I'm trying to plot a learning curve for the CatBoostClassifier. The error occurs when I fit the CatBoostClassifier into LearningCurve from yellowbrick. I think this should work since CatBoost is sklearn compatible and yellow brick is a sklearn extension.
Code snippet:
kf = RepeatedStratifiedKFold(n_splits=10, n_repeats=3, random_state=0)
sizes = np.linspace(0.2, 1.0, 10)
estimator = CatBoostClassifier(
iterations=42, learning_rate=0.3, max_depth=10)
visualizer = LearningCurve(
estimator, cv=kf, scoring='accuracy', train_sizes=sizes, n_jobs=-1
)
visualizer.fit(X, y)
visualizer.show()
Error:
... yellowbrick.exceptions.YellowbrickTypeError: Cannot detect the model name for non estimator: ''
Any suggestions?
You can use wrapper for third party estimator, more details. I’ve tried and it worked. Something like this:
from yellowbrick.classifier import ROCAUC
from yellowbrick.contrib.wrapper import wrap
catboost_model = CatBoostClassifier()
model = wrap(catboost_model)
visualizer = ROCAUC(model)
visualizer.fit(X_train, y_train)
visualizer.score(X_test, y_test)
visualizer.show()