Why didn’t thy choose the best values to be 170 & 3.5?

In the first optional lab about decision trees, why dudn’t they choose these values for both min sample splits and max depth to be 170 & 3.5 ?

Thank you so much in advance

Hi @manalchetouani ,

As you can see in the first graph, the validation accuracy is more important for us and it remains kind of unchanged after epoch 30. Also, setting min_samples_split to 170 means that each internal node must have at least 170 samples before it can be split. This is quite high and can result in underfitting.

Additionally, Setting max_depth to 3.5 is not valid because depths are counted in whole numbers of levels.

Hope this helps, feel free to ask if you need further assistance!