Open
Description
Describe the bug
When defining a condition step like in the code snippet below, if one is using a parameter for the evaluation metric (commented out line) to enable switching model evaluation metric at pipeline execution creation, the pipeline upsert fails since it expects the json_path to be valid json.
Should it not be supported to have this evaluation metric be a parameter?
I am already running everything using pipeline sessions, though there does not seem to be a way to specify that in the ConditionStep.
cond_gte = ConditionGreaterThanOrEqualTo(
left=JsonGet(
step_name=step_evaluate_model.name,
property_file=evaluation_report,
# json_path = Join(on=".", values=["binary_classification_metrics", param_eval_metric, "value"]), # this doesnt work! json_path must be json serializable
json_path=f"binary_classification_metrics.{eval_metric}.value"
),
right=param_eval_threshold,
)
step_condition_check = ConditionStep(
name="evaluation-greater-than-threshold-condition",
conditions=[cond_gte],
if_steps=[step_register_model],
else_steps=[step_fail],
)
I am using sagemaker 2.207.