Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Fix #168. Enforce float32 type for split condition values for GBT models created using XGBoost #188

Merged
merged 3 commits into from
Apr 5, 2020
Merged

Fix #168. Enforce float32 type for split condition values for GBT models created using XGBoost #188

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
e13bea3
Fix #168. Enforce float32 type for split condition values for GBT mod…
izeigerman Mar 30, 2020
242165e
Parametrize NumVal with type
izeigerman Apr 3, 2020
645a49d
Fix linter error. Add a test for ast.NumVal
izeigerman Apr 4, 2020
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Jump to
Jump to file
Failed to load files.
Diff view
Diff view
2 changes: 1 addition & 1 deletion m2cgen/assemblers/boosting.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -134,7 +134,7 @@ def _assemble_tree(self, tree):
if "leaf" in tree:
return ast.NumVal(tree["leaf"])

threshold = ast.NumVal(tree["split_condition"])
threshold = ast.NumVal(np.float32(tree["split_condition"]))
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Maybe a more general solution will be to add an optional dtype constructor argument? I mean,

class NumVal(NumExpr):
    def __init__(self, value, dtype=np.float64):

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

That's a good idea 👍

split = tree["split"]
feature_idx = self._feature_name_to_idx.get(split, split)
feature_ref = ast.FeatureRef(feature_idx)
Expand Down
35 changes: 23 additions & 12 deletions tests/e2e/test_e2e.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -35,50 +35,51 @@


# Set of helper functions to make parametrization less verbose.
def regression(model):
def regression(model, test_fraction=0.02):
return (
model,
utils.get_regression_model_trainer(),
utils.get_regression_model_trainer(test_fraction),
REGRESSION,
)


def classification(model):
def classification(model, test_fraction=0.02):
return (
model,
utils.get_classification_model_trainer(),
utils.get_classification_model_trainer(test_fraction),
CLASSIFICATION,
)


def classification_binary(model):
def classification_binary(model, test_fraction=0.02):
return (
model,
utils.get_binary_classification_model_trainer(),
utils.get_binary_classification_model_trainer(test_fraction),
CLASSIFICATION,
)


def regression_random(model):
def regression_random(model, test_fraction=0.02):
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Does test_fraction increase for random datasets allow to reproduce the original issue?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes, unfortunately the default fraction produced way too few samples to be able to reproduce the issue reliably.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Got it! We definitely need to refactor testing routines to be more tunable, e.g. allow to adjust test_fraction according to programming language. Refer to #114 (comment).

return (
model,
utils.get_regression_random_data_model_trainer(0.01),
utils.get_regression_random_data_model_trainer(test_fraction),
REGRESSION,
)


def classification_random(model):
def classification_random(model, test_fraction=0.02):
return (
model,
utils.get_classification_random_data_model_trainer(0.01),
utils.get_classification_random_data_model_trainer(test_fraction),
CLASSIFICATION,
)


def classification_binary_random(model):
def classification_binary_random(model, test_fraction=0.02):
return (
model,
utils.get_classification_binary_random_data_model_trainer(0.01),
utils.get_classification_binary_random_data_model_trainer(
test_fraction),
CLASSIFICATION,
)

Expand All @@ -92,6 +93,8 @@ def classification_binary_random(model):
FOREST_PARAMS = dict(n_estimators=10, random_state=RANDOM_SEED)
XGBOOST_PARAMS = dict(base_score=0.6, n_estimators=10,
random_state=RANDOM_SEED)
XGBOOST_HIST_PARAMS = dict(base_score=0.6, n_estimators=10,
tree_method="hist", random_state=RANDOM_SEED)
XGBOOST_PARAMS_LINEAR = dict(base_score=0.6, n_estimators=10,
feature_selector="shuffle", booster="gblinear",
random_state=RANDOM_SEED)
Expand Down Expand Up @@ -170,6 +173,14 @@ def classification_binary_random(model):
classification(xgboost.XGBClassifier(**XGBOOST_PARAMS)),
classification_binary(xgboost.XGBClassifier(**XGBOOST_PARAMS)),

# XGBoost (tree method "hist")
regression(xgboost.XGBRegressor(**XGBOOST_HIST_PARAMS),
test_fraction=0.2),
classification(xgboost.XGBClassifier(**XGBOOST_HIST_PARAMS),
test_fraction=0.2),
classification_binary(xgboost.XGBClassifier(**XGBOOST_HIST_PARAMS),
test_fraction=0.2),

# XGBoost (LINEAR)
regression(xgboost.XGBRegressor(**XGBOOST_PARAMS_LINEAR)),
classification(xgboost.XGBClassifier(**XGBOOST_PARAMS_LINEAR)),
Expand Down