I attended a very interesting get-together today courtesy of the LA Machine Learning Meetup group. The guest speaker was David Gerster, VP of Data Science at BigML, a provider of online tools for performing machine learning tasks. The title of David’s talk was “Using Machine Learning to Predict Customer Churn” and he used his company’s tools on a training-set containing 10 features as input to a decision tree classifier algorithm. The BigML service includes high quality interactive visualizations like the familiar decision tree representation, but also the unique “sunburst” visualization below. By highlighting different areas on the graph, you can get a quick sense for how different feature variable values contribute to the prediction. I found the sunburst highly intuitive.
The BigML tool-set is available on a subscription basis where the fee is based on how much of the service you use. You can try BigML as much as you want where tasks under 5MB are free using their development mode. You start by uploading a training data set and the application uses it to train the algorithm. You can then use the visualization tools to explore the model and make predictions on new data, or you can export the model to one of several programming languages including Python, R, Java and even Excel. The company also offers BigML.io, a REST-style API for creating and managing BigML resources programmatically.
Machine learning as a service is a new incarnation of this growing field, and BigML is definitely worth a serious look.
Daniel – Managing Editor, insideBIGDATA