Generation of a predictive model for selection of batch sizes in performing data format conversion
US11983172B2 · kind B2 · utility
Assignee
Inventors
Key dates
| Filing date | Dec 7, 2021 |
| Grant date | May 14, 2024 |
| Priority date | — |
| Expiry date | Apr 7, 2042 |
Classification
- Technology area (CPC G)Physics
- CPC primaryG06F16/258
- WIPO fieldComputer technology
- WIPO sectorElectrical engineering
Abstract
A method includes generating a predictive model mapping batch size to estimated memory utilization based on a plurality of prior batch sizes applied for performing a corresponding plurality of prior data format conversions, and further based on a plurality of prior actual memory utilization occurring in performing the corresponding plurality of prior data format conversions. A next batch size is automatically selected for performance of a subsequent data format conversion based on the predictive model and a current memory availability. A batch of data for performance of the subsequent data format conversion is identified based on the batch of data having the next batch size. The subsequent data format conversion is performed the batch of data, and an actual amount of memory utilized in performing the subsequent data format conversion is determined. The predictive model is updated based on the next batch size and the actual amount of memory utilized.
Source: USPTO / EPO open patent data. Objective bibliographic and citation counts.