Skip to content

Algorithms

Overview

Algorithms are objects that store calculations utilized to train models.

Using Algorithm, users can define custom logic for model training, which may include:

  • Employing standard Python modeling libraries, such as XGBoost, Scikit-learn, TensorFlow, ONNX, H2O, etc.

  • Conducting hyperparameter tuning

  • Performing model evaluation

  • Implementing any other desired custom logic

Once defined, algorithms can be utilized and serve as the standard logic for model training.

How to create an Algorithm?

Algorithms can be created by accessing the Algorithms tab under Resource and clicking on Create.

How is this used?

Algorithms are used to train models. Once registered, they can serve as the standard logic utilized in an experiment for training models.

Example

Case Study: User is creating an algorithm that utilize the XGBoost library to train models. This algorithm accepts train and test datasets, as well as user-inputted hyper-parameters, it leverages XGBClassifier to conduct model training, and produces both the trained model in PMML format and the KS value on the test dataset as outputs.

  • User registers the algorithm in the Algorithms tab within the Resource module.

    • Description

    description

    • Definition

      • Input Datasets

      input_dataset

      • Parameters

      parameter

      • Training Logic

      training_logic

      • Outputs

      output

    • Lineage

    lineage

  • User request approval of the algorithm which, if granted will make it available for use across the platform