COMPUTER-IMPLEMENTED SYSTEM FOR HIERARCHICAL UNCONSTRAINING IN DATA PROCESSES

摘要

Exemplary embodiments are generally directed to methods, mediums, and systems for correcting censored or constrained historical data with various possible types of computing devices, including cloud-based devices, personal computing devices, and edge-based devices. The corrected data may be used in forecasting, for example to forecast demand for a limited resource. In some embodiments, the data is modeled at a higher level of granularity than an individual record. The aggregated demand may then be pro-rated over a group of categories or users where a given category of users that might be small or nonexistent over a certain time frame may be better accommodated. Moreover, it may be easier or more efficient to make assumptions and employ computing resources at the aggregate level.

主权项

1. An apparatus comprising:
a processor; and a storage to store instructions that, when executed by the processor, cause the processor to: access a plurality of records from one or more storage devices, the plurality of records associated with a resource divided into a predetermined number of units, the plurality of records subject to a maximum cutoff threshold above which further records are not created, at least the resource and the units forming a hierarchy having a lower hierarchical level corresponding to the units and a higher hierarchical level corresponding to the resource; access a plurality of indicator variables that indicate a demand for a plurality of units of the resource among a plurality of users; model demand for the plurality of units of the resource by the plurality of users at a hierarchical level higher than the units of the resource; apply a censored regression model to the modeled demand based on the plurality of indicator variables to generate a partial censoring adjusted unconstrained demand, the applying comprising multiplying one or more observations of the indicator variables by one or more regression coefficients for the modeled demand as estimated by a maximum likelihood estimator (MLE); use the censored regression model to adjust the plurality of records to account for an amount of constrained demand above the maximum cutoff threshold by distributing the partial-censoring-adjusted unconstrained demand to a hierarchical level lower than the hierarchical level at which the demand was modeled; and transmit or store information relating to the amount of constrained demand.