Imagine that you are working on your first supervised learning project. You are suppose to do a regression or a classification. How do you start predicting/classifying?
The classical way of dealing with the above issue is to exhaust as many models of regression/classification until you have found a match. A model or a set of models that can predict labels or quantities with high accuracy. But anyone who has gone through this would know how time consuming it is and what about cognitive dissonance? Those feelings of inadequacy : Have I exhausted enough methods? Is there a better model that I have not experimented with?
The struggle is over. Lazy predict is now here. This is an amazing python package built to evaluate close to 30 sklearn classifiers and regressors for a given dataset. It is simple to use and uses the same line of codes taken to run a particular classifier or regressor model.
Thus now one can select models effectively and concentrate on the best methods that works for the dataset without having to worry about the time nor the insufficiency of the models explored. I think this is amazing for anyone who wants to run a quick analysis on their dataset.
Unfortunately we only have lazy predict for supervised data at the moment. Maybe there will be more developments later. Another disadvantage of it is the inbuilt feature selection and data curation. The user does not have a say in the methods that lazy predict uses to curate data and for feature selection. However, the code is available for the public in github and anyone who wants to customize it may do so. Nevertheless it is still one of the most amazing tools we have at the moment for supervised problems.
I hope anyone who reads this gives it a try!
Comments