validate

Deployment gate. Check a model against performance rules and optionally compare to a baseline model. Returns pass/fail with reasons.

Signature

ml.validate(model, *, test, rules=None, baseline=None, tolerance=0.0)

ml_validate(model, test, rules = NULL, baseline = NULL, tolerance = 0.0)

Parameters

Parameter	Type	Default	Description
`model`	Model	—	Model to validate
`test`	DataFrame	—	Test data
`rules`	dict \| None	`None`	Performance thresholds, e.g. `{"accuracy": ">0.85"}`
`baseline`	Model \| None	`None`	Previous model to check for regressions
`tolerance`	float	`0.0`	Allowed degradation from baseline

Returns

ValidateResult with:

.passed — True if all rules pass and no regressions
.failures — list of failed rules
.improvements / .degradations — vs baseline

Examples

With rules

result = ml.validate(
    model,
    test=s.test,
    rules={"accuracy": ">0.80", "roc_auc": ">=0.85"},
)
print(result.passed)   # True or False
print(result.failures) # [] or ["accuracy: 0.78 < 0.80"]

result <- ml_validate(
  model, s$test,
  rules = list(accuracy = ">0.80", roc_auc = ">=0.85")
)
result$passed
result$failures

Against a baseline

result = ml.validate(
    new_model,
    test=s.test,
    baseline=old_model,
)

result <- ml_validate(new_model, s$test, baseline = old_model)

drift →