First time posting here in the community, I hope I did it right. Just wanted to share with you all some interesting scikit-learn tools that I came across whilst completing this project which are not thaught in the relevant DQ missions. Once I was at the random forest model implementation step, I felt I needed to dig deeper into how to optimize all the hyperparameters and googling some relevant articles, I found a very interesting post on Towards data Science by Will Koehrsen. It uses the sklearn RandomizedSearchCV and GridSearchCV methods. The first attempts randomly a number (defined by user) of combinations of hyperparams in ranges given by the user, the second tries all the combinations of hyperparams passed (for fine tuning). Some attributes allow you to call then the best params set and the related estimator. Although I didn’t get a significant increase in accuracy compared to some “manual tuning”, I believe it’s worth to know these methods for future use. A significant con, run times were very long for me with 500 iterations.
Link to Towards data science post
Predicting Bike Rentals.ipynb (72 Bytes)