Model Tuning

What is Model Tuning?

All machine learning algorithms have a “default” set of hyperparameters, which Machine Learning Mastery defines as “configuration variables that are internal to the model and whose value can be estimated from data.” Different algorithms consist of different hyperparameters. For example, regularized regression models have coefficients penalties, decision trees have a set number of branches, and neural networks have a set number of layers. When building models, analysts and data scientists choose the default configuration of these hyperparameters after running the model on several datasets.

While the generic set of hyperparameters for each algorithm provides a starting point for analysis and will generally result in a well-performing model, it may not have the optimal configurations for your particular dataset and business problem. In order to find the best hyperparameters for your data, you need to tune them.

Tuning is usually a trial-and-error process by which you change some hyperparameters (for example, the number of trees in a tree-based algorithm or the value of alpha in a linear algorithm), run the algorithm on the data again, then compare its performance on your validation set in order to determine which set of hyperparameters results in the most accurate model.

Why is Model Tuning important?

Tuning allows you to customize your models so they generate the most accurate outcomes and give you highly valuable insights into your data, enabling you to make the most effective business decisions.

Model Tuning + DataRobot

DataRobot has several features that help tune your model without requiring extensive manual tweaking and coding. Once you choose your target variable, DataRobot runs a bracket-style competition of algorithms on your dataset and searches through a range of values for each hyperparameter. When the algorithm has finished running, you can select the model and visualize the results of the grid search in the Advanced Tuning section. If you want to try another set of hyperparameters to see if you can beat the initial level of accuracy, you can change the value and immediately run a new model. When it’s done, it will appear on the DataRobot Leaderboard and you can compare its performance with the original model.


However, the DataRobot platform doesn’t require you to do any manual model tuning. It automatically runs dozens of models with preset hyperparameters that our data scientists have thoroughly tested to make sure they result in highly accurate models, so you can focus on choosing the one that is most accurate for your data. At the same time, if you want to manually tune your models, DataRobot makes it easy to do so.