testing

run_model_tests(model, test_data, predicate, *, path=None, show_progress=False, stop_after=1)

Test a model with the given test data and predicate.

This function runs the model on the test data and checks if the predictions satisfy the given predicate. If any prediction does not satisfy the predicate, a TestFailed exception is raised. This exception contains the input data and prediction that failed the predicate and can be used as a counterexample. This method relies on the model's predict method to obtain a prediction. It does not utilize the model's type or internal structure to prove predicates.

Parameters:

Name	Type	Description	Default
model	Model	The model to test.	required
test_data	IncrementalEnvironment	The test data to use for testing the model. This must only include input features passed to the model and not the targets.	required
predicate	Predicate	The predicate used to check the model's predictions.	required
path	str | Path | None	Optional output file for failed test details. If provided, failures are written to this file instead of raising TestFailed.	None
show_progress	bool	Whether to show progress during testing. Defaults to False.	False
stop_after	int	Number of tests that need to fail before stopping. Defaults to 1. If set to 0 or negative, all tests are run regardless of failures.	1

Raises:

Type	Description
TestFailed	If the model's prediction does not satisfy the predicate and path is not provided.

Source code in src/flowcean/testing/test.py

def run_model_tests(
    model: Model,
    test_data: IncrementalEnvironment,
    predicate: Predicate,
    *,
    path: str | Path | None = None,
    show_progress: bool = False,
    stop_after: int = 1,
) -> None:
    """Test a model with the given test data and predicate.

    This function runs the model on the test data and checks if the
    predictions satisfy the given predicate. If any prediction does not
    satisfy the predicate, a TestFailed exception is raised.
    This exception contains the input data and prediction that failed the
    predicate and can be used as a counterexample.
    This method relies on the model's predict method to obtain a prediction.
    It does not utilize the model's type or internal structure to prove
    predicates.

    Args:
        model: The model to test.
        test_data: The test data to use for testing the model. This must only
            include input features passed to the model and *not* the targets.
        predicate: The predicate used to check the model's predictions.
        path: Optional output file for failed test details. If provided,
            failures are written to this file instead of raising
            ``TestFailed``.
        show_progress: Whether to show progress during testing.
            Defaults to False.
        stop_after: Number of tests that need to fail before stopping. Defaults
            to 1. If set to 0 or negative, all tests are run regardless of
            failures.

    Raises:
        TestFailed: If the model's prediction does not satisfy the
            predicate and ``path`` is not provided.
    """
    number_of_failures = 0
    failure_data: list[Data] = []
    failure_prediction: list[Data] = []
    # Run the model on the test data
    for input_data in (
        tqdm.tqdm(
            test_data,
            "Testing Model",
            total=test_data.num_steps(),
        )
        if show_progress
        else test_data
    ):
        prediction = model.predict(input_data)

        # Handle dataframes and lazyframes separately
        # Those may contain multiple rows / samples and need to be
        # sliced to get the individual samples for testing
        if isinstance(prediction, pl.LazyFrame | pl.DataFrame) and isinstance(
            input_data,
            pl.LazyFrame | pl.DataFrame,
        ):
            input_data_collected = input_data.lazy().collect()
            prediction = prediction.lazy().collect()

            test_inputs = [
                input_data_collected.slice(i, 1)
                for i in range(len(input_data_collected))
            ]
            predictions = [
                prediction.slice(i, 1) for i in range(len(prediction))
            ]

        else:
            test_inputs = [input_data]
            predictions = [prediction]

        # Check if the prediction satisfies the predicate
        for test_input, prediction in zip(
            test_inputs,
            predictions,
            strict=True,
        ):
            if not predicate(
                test_input,
                prediction,
            ):
                number_of_failures += 1
                failure_data.append(test_input)
                failure_prediction.append(prediction)

                if number_of_failures >= stop_after > 0:
                    break
        else:
            continue
        break

    # If we got any failures at this point, raise an exception or write
    # details to disk, depending on the selected output mode.
    if number_of_failures > 0:
        test_failed = TestFailed(failure_data, failure_prediction)
        if path is None:
            raise test_failed
        logger.info(
            " %s tests failed. Writing details to %s.",
            number_of_failures,
            path,
        )
        test_failed.to_file(path)
    elif number_of_failures == 0:
        logger.info("All tests passed.")
        if path is not None:
            output_path = Path(path)
            output_path.parent.mkdir(parents=True, exist_ok=True)
            output_path.write_text("All tests passed.", encoding="utf-8")