Interpret the Leaderboard¶
The Leaderboard provides useful summary information for each model built in a project and ranks them based on the chosen optimization metric—meaning the best performing models are at the top. From the Leaderboard, you can also access a variety of insight tabs to further fine-tune and evaluate your models.
To access the Leaderboard, click Models > Leaderboard at the top of the page.
This tutorial explains:
- How to read the following model summary information on the Leadboard:
- Performance metrics
- Model badges, icons, and indicators
- Recommended models
- Feature list and sample size
- How to filter the Leaderboard
Compare performance metrics¶
Models are ranked by the optimization metric chosen prior to model building, which is displayed at the top. The Leaderboard displays the model's Validation, Cross Validation, and Holdout (if unlocked) scores.
To change the optimization metric used for ranking, click the Metric dropdown and select a new metric.
Understand model icons¶
The Leaderboard provides a wealth of information for each model built in a project using various badges, tags, and indicators. The badge to the left of the model name indicates the type, and the text below describes model type and version, or whether it uses unaltered open source code.
The tags and indicators along the bottom provide quick model identifying and scoring information.
The model at the top of the list features a Recommended for deployment tag, meaning DataRobot recommends and prepares the model for deployment based on its accuracy and complexity.
Before selecting the recommended model, compare it against other models on the leaderboard using the model comparison tools to make sure it's the best model for your use case.
Starring models on the Leaderboard allows you to quickly find certain models later on.
Hover over the model you would like to star.
Click the star icon. When the star is filled in, that means you've successfully marked this model as a favorite.
Read feature list and sample size¶
The Feature List & Sample Size column displays the feature list and sample size used to build the model and allows you to retrain the model using different parameters.
Filter the Leaderboard¶
If you're looking for a specific model or model criteria, you can filter the Leaderboard to narrow down the results.
By optimiziation metric¶
To sort by Validation, Cross Validation, or Holdout scores, click the column header. When the header is blue, the Leaderboard lists models from most accurate to least accurate for the selected partition.
By starred models¶
To view all starred models, click Filter Models and select Starred Models. The Leaderboard updates to only display models that have been starred.
By sample size and feature list¶
You can also filter Leaderboard models by the feature list and sample size used to build them.
- Click the Feature List & Sample Size column header.
Select the feature list and sample size parameters of the models you would like the Leaderboard to display.
- Leaderboard reference page
- Model recommendation process
- Optimization metric details
- Create a feature list from the Data page
- Compare Leaderboard models
- Model insight tabs