Using Fusions to create better supervised models
In this guide you'll learn how to use Fusions to create better supervised models.
Fusions are composite models made of classifications or regressions, over the same kind of data, with the same objective field. You can fuse together different Supervised models - decision trees, ensembles, logistic regressions, deepnets and other fusions.
The goal of fusions is to balance out the weakenesses of it's underlying models, seeking an incremental performance. For this, users have to understand how different models work with different data and goals.
To create a fusion you must first choose what models you are going to fuse together. In the Machine Learning module, click on the Supervised model category for the first model you chose. Move your cursor to the model you want and click on the arrow to the right of it's name. Select Create Fusion.
In the "New Fusion" screen you will see a few stright-forward options:
- Model type selector and a search bar, so you can add the models you wish to fuse.
- A checkbox to show OptiML-created models in the search.
- On/off filter button, to filter search by models with the same chosen objective field.
- Weight: You can add weights to each model in your fusion. When making predictions, the system will compute a weighted average of all model predictions as per the model weights specified.
- Name: The created Fusion's name.