Marijkevandesteene / MachineLearning

repo to share progress and to manage versions of exam MachineLearning (M14)
0 stars 2 forks source link

Hyperparameter - train test split - best percentage #28

Closed Marijkevandesteene closed 4 months ago

Marijkevandesteene commented 4 months ago

What is the optimal percentage for train test split? 20 - 30%?

Learning curve for RF is afbeelding

binomaiheu commented 4 months ago

Veel onder de 80 % voor training set (4000 samples) zou ik niet gaan, misschien 75 % / 25 % inderdaad, want in die 5-fold holdout CV gaan er van de trainset nog samples van af.

Marijkevandesteene commented 4 months ago

Handled in Final Notebook