This article explains the motivations behind choosing different application types in the AI & Analytics Engine.
In this article you will learn the motivations behind choosing different application types in the AI & Analytics Engine. Specifically, when should you use the “build from scratch” option to create an app as compared to the “template” option of creating an app.
Note: If you want to learn more about apps, see this article.
A template is a structured approach, designed to solve a specific business problem. It offers a more streamlined way of tackling the problem, by providing guidance through all the stages of the solution, including data ingestion, analysis, processing, modeling, deployment, and consumption of outputs.
For more information about what templates are, see this article.
Build from scratch
Build from scratch is a more manual method of creating a classification, regression or clustering app. It is not limited to any specific business problem. Since the process is manual, the user may need to perform additional tasks, such as creating data preparation recipes, or manually triggering a trained pipeline for predictions outside the platform etc.
For more information about what "build from scratch" is, see this article.
When to use “build from scratch” vs “template”?
If a template that matches your business problems is available on the platform, it’s always preferable to choose it, because:
It is easier to use.
It guides the user through all steps.
Automatically prepares the data and trains one or more models.
May have (depending on the template) additional ML-Ops capabilities, such as scheduled querying of user’s data sources and triggering of prediction pipelines over new data).
Requires less time to solve the problem from end to end due to extensive automation.
If none of the templates suit your particular business requirement, you should choose the “build from scratch” option instead. The advantage of this choice is mainly greater flexibility:
Can use any tabular datasets and schemas.
Control over data preparation recipes.
Manual selection of models for training.
When possible, it is recommended to choose “template”, otherwise, use the “build from scratch” option instead.