Developing machine learning models may require reuse of features across existing feature pipelines. It is important that those features are not created repeatedly because not only will it require extra resources for computation and development, it also incurs unnecessary cost for maintaining features in different pipelines.

In this webinar, we will introduce the "feature view" from Hopsworks, which is the API for ML model development. We will present how the feature view reuses features produced by different feature pipelines, and avoids training-serving skew by supporting training data creation and feature serving with model dependent transformation functions.