The AI & Analytics Engine provides a large selection of machine learning algorithms, and the Model Recommender helps the user select the best ML algorithm for their dataset.
The Engine’s AI-driven recommender predicts how machine learning algorithms will perform on a particular dataset even before training has begun, and ranks them according to their estimated performance. This means that instead of applying the more taxing approach where hundreds of models are trained, users can simply focus on the top few recommended options.
Model recommendation when creating a new App
On the Engine, users can specify the number of algorithms to train in the “build models“ dialog when creating a new App – in this case, the model recommender estimates the performance for all supported algorithms, and only selects the best-performing algorithms for training. Users can optionally define an upper bound for the training time and/or a lower bound for prediction quality. The algorithms with an estimated training time or prediction quality outside the specified bounds will be filtered out from the training list.
Model recommendation when creating an App
Model recommendation when training more models
Users can train more models in an existing App. Before they determine which algorithms to use, the model recommender gives a detailed comparison based on the estimated prediction quality and training time.
Model recommendation when training more model(s) in an existing App