Predictive Models

This page covers technical details about creating your own predictive models

Predictive models are hosted machine learning models on Terrene. Currently Terrene supports to specialized models and more generic models.

Model Training

When you select a TFrame that is compatible with your model (contains both input and output variables of your model), Terrene will automatically suggest trainers that are appropriate for the job. On top of the normal regression and classifier models, Terrene has two specialized model types:

1. Anomaly Detection Model

This model will only work with one label (ouput variable) and for it to be suggested by Terrene, your output variable will only need to have a range from 0 (not anomaly) to 1 (anomaly).

2. Probabilist Model

This model will work with only one label (output variable)

Batch Prediction

Terrene supports consuming a TFrame and then writing its predictions to it. The predictions column name will follow the format specified in the Format Output Variable setting.

The default output variable format is {var}__predicted__{object_id} . {var} and {object_id} will automatically be replaced by the name of the variable being predicted and the object_id of the predictive mode respectively.

Recently Trained Models

You can view a list of recently trained instances of the model and revert back to an earlier version of the model if needed.

Predictions Store

All predictions made by the mode will automatically be logged into the Explore Predictions tab.