* Add more experiments to models_class1_experiments
* Add `--quantitative-only` flag to training script
* Add `--no-throw` flag to prediction script

Ensemble models: return reference (using Kubeface remote_object support) to trained models instead of trained models themselves