| CPC G06N 20/00 (2019.01) [G06F 8/65 (2013.01); G06F 18/2113 (2023.01); G06V 10/757 (2022.01); G06F 8/71 (2013.01)] | 20 Claims |
|
1. A computer-implemented method comprising:
identifying a machine learning model with missing lineage;
generating a creation event and deployment event for the machine learning model;
generating a batch scoring event for the machine learning model, wherein the generating the batch scoring event comprises:
identifying an amount of target batch records for each target batch and an amount of source batch records for each source batch for the machine learning model;
comparing the amount of source batch records for each source batch;
determining, based on the comparing, whether a plurality of source batch records have an equal amount of records; and
mapping, based on the determining, each source batch to each corresponding target batch;
generating a version change event for the machine learning model, wherein generating the version change event comprises:
identifying a plurality of predicted data points with a model confidence less than or equal to a threshold model confidence value,
rescoring a plurality of predicted data points based on the machine learning model at a second time period, resulting in an updated plurality of predicted data points,
determining that the updated plurality of predicted data points are significantly different than the plurality of predicted data points, wherein the determining comprises:
calculating a difference factor between the updated plurality of predicted data points and the plurality of predicted data points, and
comparing the difference factor to a threshold difference factor, and
inferring, based on the determining, that there is a new version of the machine learning model; and
creating a lineage path for the machine learning model based on the batch scoring event, the creation event, the deployment event, and the version change event.
|