Wrap Up: Custom ML Models#

Congratulations!#

You’ve just completed the Custom ML Models course, where you learned about ways to share your code with other team members on the same Dataiku instance.

Here are a few of the main takeaways from this course:

  • Within the visual ML tool of Dataiku, you have the ability to apply custom feature preprocessing and custom algorithms to your data, in addition to the available built-in methods and algorithms.

  • Custom preprocessors used in the code editor of the visual ML tool must be scikit-learn compatible and must have a fit() and a transform() method.

  • Custom algorithms used in the code editor of the visual ML tool must be scikit-learn compatible and must have a fit() and a predict() method. In addition, classifiers can implement a predict_proba() method and must have a classes_ attribute.

  • It is best practice to define preprocessors and custom algorithms in libraries and instantiate them in the code editor. You can also import preprocessors and algorithms that are included in the runtime code environment used by the visual ML tool.

  • Finally, you can also access custom models by using a prediction algorithm component of a plugin.

Learn more#

To learn about using code to create and package extended functionalities as plugins, be sure to check out the Plugin Development course in the Developer Learning path!