Does anyone know why you get different results when you split the dataset by yourself and when you split it with train_test_split?
splitting with the train_test_split: tr
