Evaluating Modifications to Features Used by Machine Learned Models Applied by an Online System

摘要

An online system identifies an additional feature to evaluate for inclusion in a machine learned model. The additional feature is based on characteristics of one or more dimensions of information maintained by the online system. To generate data for evaluating the additional feature, the online system generates various partitions of stored data, where each partition includes characteristics associated with one or more dimensions on which the additional feature is based. Using values of characteristics in a partition, the online system generates values for the additional feature and includes the values of the additional feature in the partition. Values for the additional feature are generated for various partitions based on the values of characteristics in each partition. The online system combines multiple partitions that include values for the additional feature to generate a training set for evaluating a machine learned model including the additional feature.

主权项

1. A method, comprising:
maintaining a model at an online system, the model receiving a plurality of features and generating an output based on the received features; identifying an additional feature for the model, a value for the additional feature based on values of one or more characteristics of one or more dimensions maintained by the online system at one or more times prior to a current time; obtaining data stored by the online system and including characteristics of one or more dimensions; generating a plurality of different partitions of the data stored by the online system based on the one or more dimensions, each partition of data including values of one or more characteristics of the one or more dimensions on which the additional feature is based; modifying each partition of data to include one or more values for the additional feature, a value of the additional feature included in a partition of data determined from values of one or more characteristics of the one or more dimensions on which the additional feature is based that are included in the partition of data and are associated with one or more times prior to the current time; generating a training set by combining the modified partitions of data; and applying a modified model including the additional feature to the training set to generate one or more results.