MLModelConfigVersionOutput
An immutable snapshot of a completed ML training run
Fields
Unique version identifier
Parent ML model config ID
Sequential version number (1, 2, 3, ...)
Type of ML model
Possible values (3)
REGRESSIONRegression model (predicts numeric values)
CLASSIFICATIONClassification model (predicts categories)
FORECASTTime-series forecast model (predicts future values over a horizon)
Blob storage key for the trained model
Target column used for training
Feature columns used for training
Hyperparameters used for this training run
Metrics from this training run
Feature importance from this training run
Per-round evaluation history
Training duration in milliseconds
Number of training rows used
How this run's model scored on rows held out from its training, against a baseline and recent versions; null for versions trained before holdout evaluation
Identifies the data source and pipeline stages that produced the training rows; versions with different fingerprints read different data
When this version was created (ISO8601)
Referenced by (3)
Exact type SDL
"""An immutable snapshot of a completed ML training run"""
type MLModelConfigVersionOutput {
"""Unique version identifier"""
id: MLModelConfigVersionID!
"""Parent ML model config ID"""
mlModelConfigId: MLModelConfigID!
"""Sequential version number (1, 2, 3, ...)"""
version: Int!
"""Type of ML model"""
modelType: MLModelType!
"""Blob storage key for the trained model"""
trainedModelID: String!
"""Target column used for training"""
targetColumn: String!
"""Feature columns used for training"""
featureColumns: [String!]!
"""Hyperparameters used for this training run"""
hyperparameters: MLHyperparameters
"""Metrics from this training run"""
trainingMetrics: TrainingMetrics
"""Feature importance from this training run"""
featureImportance: [MLFeatureImportance!]
"""Per-round evaluation history"""
evaluationHistory: [MLEvaluationEntry!]
"""Training duration in milliseconds"""
trainingDurationMs: Int
"""Number of training rows used"""
trainingRowCount: Int
"""
How this run's model scored on rows held out from its training, against a baseline and recent versions; null for versions trained before holdout evaluation
"""
evaluation: MLModelEvaluation
"""
Identifies the data source and pipeline stages that produced the training rows; versions with different fingerprints read different data
"""
pipelineFingerprint: String
"""When this version was created (ISO8601)"""
createdAt: ISO8601!
}