Learn to avoid the six most serious mistakes related to machine learning theory that beginners often make through Sklearn.
Image by me with Leonardo AI
Often, Sklearn throws big red error messages and warnings when you make a mistake. These messages suggest something is seriously wrong with your code, preventing the Sklearn magic from doing its job.
But what happens if you don’t get any errors or warnings? Does that mean you are crushing it so far? Not necessarily. Many knobs and dials make Sklearn the greatest ML library, its world-class code design being an example.
The mistakes while writing Sklearn code can easily be fixed. What can go unnoticed is the mistakes related to the internal logic and ML theory that powers Sklearn algorithms and transformers.
These mistakes are especially more common and subtle when you are a beginner. So this post will be about the six such mistakes I made and learned to avoid when I was a beginner myself.
. Using fit or fit_transform everywhere
Let’s start with the most serious mistake — a mistake that is related to data leakage. Data leakage is subtle and can be destructive to model performance.
It occurs when information that would not be available at prediction time is used during the model training. Data leakage causes models to give very optimistic results, even in cross-validation, but perform terribly when testing on actual novel data.