"""Data check to check if any set features are likely to be multicollinear."""
from evalml.data_checks import (
DataCheck,
DataCheckMessageCode,
DataCheckWarning,
)
from evalml.utils import infer_feature_types
[docs]class MulticollinearityDataCheck(DataCheck):
"""Check if any set features are likely to be multicollinear.
Args:
threshold (float): The threshold to be considered. Defaults to 0.9.
"""
def __init__(self, threshold=0.9):
if threshold < 0 or threshold > 1:
raise ValueError("threshold must be a float between 0 and 1, inclusive.")
self.threshold = threshold
[docs] def validate(self, X, y=None):
"""Check if any set of features are likely to be multicollinear.
Args:
X (pd.DataFrame): The input features to check.
y (pd.Series): The target. Ignored.
Returns:
dict: dict with a DataCheckWarning if there are any potentially multicollinear columns.
Example:
>>> import pandas as pd
>>> col = pd.Series([1, 0, 2, 3, 4])
>>> X = pd.DataFrame({"col_1": col, "col_2": col * 3})
>>> y = pd.Series([1, 0, 0, 1, 0])
>>> multicollinearity_check = MulticollinearityDataCheck(threshold=0.8)
>>> assert multicollinearity_check.validate(X, y) == {
... "errors": [],
... "warnings": [{'message': "Columns are likely to be correlated: [('col_1', 'col_2')]",
... "data_check_name": "MulticollinearityDataCheck",
... "level": "warning",
... "code": "IS_MULTICOLLINEAR",
... 'details': {'columns': [('col_1', 'col_2')]}}],
... "actions": []}
"""
results = {"warnings": [], "errors": [], "actions": []}
X = infer_feature_types(X)
mutual_info_df = X.ww.mutual_information()
if mutual_info_df.empty:
return results
above_threshold = mutual_info_df.loc[
mutual_info_df["mutual_info"] >= self.threshold
]
correlated_cols = [
(col_1, col_2)
for col_1, col_2 in zip(
above_threshold["column_1"], above_threshold["column_2"]
)
]
if correlated_cols:
warning_msg = "Columns are likely to be correlated: {}"
results["warnings"].append(
DataCheckWarning(
message=warning_msg.format(correlated_cols),
data_check_name=self.name,
message_code=DataCheckMessageCode.IS_MULTICOLLINEAR,
details={"columns": correlated_cols},
).to_dict()
)
return results