What is the recommended way to reduce model training time?

To cut down on model training time, you can use Kumo's model planner to set the run mode for your model plan. By default, Kumo decides the size of the search space so that the search completes in a reasonable amount of time, yielding a close-to-optimal result. This happens automatically under the Normal run mode; however, depending on your budget for training time, you may configure a longer or shorter training time duration.

Select the run mode that best suits your particular scenario:

• Normal: Default value.
• Fast: Speeds up the search process—typically about 4x faster than using the normal mode.
• Best: Typically takes 4x the time used by the normal mode.

Keep in mind that there is a trade-off between search time and optimal search results.

Making Your Jobs Run Faster

If you already know what kind of model architecture you want (e.g., based on your experience writing similar predictive queries on your dataset), you can use the model planner to skip the full AutoML architecture search, and focus on optimizing on a very narrow portion of the search space. For example, if you only run one experiment instead of eight, you can make your job potentially eight times faster, and increase your productivity as a data scientist.

Maximizing Performance

Additionally, if an additional 1-5% of performance is crucial, you can use the Model Planner to eke out more performance (with a likely cost of increased job runtime or other lost functionality). For example, the default model plan typically limits the number of channels to 256, as anything beyond this point usually results in performance benefit that do not outweigh the cost; however, you are free to push the limit. Advanced users can configure settings like the refit option, which trains over the entire dataset, at the cost of losing evaluation metrics on the holdout dataset.

Learn More:

• Model Planner and AutoML