Data science algorithms
You can use the Algorithms tab to view all available algorithms for custom data science models. You have the option of selecting ready-to-use algorithms for custom data science models, or you can create custom algorithms.
To enable this feature, please create a service request at My Oracle Support.
Custom algorithms that you use for custom models must be in Python. Additional supported languages are planned for future Unity releases.
Custom algorithms explained
You have the option of creating custom algorithms and configuring the parameters of the algorithm that meets your specific needs. You can then use the custom algorithm in a custom data science model. Learn more about Creating custom data science models.
Parameters of custom algorithms
When creating custom algorithms, you will need to define the following parameters for the algorithm:
-
Purpose: Define the algorithm for Scoring and Training, or only Scoring.
-
Docker image: Import the algorithm code as a docker image.
-
JSON file: You will need to upload a JSON file that will be loaded during scoring.
-
Algorithm parameters: Allow you to tweak the model according to the required content at any point in time. For example, you can configure a lookback window and provide several options with lookback windows of differing number of days.
-
Hyperparameters: Allow you to configure adjustable parameters that can be tuned into order to get a model with optimal performance. For example, you can configure the hyperparameters such as learning rate or tuning rate.
Create and use custom algorithms
To create and use a custom algorithm, you will need to do the following:
-
Follow the steps for Creating custom algorithms.
-
Follow the steps for Creating custom data science models. When configuring the model, select the custom algorithm you selected.
After completing all the steps for creating and using a custom model, the custom algorithm will be leveraged to train and score the custom model.