This guide will show you how to use the model recommender in the Engine.
When training a new model, user has the choice of selecting which algorithm/model for training. For new users, this decision can be overwhelming as there is a long list of algorithms available.
The model recommender
To assist users in making this decision, the Engine provides a model recommender that estimates the performance of each of these algorithms, together in consideration with your dataset, the:
- Predictive performance - percentage value of how close your model's predictions will be to the true values. A 100% score indicates that the model is estimated to be able to predict without any errors.
- Prediction time - duration needed for trained model to generate prediction on new data.
- Training time - duration needed to finishing training the model on the dataset
A higher rating/estimate indicates that the model has a better or faster performance.
Sort model templates based on recommendations
Based on your priority, you can sort the model templates based on the estimated performance.
Simply select from the drop-down list which metric do you want to sort by. The models will be sorted from best-performing to worst-performing.