Off-line processing and modeling
Last updated
Last updated
Process of off-line aggreagetes computing and modeling should be repeated at fixed intervals (induced automatically or manually). For every model an analytical table is created. The table contains 1 line for each user, that fulfils trigger condition and has counted target. For every analytical table, the Automatic Business Modeler builds a model. Deployed models and information about used variables (model's signature) are saved to metadata. The method inducing off-line processing starts an API ABM method, that enable to build a model in only 1 query (for more information see ). A table with aggregates is an input table to ABM process, that selects suitable variables and calculates optimal model automatically. This method also enables saving the aggregates table for further analysis or manual modeling. For this purpose, a GDBase alias and a table name must be given.
Process of building models can be scheduled. Scheduling is done through inducing an API adding model request (see . After first build of a model, the task is scheduled for the next day. If model building succeed and the model will be implemented (accurate number of positive targets, accurate quality of the model based on ROC curve), then next model building is scheduled for the next week.