For many machine learning problems, simply running a model out-of-the-box and getting a prediction is not enough; you want the best model with the most accurate prediction. One way to perfect your model is with hyperparameter tuning, which means optimizing the settings for that specific model. In this course, you will work with the caret, mlr and h2o packages to find the optimal combination of hyperparameters in an efficient manner using grid search, random search, adaptive resampling and automatic machine learning (AutoML). Furthermore, you will work with different datasets and tune different supervised learning models, such as random forests, gradient boosting machines, support vector machines, and even neural nets. Get ready to tune!
Introduction to hyperparametersFree
Why do we use the strange word "hyperparameter"? What makes it hyper? Here, you will understand what model parameters are, and why they are different from hyperparameters in machine learning. You will then see why we would want to tune them and how the default setting of caret automatically includes hyperparameter tuning.
Hyperparameter tuning with caret
In this chapter, you will learn how to tune hyperparameters with a Cartesian grid. Then, you will implement faster and more efficient approaches. You will use Random Search and adaptive resampling to tune the parameter grid, in a way that concentrates on values in the neighborhood of the optimal settings.
Hyperparameter tuning with mlr
Here, you will use another package for machine learning that has very convenient hyperparameter tuning functions. You will define a Cartesian grid or perform Random Search, as well as advanced techniques. You will also learn different ways to plot and evaluate models with different hyperparameters.
Hyperparameter tuning with h2o
In this final chapter, you will use h2o, another package for machine learning with very convenient hyperparameter tuning functions. You will use it to train different models and define a Cartesian grid. Then, You will implement a Random Search use stopping criteria. Finally, you will learn AutoML, an h2o interface which allows for very fast and convenient model and hyperparameter tuning with just one function.
Shirin Elsinghorst (formerly Glander)
Data Scientist @ codecentric
I’m Shirin, a biologist turned bioinformatician turned data scientist. During my PhD and Postdoc I worked with Next Generation Sequencing data to analyze diseases like arthritis. However, I then chose to become a data scientist for a German IT company, called codecentric. In this capacity, I have been working on many different projects, e.g. building fraud detection models, creating a chatbot, implementing predictive maintenance, and more. My tool of choice for data analysis so far has been R but I also work with Python. I am also very passionate about teaching and sharing knowledge, so I give workshops, talk at conferences or meetups, write blog posts and organize the MünsteR R-users group.