performing a single (non-batched) world step, handling one set of parameters at a time.

BatchedStepWorldCB​

BatchedStepWorldCB

A callable for performing a batched world step, processing multiple parameter sets or optimization nodes simultaneously. Returns structured observation, measurement, and metadata dictionaries.