koog-agents-optimization/ai.koog.agents.optimization.training/TrainingResources

TrainingResources

class TrainingResources<Input, Output, InputLabel>(val trackedAgent: GraphAIAgent<Input, Output>, val dataset: TrainSet<Input, InputLabel>, val substepPromptExecutor: PromptExecutor, val metric: (TrainSetItem<Input, InputLabel>, Output) -> Double, val threshold: Double, val serializers: DatasetExecutionSerializers<Input, Output, InputLabel>, val actionLogTruncation: ActionLogTruncation, val retriesPolicy: RetryPolicy, val failureAnalyzer: FailureAnalyzer, val capturingExecutorFactory: (PromptExecutor) -> CapturingPromptExecutor, val consumptionCollector: () -> LLMConsumption? = { null }, val runAgentAttempt: suspend (agent: GraphAIAgent<Input, Output>, input: Input) -> Pair<GraphAIAgent<Input, Output>, Output> = { agent, input -> agent to agent.run(input) }, val withItemContext: suspend (itemId: String, block: suspend () -> Unit) -> Unit = { _, block -> block() })

Shared execution resources created once per training session.

Holds the tracked agent, the injected execution collaborators (consumption capture, failure analysis, per-item context, agent invocation), and configuration shared across all scope instances.

These collaborators (failureAnalyzer, consumptionCollector, capturingExecutorFactory, runAgentAttempt, withItemContext) are supplied by the app; the defaults make the resources usable on their own (no consumption, no provider-specific recognition, plain agent runs).

Constructors

TrainingResources

constructor(trackedAgent: GraphAIAgent<Input, Output>, dataset: TrainSet<Input, InputLabel>, substepPromptExecutor: PromptExecutor, metric: (TrainSetItem<Input, InputLabel>, Output) -> Double, threshold: Double, serializers: DatasetExecutionSerializers<Input, Output, InputLabel>, actionLogTruncation: ActionLogTruncation, retriesPolicy: RetryPolicy, failureAnalyzer: FailureAnalyzer, capturingExecutorFactory: (PromptExecutor) -> CapturingPromptExecutor, consumptionCollector: () -> LLMConsumption? = { null }, runAgentAttempt: suspend (agent: GraphAIAgent<Input, Output>, input: Input) -> Pair<GraphAIAgent<Input, Output>, Output> = { agent, input -> agent to agent.run(input) }, withItemContext: suspend (itemId: String, block: suspend () -> Unit) -> Unit = { _, block -> block() })

Properties

actionLogTruncation

val actionLogTruncation: ActionLogTruncation

Truncation applied to action-log strings recorded during the session.

capturingExecutorFactory

val capturingExecutorFactory: (PromptExecutor) -> CapturingPromptExecutor

Wraps a delegate executor into a consumption-capturing one, created once per leaf prompt call.

consumptionCollector

val consumptionCollector: () -> LLMConsumption?

Returns the consumption accumulated by the tracked agent since the last call (cleared each call).

dataset

val dataset: TrainSet<Input, InputLabel>

The training dataset iterated by the DSL.

failureAnalyzer

val failureAnalyzer: FailureAnalyzer

Recognizes caught exceptions into the project's failure taxonomy.

metric

val metric: (TrainSetItem<Input, InputLabel>, Output) -> Double

Scores an agent's output against its dataset item; higher is better.

retriesPolicy

val retriesPolicy: RetryPolicy

Session-level retry policy applied inside leaf primitives (runAgent / executeWithTrackedPromptExecutor).

runAgentAttempt

val runAgentAttempt: suspend (agent: GraphAIAgent<Input, Output>, input: Input) -> Pair<GraphAIAgent<Input, Output>, Output>

Runs one agent attempt; returns the (possibly re-prepared) agent and its output.

serializers

val serializers: DatasetExecutionSerializers<Input, Output, InputLabel>

Serializers for dataset items and agent outputs used when recording runs.

substepPromptExecutor

val substepPromptExecutor: PromptExecutor

Executor used for the session's leaf prompt calls (executePrompt and variants).

threshold

val threshold: Double

Score at or above which a run counts as solved.

trackedAgent

val trackedAgent: GraphAIAgent<Input, Output>

The agent being trained/evaluated; the default target of runAgent.

withItemContext

val withItemContext: suspend (itemId: String, block: suspend () -> Unit) -> Unit

Establishes any per-item context (e.g. tracing) around a dataset item's processing.