Skip to main content

Model TargetSpec

class TargetSpec(BaseModel)  # adk.models.model

Pydantic model for one entry in target_specifications.json and GenieModel.target_specifications. Describes what targets an agent was trained for and how observations are ordered.

note

Not the same as the runtime TargetSpec inside OptimizationContext. Model specs are authored on disk; runtime specs are built by the ADK when an optimization starts.

Definition​

class TargetSpec(BaseModel):
    label: str
    kind: RuntimeType
    points: int
    precisions: float | list[float] | list[list[float]]
    precision_kind: PrecisionType = PrecisionType.Absolute
    fn: Fn = Fn.Range
    description: str = ""
    order: int

In JSON, fn, kind, and precision_kind are usually strings (see below). The Python types are Fn, RuntimeType, and PrecisionType enums.

Members​

label​

label: str

A unique identifier which can be used to map user defined expressions (on the web interface) to actual agent specifications.

description​

description: str = ""

Human-readable description describing the purpose of the target specification. For informal purposes only.

fn​

fn: Fn

Objective function for the target. JSON accepts "min", "max", "range", or "equals" (or integer codes 1–4).

kind​

kind: RuntimeType

Data shape for the target. JSON accepts "scalar", "vector", or "series" (or integer codes 0–2). Two-dimensional data is represented as series.

points​

points: int

Number of points required by the target. For vector and series targets, this represents the length. For series, this is the total number of elements across all rows. Not validated when kind is scalar.

precisions​

precisions: float | list[float] | list[list[float]]

Tolerance or precision for optimization:

  • When kind is scalar: a single float
  • When kind is vector: a list of floats (one per point)
  • When kind is series: a 2D list matching the series shape

Only applicable for range and equals objective functions.

precision_kind​

precision_kind: PrecisionType = PrecisionType.Absolute

Whether precisions are absolute values or percentages of the target mean. JSON accepts "absolute" / "percent" (or 1 / 2). Used when flattening target spaces in OptimizationEnv.

order​

order: int

Used to sort observations before they are provided to the agent; must be unique across all target specifications on a model.