Resource
ml_model_config
ML model configuration for training pipelines
Configuration
sourceml_model_documentRequiredMachine learning model document, including its data pipeline.
ml_model_document fields
ML configuration. Training data and operational state are not part of this document.
autoPublishOptionalbackendOptional- Pattern
"^([xX][gG][bB][oO][oO][sS][tT]|[aA][uU][gG][uU][rR][sS])$"
dataQuerySQLOptionaldescriptionOptionalformatRequired- Allowed values
["1"]
nameRequired- Pattern
"^[a-z][a-z0-9_-]{0,63}$"
specOptionaltabularOptionaltimeSeriesOptionalOnly the fields listed here are accepted.
All of these apply
Choose exactly one shape
specRequiredtabularRequiredspecRequiredtimeSeriesRequiredOnly the fields listed here are accepted.
publishedVersionIdMLModelConfigVersionIDOptionalID of the version currently used for predictions
written by `publishMLModelVersion`, not by this resource's update
Read-only values
Show values
authoringRevisionInt!Read onlyMonotonic source/configuration revision used to reject stale authoring updates; independent of trained model versions.
computed by the platform
idMLModelConfigID!Read onlyUnique identifier
computed by the platform
tenantIdTenantID!Read onlyTenant this config belongs to
computed by the platform
nameString!Read onlyConfig name
computed by the platform
descriptionStringRead onlyConfig description
computed by the platform
modelTypeMLModelType!Read onlyType of ML model
Allowed values
REGRESSION — Regression model (predicts numeric values)
CLASSIFICATION — Classification model (predicts categories)
FORECAST — Time-series forecast model (predicts future values over a horizon)
computed by the platform
backendMLBackend!Read onlyEngine that trains/serves this model
Allowed values
XGBOOST — XGBoost gradient-boosted trees (tabular regression/classification)
AUGURS — Augurs time-series engine (forecasting)
computed by the platform
dataQuerySQLStringRead onlyMicroSQL query for data retrieval. Each SQL page has a 30-second statement timeout; preview sampling does not limit the full training query.
computed by the platform
specMLModelSpec!Read onlyConcept-specific configuration (tabular or time-series)
computed by the platform
statusMLModelTrainingStatus!Read onlyStatus of the latest training run. A published version remains available for serving while retraining or after a failed run.
Allowed values
IDLE — Created but never trained
TRAINING — Currently training
TRAINED — Successfully trained and ready for predictions
FAILED — Last training attempt failed
computed by the platform
trainedModelIDStringRead onlyBlob storage key for the trained model
computed by the platform
lastTrainedAtStringRead onlyWhen the model was last trained (ISO8601)
computed by the platform
trainingDurationMsIntRead onlyDuration of last training run in milliseconds
computed by the platform
trainingMetricsTrainingMetricsRead onlyMetrics from last training run
computed by the platform
trainingErrorStringRead onlyError message if last training failed; SQL source pages have a 30-second statement timeout
computed by the platform
featureImportance[MLFeatureImportance!]Read onlyFeature importance from trained model
computed by the platform
evaluationHistory[MLEvaluationEntry!]Read onlyPer-round evaluation history
computed by the platform
csvFilenameStringRead onlyOriginal CSV filename (if CSV data source)
computed by the platform
csvSizeBytesIntRead onlySize of uploaded CSV in bytes
computed by the platform
hasTrainingCSVBoolean!Read onlyWhether a CSV file has been uploaded as training data
computed by the platform
csvPreviewRows[[String!]!]Read onlyFirst few rows of CSV data for preview (header + up to 10 rows)
computed by the platform
inferenceSchemaStringRead onlyJSON Schema describing the model's input features (types, enum values). Generated after training.
computed by the platform
emittedSchema[MLEmittedColumn!]Read onlyThe pipeline's emitted columns + types (cached from the last preview of the saved config)
computed by the platform
publishedVersionNumberIntRead onlyVersion number of the published version (e.g. 1, 2, 3)
computed by the platform
autoPublishBoolean!Read onlyWhether new versions are automatically published after training
computed by the platform
isSystemBoolean!Read onlyWhether this model is system-managed (auto-created and auto-retrained by the platform)
computed by the platform
createdAtString!Read onlyCreation timestamp (ISO8601)
computed by the platform
updatedAtString!Read onlyLast update timestamp (ISO8601)
computed by the platform
Lifecycle
The CLI can create this resource. The platform assigns its id.
When replacement is needed, the existing resource is removed before its replacement is created.