Ejemplo n.º 1
0
def test_log_api_call_count(wandb: mock.MagicMock) -> None:

    wandb.sdk.wandb_run.Run = mock.MagicMock

    study = optuna.create_study()
    wandbc = WeightsAndBiasesCallback()

    @wandbc.track_in_wandb()
    def _decorated_objective(trial: optuna.trial.Trial) -> float:
        result = _objective_func(trial)
        wandb.run.log({"result": result})
        return result

    target_n_trials = 10
    study.optimize(_objective_func,
                   n_trials=target_n_trials,
                   callbacks=[wandbc])
    assert wandb.run.log.call_count == target_n_trials

    wandbc = WeightsAndBiasesCallback(as_multirun=True)
    wandb.run.reset_mock()

    study.optimize(_decorated_objective,
                   n_trials=target_n_trials,
                   callbacks=[wandbc])

    assert wandb.run.log.call_count == 2 * target_n_trials

    wandb.run = None
    study.optimize(_objective_func,
                   n_trials=target_n_trials,
                   callbacks=[wandbc])
    assert wandb.init().log.call_count == target_n_trials
Ejemplo n.º 2
0
def test_run_initialized(wandb: mock.MagicMock) -> None:

    wandb.sdk.wandb_run.Run = mock.MagicMock

    n_trials = 10
    wandb_kwargs = {
        "project": "optuna",
        "group": "summary",
        "job_type": "logging",
        "mode": "offline",
        "tags": ["test-tag"],
    }

    WeightsAndBiasesCallback(metric_name="mse",
                             wandb_kwargs=wandb_kwargs,
                             as_multirun=False)
    wandb.init.assert_called_once_with(project="optuna",
                                       group="summary",
                                       job_type="logging",
                                       mode="offline",
                                       tags=["test-tag"])

    wandbc = WeightsAndBiasesCallback(metric_name="mse",
                                      wandb_kwargs=wandb_kwargs,
                                      as_multirun=True)
    wandb.run = None

    study = optuna.create_study(direction="minimize")
    _wrapped_func = wandbc.track_in_wandb()(lambda t: 1.0)
    wandb.init.reset_mock()
    trial = optuna.create_trial(value=1.0)
    _wrapped_func(trial)

    wandb.init.assert_called_once_with(project="optuna",
                                       group="summary",
                                       job_type="logging",
                                       mode="offline",
                                       tags=["test-tag"])

    wandb.init.reset_mock()
    study.optimize(_objective_func, n_trials=n_trials, callbacks=[wandbc])

    wandb.init.assert_called_with(project="optuna",
                                  group="summary",
                                  job_type="logging",
                                  mode="offline",
                                  tags=["test-tag"])

    assert wandb.init.call_count == n_trials

    wandb.init().finish.assert_called()
    assert wandb.init().finish.call_count == n_trials
Ejemplo n.º 3
0
def test_multiobjective_values_registered_on_epoch_with_logging(
        wandb: mock.MagicMock, metrics: Union[str, Sequence[str]],
        expected: List[str]) -> None:

    wandbc = WeightsAndBiasesCallback(as_multirun=True, metric_name=metrics)

    @wandbc.track_in_wandb()
    def _decorated_objective(trial: optuna.trial.Trial) -> Tuple[float, float]:
        result0, result1 = _multiobjective_func(trial)
        wandb.run.log({"result0": result0, "result1": result1})
        return result0, result1

    study = optuna.create_study(directions=["minimize", "maximize"])
    study.enqueue_trial({"x": 2, "y": 24})
    study.optimize(_decorated_objective, n_trials=1, callbacks=[wandbc])

    logged_in_decorator = wandb.run.log.mock_calls[0][1][0]
    logged_in_callback = wandb.run.log.mock_calls[1][1][0]

    assert len(wandb.run.log.mock_calls) == 2
    assert list(logged_in_decorator) == ["result0", "result1"]
    assert list(logged_in_callback) == expected

    call_args = wandb.run.log.call_args
    assert call_args[1] == {"step": 0}
Ejemplo n.º 4
0
def test_multiobjective_values_registered_on_epoch(
    wandb: mock.MagicMock,
    metrics: Union[str, Sequence[str]],
    as_multirun: bool,
    expected: List[str],
) -> None:
    def assert_call_args(log_func: mock.MagicMock, as_multirun: bool) -> None:
        call_args = log_func.call_args
        assert list(call_args[0][0].keys()) == expected
        assert call_args[1] == {"step": None if as_multirun else 0}

    wandb.sdk.wandb_run.Run = mock.MagicMock

    if as_multirun:
        wandb.run = None
        log_func = wandb.init().log
    else:
        log_func = wandb.run.log

    study = optuna.create_study(directions=["minimize", "maximize"])
    wandbc = WeightsAndBiasesCallback(metric_name=metrics,
                                      as_multirun=as_multirun)

    study.optimize(_multiobjective_func, n_trials=1, callbacks=[wandbc])
    assert_call_args(log_func, as_multirun)
Ejemplo n.º 5
0
def test_values_registered_on_epoch_with_logging(wandb: mock.MagicMock,
                                                 metric: str,
                                                 expected: List[str]) -> None:

    wandb.sdk.wandb_run.Run = mock.MagicMock

    study = optuna.create_study()
    wandbc = WeightsAndBiasesCallback(metric_name=metric, as_multirun=True)

    @wandbc.track_in_wandb()
    def _decorated_objective(trial: optuna.trial.Trial) -> float:
        result = _objective_func(trial)
        wandb.run.log({"result": result})
        return result

    study.enqueue_trial({"x": 2, "y": 25})
    study.optimize(_decorated_objective, n_trials=1, callbacks=[wandbc])

    logged_in_decorator = wandb.run.log.mock_calls[0][1][0]
    logged_in_callback = wandb.run.log.mock_calls[1][1][0]
    assert len(wandb.run.log.mock_calls) == 2
    assert list(logged_in_decorator) == ["result"]
    assert list(logged_in_callback) == expected

    call_args = wandb.run.log.call_args
    assert call_args[1] == {"step": 0}
Ejemplo n.º 6
0
def test_multiobjective_raises_on_name_mismatch(wandb: mock.MagicMock,
                                                metrics: List[str]) -> None:

    wandbc = WeightsAndBiasesCallback(metric_name=metrics)
    study = optuna.create_study(directions=["minimize", "maximize"])

    with pytest.raises(ValueError):
        study.optimize(_multiobjective_func, n_trials=1, callbacks=[wandbc])
Ejemplo n.º 7
0
def test_multiobjective_attributes_set_on_epoch(wandb: mock.MagicMock) -> None:

    wandb.config.update = mock.MagicMock()

    wandbc = WeightsAndBiasesCallback()
    study = optuna.create_study(directions=["minimize", "maximize"])
    study.optimize(_multiobjective_func, n_trials=1, callbacks=[wandbc])

    expected = {"direction": ["MINIMIZE", "MAXIMIZE"]}
    wandb.config.update.assert_called_once_with(expected)
Ejemplo n.º 8
0
def test_log_api_call_count(wandb: mock.Mock) -> None:

    wandb.log = mock.MagicMock()

    wandbc = WeightsAndBiasesCallback()
    target_n_trials = 10
    study = optuna.create_study()
    study.optimize(_objective_func,
                   n_trials=target_n_trials,
                   callbacks=[wandbc])
    assert wandb.log.call_count == target_n_trials
Ejemplo n.º 9
0
def test_values_registered_on_epoch(wandb: mock.Mock) -> None:

    wandb.log = mock.MagicMock()

    wandbc = WeightsAndBiasesCallback()
    study = optuna.create_study()
    study.optimize(_objective_func, n_trials=1, callbacks=[wandbc])

    kall = wandb.log.call_args
    assert list(kall[0][0].keys()) == ["x", "y", "value"]
    assert kall[1] == {"step": 0}
Ejemplo n.º 10
0
def test_values_registered_on_epoch(wandb: mock.Mock, metric: str,
                                    expected: List[str]) -> None:

    wandb.log = mock.MagicMock()

    wandbc = WeightsAndBiasesCallback(metric_name=metric)
    study = optuna.create_study()
    study.optimize(_objective_func, n_trials=1, callbacks=[wandbc])

    kall = wandb.log.call_args
    assert list(kall[0][0].keys()) == expected
    assert kall[1] == {"step": 0}
Ejemplo n.º 11
0
def test_multiobjective_values_registered_on_epoch(
        wandb: mock.Mock, metrics: Union[str, Sequence[str]],
        expected: List[str]) -> None:

    wandb.log = mock.MagicMock()

    wandbc = WeightsAndBiasesCallback(metric_name=metrics)
    study = optuna.create_study(directions=["minimize", "maximize"])
    study.optimize(_multiobjective_func, n_trials=1, callbacks=[wandbc])

    kall = wandb.log.call_args
    assert list(kall[0][0].keys()) == expected
    assert kall[1] == {"step": 0}
Ejemplo n.º 12
0
def test_run_initialized(wandb: mock.MagicMock) -> None:

    wandb_kwargs = {
        "project": "optuna",
        "group": "summary",
        "job_type": "logging",
        "mode": "offline",
    }

    WeightsAndBiasesCallback(
        metric_name="mse",
        wandb_kwargs=wandb_kwargs,
    )

    wandb.init.assert_called_once_with(
        project="optuna",
        group="summary",
        job_type="logging",
        mode="offline",
    )
Ejemplo n.º 13
0
def test_attributes_set_on_epoch(wandb: mock.MagicMock,
                                 as_multirun: bool) -> None:

    wandb.sdk.wandb_run.Run = mock.MagicMock

    expected_config: Dict[str, Any] = {"direction": ["MINIMIZE"]}
    trial_params = {"x": 1.1, "y": 2.2}
    expected_config_with_params = {**expected_config, **trial_params}

    study = optuna.create_study(direction="minimize")
    wandbc = WeightsAndBiasesCallback(as_multirun=as_multirun)

    if as_multirun:
        wandb.run = None

    study.enqueue_trial(trial_params)
    study.optimize(_objective_func, n_trials=1, callbacks=[wandbc])

    if as_multirun:
        wandb.init().config.update.assert_called_once_with(
            expected_config_with_params)
    else:
        wandb.run.config.update.assert_called_once_with(expected_config)
Ejemplo n.º 14
0
def test_multiobjective_raises_on_type_mismatch(metrics: Any) -> None:

    with pytest.raises(TypeError):
        WeightsAndBiasesCallback(metric_name=metrics)