TrialSpec is the specification of a Trial.
Name | Type | Description | Notes |
---|---|---|---|
early_stopping_rules | list[V1beta1EarlyStoppingRule] | Rules for early stopping techniques. Each rule should be met to early stop Trial. | [optional] |
failure_condition | str | Condition when trial custom resource is failed. Condition must be in GJSON format, ref https://github.com/tidwall/gjson. For example for BatchJob: status.conditions.#(type=="Failed")# | #(status=="True")# |
labels | dict(str, str) | Labels that provide additional metadata for services (e.g. Suggestions tracking) | [optional] |
metrics_collector | V1beta1MetricsCollectorSpec | [optional] | |
objective | V1beta1ObjectiveSpec | [optional] | |
parameter_assignments | list[V1beta1ParameterAssignment] | Key-value pairs for hyperparameters and assignment values. | [optional] |
primary_container_name | str | Name of training container where actual model training is running | [optional] |
primary_pod_labels | dict(str, str) | Label that determines if pod needs to be injected by Katib sidecar container | [optional] |
retain_run | bool | Whether to retain the trial run object after completed. | [optional] |
run_spec | object | [optional] | |
success_condition | str | Condition when trial custom resource is succeeded. Condition must be in GJSON format, ref https://github.com/tidwall/gjson. For example for BatchJob: status.conditions.#(type=="Complete")# | #(status=="True")# |