Solution to end-to-end feature engineering automation

Patent metadata

Field	Value
Publication number	US-2021158073-A1
Application number	US-201916691294-A
Country	US
Kind code	A1
Filing date	Nov 21, 2019
Priority date	Nov 21, 2019
Publication date	May 27, 2021
Grant date	—

Abstract

Aspects of the present disclosure involve systems, methods, devices, and the like for an end-to-end solution to auto-identifying features. In one embodiment, a novel architecture is presented that enables the identification of optimal features and feature processes for use by a machine learning model. The novel architecture introduces a feature orchestrator for managing, routing, and retrieving the data and features associated with analytical job request. The novel architecture also introduces a feature store designed to identify, rank, and store the features and data used in the analysis. To aid in identifying the optimal features and feature processes, a training system may also be included in the solution which can perform some of the training and scoring of the features.

First claim

What is claimed is:

1 . A system comprising:
a non-transitory memory storing instructions; and a processor configured to execute instructions to cause the system to:
receiving, via a wireless network communication, a request for a first feature process evaluation at a feature orchestrator;
determining, by an analytical session manager, a plurality of feature processes and data associated with the feature process evaluation request;
determining, a first set of features for a machine learning model, based in part on a training using the plurality of processes and data determined;
processing, a second feature process evaluation using at least some of the first set of features associated with the first feature process; and
determining, a plurality of optimal features for the first and second feature process evaluations.

2 . The system of claim 1 , executing instructions further causes the system to:
associating the first set of features determined with the second feature process evaluation based in part on an evaluation type.

3 . The system of claim 1 , executing instructions further causes the system to:
associating, by a feature store module, an analytical session identification number with the feature process evaluation, wherein the determining of the plurality of processes and data is associated to the analytical session identification number.

4 . The system of claim 3 , wherein the analytical session identification number is determined by the feature orchestrator.

5 . The system of claim 1 , wherein the features with a higher feature score are promoted by a feature store module for use with the second feature process evaluation.

6 . The system of claim 5 , wherein the analytical session identification number isolates the first feature process evaluation from another feature process evaluation.

7 . The system of claim 6 , wherein the first feature process evaluation and the another feature process evaluation run in parallel.

8 . A method comprising:
in response to receipt of a request to run an analytical session for feature process evaluation, determining an analytical session identification number; transmitting the analytical session number to a feature store module and consumer module; determining and retrieving corresponding features and feature processes based in part on the feature process evaluation requested by the feature store module; transmitting, by the consumer module, the corresponding features and feature processes to a training module; determining top features and feature processes by the training module; and promoting the top features and feature processes to a feature orchestrator.

9 . The method of claim 8 , further comprising:
routing, by the feature orchestrator, a request to the feature store module to determine the analytical session identification number.

10 . The method of claim 8 , wherein the feature orchestrator associates the promoted features and feature processes with a type of feature process evaluation, and the feature orchestrator uses the promoted features and feature processes in another feature process evaluation of a same type.

11 . The method of claim 8 , further comprising:
retiring, the features and feature processes not determined to be top feature and feature processes in the feature process evaluation.

12 . The method of claim 8 , wherein the training module uses labels and modules associated with feature process evaluation.

13 . The method of claim 8 , wherein the feature orchestrator includes a ranking engine used for determining the top features.

14 . The method of claim 8 , wherein the promoting occurs by a feature registry located within the feature store module.

15 . The method of claim 8 , further comprising:
determining the top features are optimal features when a machine learning model converges.

16 . A non-transitory machine-readable medium having stored thereon machine-readable instructions executable to cause a machine to perform operations comprising:
receiving a request for a first feature process evaluation at a feature orchestrator; determining a plurality of feature processes and data associated with the feature process evaluation request; determining a first set of features for a machine learning model, based in part on a training using the plurality of processes and data determined; processing a second feature process evaluation using at least some of the first set of features associated with the first feature process; and determining a plurality of optimal features for the first and second feature process evaluations.

17 . The non-transitory medium of claim 16 , further comprising:
associating the first set of features determined with the second feature process evaluation based in part on an evaluation type.

18 . The non-transitory medium of claim 16 , further comprising:
associating, by a feature store module, an analytical session identification number with the feature process evaluation, wherein the determining of the plurality of processes and data is associated to the analytical session identification number.

19 . The non-transitory medium of claim 16 , wherein the analytical session identification number is determined by the feature orchestrator.

20 . The non-transitory medium of claim 16 , wherein the features with a higher feature score are promoted by a feature store module for use with the second feature process evaluation.

Assignees

• Paypal Inc

Inventors

• Xie Kai
• Ho Chun Kiat
• Song Lin
• Neo Rey
• Soenaryo Edwin Boaz
• Bui Kim Dung

Classifications

• G06V10/771 Primary

  Feature selection, e.g. selecting representative features from a multi-dimensional feature space · CPC title

• G06F18/2113

  by ranking or filtering the set of features, e.g. using a measure of variance or of feature cross-correlation · CPC title

• G06F18/2185

  the supervisor being an automated module, e.g. intelligent oracle · CPC title

• G06F18/2148

  characterised by the process organisation or structure, e.g. boosting cascade · CPC title

• G06N20/00

  Machine learning · CPC title