Hello Sk-Learn team,
At the end of this exercise notebook, it appears that the constant prediction accurracies are 25/75-ish% (if I’m not mistaken).
What I understand is that the train data is composed of 25% people above 50K and 75% under.
Does it have any link with the 25/75% splitting of the train/test data ?
Thank you in advance !
Hello,
You can find the exact fraction of class ">50K" with respect to class "<=50K" in the training set by running
target_train.value_counts()[1]/target_train.value_counts()[0]
after your train-test split.
You can repeat the experiment by setting different values of random_state in the argument of the train_test_split. You can alternatively experiment passing different test sizes, e.g. test_size=10.
You will notice that the proportion of the train-test split has nothing to do with the prediction of the Dummy Classifier in general.
Ok, thank you, it is clear !