System and method for reproducible machine learning
AssigneeRoyal Bank of Canada
InventorsWeiguang DING, Yanshuai CAO
Life of this patent
Estimating the term…
Abstract
Systems and methods for computationally generating a set of more “stable” configuration default values that are used for traceability and improving reproducibility of machine learning approaches. Hash values are generated based on a merged/modified configuration and both configuration content and hash are stored together in one or more data structures. These data structures can be used to link back to the actual values used in experiments.
Claims20
A system for generating one or more data structures representative of one or more factors used in obtaining one or more outputs from machine learning program, the system comprising:
- a machine learning pipeline input receiver configured to process one or more input files for the machine learning program and extract time-encoded data sets representative of:
- a data path or content, source code, hyperparameter configuration, and a software environment;
- a hashing processor configured for generating a plurality of hash values corresponding to the data path or content, the source code, the hyper parameter configuration, and the software environment;
- and
- a data storage configured to store the plurality of hash values linked to one or more corresponding output files of execution of the machine learning program and to generate the one or more data structures representative of the one or more factors used in obtaining the corresponding one or more outputs.
Non-transitory computer readable medium storing instructions executable by a processor to configure the processor to:
- generate one or more data structures representative of one or more factors used in obtaining one or more outputs from machine learning program;
- process one or more input files for the machine learning program and extract time-encoded data sets representative of:
- a data path or content, source code, hyperparameter configuration, and a software environment;
- generate a plurality of hash values corresponding to the data path or content, the source code, the hyper parameter configuration, and the software environment;
- and
- store the plurality of hash values linked to one or more corresponding output files of execution of the machine learning program and to generate the one or more data structures representative of the one or more factors used in obtaining the corresponding one or more outputs.
Dependencies are inferred from the claim text.