Model Recommendation

During the Autopilot modeling process, DataRobot will recommend up to three well-performing models.

Warning

Model recommendations are only generated when you run full Autopilot.

One of them (the most accurate individual, non-blender model) will be prepared for deployment. In the preparation process, DataRobot:

  1. Calculates feature impact for the selected model and uses it to generate a reduced feature list.
  2. Retrains the selected model on the reduced feature list. If the new model performs better than the original model, DataRobot uses the new model for the next stage. Otherwise, the original model is used.
  3. Retrains the selected model on a higher sample size. If the new model performs better than the original model, DataRobot selects it as Recommended for Deployment. Otherwise, the original model is selected.

Note

The higher sample size DataRobot uses in Step 3 is either:

  1. Up to holdout if the training sample size does not exceed the maximum Autopilot size threshold: sample size is the training set plus the validation set (for TVH) or 5-folds (for CV). In this case, DataRobot compares retrained and original models on the holdout score.
  2. Up to validation if the training sample size does exceed the maximum Autopilot size threshold: sample size is the training set (for TVH) or 4-folds (for CV). In this case, DataRobot compares retrained and original models on the validation score.

The four types of recommendations are the following:

  • Recommended for Deployment. This is the most accurate individual, non-blender model on the Leaderboard. This model is ready for deployment.
  • Recommended (deprecated, will be removed in 2.14). This is the most accurate individual, non-blender model on the Leaderboard.
  • Most Accurate. Based on the validation or cross-validation results, this model is the most accurate model overall on the Leaderboard (in most cases, a blender).
  • Fast & Accurate. This is the most accurate individual model on the Leaderboard that passes a set prediction speed guidelines. If no models meet the guideline, the badge is not applied.

Retrieve all recommendations

The following code will return all models recommended for the project.

import datarobot as dr

recommendations = dr.ModelRecommendation.get_all(project_id)

Retrieve a default recommendation

If you are unsure about the tradeoffs between the various types of recommendations, DataRobot can make this choice for you. The following route will return the Recommended for Deployment model to use for predictions for the project.

import datarobot as dr

recommendation = dr.ModelRecommendation.get(project_id)

Retrieve a specific recommendation

If you know which recommendation you want to use, you can select a specific recommendation using the following code.

import datarobot as dr

recommendation_type = dr.enums.RECOMMENDED_MODEL_TYPE.FAST_ACCURATE
recommendations = dr.ModelRecommendation.get_all(project_id, recommendation_type)