def test_gridsearch_can_log_with_context_manager(self,
feature_union_classifier,
train_iris_dataset,
tmp_path):
classifier = Model(feature_union_classifier)
classifier.config.RUN_DIR = tmp_path
with classifier.log("gridsearch_union_test"):
_, _ = classifier.gridsearch(
train_iris_dataset,
param_grid={"estimator__penalty": ["l1", "l2"]})
def test_gridsearch_uses_default_metric(self, classifier: Model,
train_iris_dataset):
model, results = classifier.gridsearch(
train_iris_dataset, param_grid={"penalty": ["l1", "l2"]})
assert len(results) == 2
assert results[0].metrics.score >= results[1].metrics.score
assert results[0].metrics.name == "accuracy"
assert isinstance(model, Model)
def test_gridsearch_model_does_not_fail_when_run_twice(
self, pipeline_logistic: Pipeline, train_iris_dataset):
model = Model(pipeline_logistic)
best_model, results = model.gridsearch(
train_iris_dataset,
param_grid={"estimator__penalty": ["l1", "l2"]})
assert isinstance(best_model.estimator, Pipeline)
assert 2 == len(results)
for result in results:
assert isinstance(result, Result)
best_model, results = model.gridsearch(
train_iris_dataset,
param_grid={"estimator__penalty": ["l1", "l2"]})
assert isinstance(best_model.estimator, Pipeline)
assert 2 == len(results)
for result in results:
assert isinstance(result, Result)
def test_gridsearch_best_model_is_not_fitted_if_refit_is_not_true(
self, pipeline_logistic: Pipeline, train_iris_dataset: Dataset):
model = Model(pipeline_logistic)
model, results = model.gridsearch(
train_iris_dataset,
param_grid={"estimator__penalty": ["l1", "l2"]},
refit=False,
)
with pytest.raises(MLToolingError,
match="You haven't fitted the estimator"):
model.make_prediction(data=train_iris_dataset, idx=1)
def test_gridsearch_model_returns_as_expected(self,
pipeline_logistic: Pipeline,
train_iris_dataset):
model = Model(pipeline_logistic)
model, results = model.gridsearch(
train_iris_dataset,
param_grid={"estimator__penalty": ["l1", "l2"]})
assert isinstance(model.estimator, Pipeline)
assert 2 == len(results)
for result in results:
assert isinstance(result, Result)
def test_gridsearch_can_take_multiple_metrics(self, classifier: Model,
train_iris_dataset):
model, results = classifier.gridsearch(
train_iris_dataset,
param_grid={"penalty": ["l1", "l2"]},
metrics=["accuracy", "roc_auc"],
)
assert len(results) == 2
assert results[0].metrics.score >= results[1].metrics.score
for result in results:
assert len(result.metrics) == 2
assert "accuracy" in result.metrics
assert "roc_auc" in result.metrics
assert result.metrics.name == "accuracy"
assert result.metrics.score == result.metrics[0].score
