BigML’s first release of the year is here! Join us on Wednesday, January 31, 2018, at 10:00 AM PT (Portland, Oregon. GMT -08:00) / 07:00 PM CET (Valencia, Spain. GMT +01:00) for a FREE live webinar to discover the latest version of the BigML platform. We will be presenting two new features: operating thresholds for classification models to fine tune the performance of your models; and organizations, a convenient collaborative space that breaks down silos and makes it easy and efficient for any company to adopt Machine Learning across their entire corporate structure.

During fall 2017, the BigML Team has implemented operating thresholds, now available from the BigML Dashboard, API, and from WhizzML for automation. The ability to set an operating threshold is a key feature that allows you to tell BigML to be more or less aggressive when predicting a class for an instance. Setting the right threshold for the positive class when evaluating your classification models can be especially useful when your dataset has one class that is particularly rare. Because it makes up a small portion of the overall data, models will rarely predict this class by default. If this “minority class” is the class of interest (as is the case in fraud detection), setting an appropriate operating threshold can assure the model predicts the minority class with reasonable frequency. Applying operating thresholds is especially useful in domains like fraud detection or medical diagnosis, where the consequences of some classifications may have prohibitive costs associated.

The ultimate goal of any BigML resource is making predictions. Now, BigML provides three types of thresholds, one for each of the certainty measures that BigML offers for your classification model predictions: probabilities, confidences, and votes. All classification models (Decision Trees, Ensembles, Deepnets, and Logistic Regressions) return a per-class probability along with the prediction. Thus you can apply a probability threshold for the positive class for any model. Decision trees and non-boosted ensembles predictions also come with a per-class confidence, a pessimistic measure of the model certainty, so you can apply a confidence threshold for these models. Finally, only for non-boosted ensembles, BigML offers another metric called votes that takes into account the percentage of trees in the ensemble to predict a given class. As an alternative to the probability or the confidence, you can also apply a vote threshold for these models. These three metrics will be explained in detail during our webinar.

Finally, BigML is bringing organizations to the BigML Dashboard, a space where several customers can work on the same projects, using the same Dashboard, but at different levels of privileges. Organizations are ideal for teams that want to work collaboratively and more efficiently to get the most out of their Machine Learning models.

Want to know more about operating thresholds and organizations?

If you have any questions or you’d like to learn more about how operating thresholds and organizations work, please visit the dedicated release page. It includes a series of blog posts, the BigML Dashboard and API documentation, the webinar slideshow as well as the full webinar recording.