def test_persist_experts_in_physical_partition(train_data, grid_search, tmp_path):
partition_columns = ["Product"]
results = select_model(
train_data,
target_col_name="Quantity",
partition_columns=partition_columns,
grid_search=grid_search,
)
persist_experts_in_physical_partition(
tmp_path,
results,
persist_cv_results=True,
persist_cv_data=True,
persist_model_reprs=True,
persist_best_model=True,
persist_partition=True,
persist_model_selector_results=True,
)
files = os.listdir(tmp_path)
assert len(files) == 18
for result in results:
with open(os.path.join(tmp_path, result.partition_hash + ".cv_results"), "rb") as file:
cv_results = pickle.load(file)
assert isinstance(cv_results, type(result.cv_results))
assert all(cv_results.columns == result.cv_results.columns)
with open(os.path.join(tmp_path, result.partition_hash + ".cv_data"), "rb") as file:
cv_data = pickle.load(file)
assert isinstance(cv_data, type(result.cv_data))
assert all(cv_data.columns == result.cv_data.columns)
with open(os.path.join(tmp_path, result.partition_hash + ".model_reprs")) as file:
model_reprs = json.load(file)
assert isinstance(model_reprs, type(result.model_reprs))
assert model_reprs == result.model_reprs
with open(os.path.join(tmp_path, result.partition_hash + ".best_model"), "rb") as file:
model = pickle.load(file)
assert isinstance(model, type(result.best_model))
assert str(model.get_params()) == str(result.best_model.get_params())
with open(os.path.join(tmp_path, result.partition_hash + ".partition")) as file:
partition = json.load(file)
assert partition == result.partition
with open(
os.path.join(tmp_path, result.partition_hash + ".model_selector_result"),
"rb",
) as file:
model_selector_result = pickle.load(file)
assert isinstance(model_selector_result, type(result))
assert str(model_selector_result.__dict__) == str(result.__dict__)
def test_load_expert(train_data, grid_search, tmp_path):
partition_columns = ["Product"]
results = select_model(
train_data,
target_col_name="Quantity",
partition_columns=partition_columns,
grid_search=grid_search,
)
persist_experts_in_physical_partition(
tmp_path,
results,
persist_cv_results=True,
persist_cv_data=True,
persist_model_reprs=True,
persist_best_model=True,
persist_partition=True,
persist_model_selector_results=True,
)
files = os.listdir(tmp_path)
assert len(files) == 18
for result in results:
cv_results = _load_file(partition_label=result.partition,
path=tmp_path,
expert_type="cv_results")
assert isinstance(cv_results, type(result.cv_results))
assert all(cv_results.columns == result.cv_results.columns)
cv_data = _load_file(partition_label=result.partition,
path=tmp_path,
expert_type="cv_data")
assert isinstance(cv_data, type(result.cv_data))
assert all(cv_data.columns == result.cv_data.columns)
model_reprs = _load_file(partition_label=result.partition,
path=tmp_path,
expert_type="model_reprs")
assert isinstance(model_reprs, type(result.model_reprs))
assert model_reprs == result.model_reprs
pkl_model = _load_file(partition_label=result.partition,
path=tmp_path,
expert_type="best_model")
assert isinstance(pkl_model, type(result.best_model))
assert str(pkl_model.get_params()) == str(
result.best_model.get_params())
partition = _load_file(partition_label=result.partition,
path=tmp_path,
expert_type="partition")
assert isinstance(partition, type(result.partition))
assert partition == result.partition
model_selector_result = _load_file(
partition_label=result.partition,
path=tmp_path,
expert_type="model_selector_result",
)
assert isinstance(model_selector_result, type(result))
assert str(model_selector_result.__dict__) == str(result.__dict__)
# with partition_hash
pkl_model = _load_file(
partition_hash=result.partition_hash,
path=tmp_path,
expert_type="best_model",
)
assert isinstance(pkl_model, type(result.best_model))
assert str(pkl_model.get_params()) == str(
result.best_model.get_params())
with pytest.raises(ValueError):
_load_file(
partition_label=result.partition,
partition_hash=result.partition_hash,
path=tmp_path,
expert_type="best_model",
)
with pytest.raises(ValueError):
_load_file(path=tmp_path, expert_type="best_model")
