Overview
Model settings are a powerful feature for managing how ML Apps learn & make decisions as they progress throughout your product development lifecycle. Saving model settings, unless pausing a model, may trigger new model(s) to build overnight.
Model Settings are comprised of the following sections:
This article provides more information about how you can adjust Model Settings.
General Model Settings
General Model Settings govern how your ML App should generate new models. From the Models tab, select Model Settings and the General tab will open.
Below is a summary of each setting's functionality.
Look Back Date Range: Change from how far in the past the ML App should train new models. Another way to think of this is to consider how recent the data should be in training new models.
Impute Performance Goal Data: Replace missing API data with substituted values while your ML App continues learning
Rebalance Data: Optimize predictions on less common decisions
Model Considerations: Control types of models that are built (default or custom).
Build New Models Automatically: Select if models build automatically (changes dynamically) or are paused (current model remains in use). If model building is paused, you will thus have the options to drive predictions and decisions based on the model or on the guardrails.
Feature-Specific Model Settings
Control which features should and should not inform your models. From the Models tab, select Model Settings and then the Model Features tab.
Custom Modeling Plans
Control the types of models that are built (default or custom). From the Models tab, select Model Settings and then the Modeling Plans tab. When Custom is selected, define the model specification parameters and add/remove model specifications
Gradient Boost Machine Model: Commonly used in regression and classification tasks, gradient boost machine models make few assumptions about the data and build a sequence of decision trees. After building a tree, the algorithm evaluates the forest against the training data and then reweights the data to focus on poorly-predicted records.
XGBoost Model: Quite literally an extreme gradient boost machine model, aims to provide a scalable, portable and distributed gradient boosting model with greater complexity.
Random Forest Model: Multiple learning algorithms are used for classification, regression, and other tasks that operate by constructing a multitude of decision trees at training time.
Generalized Linear Model: Generalizes linear regression by relating linear model to the response variable and allowing the magnitude of the variance of each measurement to be a function of its predicted value.
Multi-level Neural Net Model: Power your ML App with "deep learning" models. These models may perform better in specific cases than the standard machine learning models.