IExperimentTrainer backed by the TrainingEngine. For each
iteration it resolves the pipeline (via an injected IPipelineResolver),
trains it through the engine (producing an immutable Draft MJ: ML Models +
its MJ: ML Training Runs leaf linked to the iteration via
experimentSessionIterationId), and extracts the normalized holdout Score that
drives the leaderboard: on the plan's SuccessMetric, or on the model's primary
proper score when its holdout gave none on it (ExperimentScores.ExtractNormalizedScore).
session's remaining compute seconds become the run's timeout (the engine takes the smaller of them and the
protocol's compute budget); a session with no seconds bound leaves the protocol's.
Throws
TrainingAttemptFailedError when the attempt failed after its run row existed, carrying the run's compute
IExperimentTrainer backed by the TrainingEngine. For each iteration it resolves the pipeline (via an injected IPipelineResolver), trains it through the engine (producing an immutable Draft
MJ: ML Models+ itsMJ: ML Training Runsleaf linked to the iteration viaexperimentSessionIterationId), and extracts the normalized holdout Score that drives the leaderboard: on the plan'sSuccessMetric, or on the model's primary proper score when its holdout gave none on it (ExperimentScores.ExtractNormalizedScore).