These WhizzML script will let the user run a fully automated Machine Learning pipeline in BigML. It needs the AutoML Library, more info here.

From a train, holdout and test datasets, the following tasks will be automatically done:

• Unsupervised Models Generation: Creating the following unsupervised models: Cluster, Anomaly Detector, Association Discovery (with leverage and lift as search_strategy), PCA and Topic Model (for datasets that contain text fields).

• Feature Generation: Using the unsupervised models created previously to append automatically generated new features to all the user given datasets.

• Feature Selection: Reducing, automatically, the number of fields of the datasets using the Recursive Feature Eliminination algorithm. This step is bypassed if the total number of fields is lower than 50 (or 30 if shallow_search is True)

• Model Selection Using OptiML to find the best models and using the top 3 models to create a Fusion model to predict all the test dataset instances. The script will return an evaluation of the final model too.

Find more info in package readme and script readme