System and method for constructing top-performing pipelines using hierarchical configuration space

Patent metadata

Field	Value
Publication number	US-2024330756-A1
Application number	US-202318194601-A
Country	US
Kind code	A1
Filing date	Mar 31, 2023
Priority date	Mar 31, 2023
Publication date	Oct 3, 2024
Grant date	—

Abstract

A computer-implemented method for developing a hierarchical machine-learning pipeline can include receiving a hierarchy specification, a set of estimators for the root node, and one or more transformer options for each of the transformer nodes. The hierarchy specification provides a configuration of the root node, transformer nodes, and edges interconnecting the root and transformer nodes. A rank can be obtained for each estimator in the root node. A hierarchy pipeline traverser can then traverse a first child layer of the transformer nodes connected to the root node via one of the edges. A first ranked list of pathways can be determined with respect to the one or more transformer options selected for the first child layer and at least one selected estimator of the root node.

First claim

What is claimed is:

1 . A computer-implemented method for developing a hierarchical machine-learning pipeline, comprising:
receiving a hierarchy specification, the hierarchy specification providing a root node, transformer nodes, and edges interconnecting the root node and the transformer nodes; receiving a set of estimators for the root node, and one or more transformer options for each of the transformer nodes; obtaining a rank for each estimator in the root node; traversing, using a hierarchy pipeline traverser, a first child layer of the transformer nodes connected to the root node via one of the edges; and determining a first ranked list of pathways with respect to the one or more transformer options selected for the first child layer and at least one selected estimator of the root node.

2 . The computer-implemented method of claim 1 , further comprising obtaining a ranked list of traversal options at the root node.

3 . The computer-implemented method of claim 1 , further comprising selecting a predefined number of estimators based on the rank for the at least one selected estimator of the root node.

4 . The computer-implemented method of claim 1 , further comprising:
traversing, using the hierarchy pipeline traverser, a second child layer of the transformer nodes connected to either the first child layer or the root node by one of the edges; and determining a second ranked list of pathways with respect to the one or more transformer options selected for the first child layer, the second child layer, and at least one selected estimator of the root node.

5 . The computer-implemented method of claim 1 , further comprising selecting the set of estimators for the root node and the one or more transformer options for each of the transformer nodes from a knowledge base.

6 . The computer-implemented method of claim 1 , further comprising placing preferred ones of the transformer nodes in a layer of the hierarchy specification closer to the root node as compared to non-preferred ones of the transformer nodes.

7 . The computer-implemented method of claim 1 , further comprising placing the transformer nodes having a greater time and/or memory complexity in a layer of the hierarchy specification farther away from the root node.

8 . The computer-implemented method of claim 7 , wherein the time and/or memory complexity of the transformer nodes is received from a knowledge base.

9 . The computer-implemented method of claim 1 , further comprising providing annotations for each of the edges, wherein the annotations include an insert edge, an append edge, or a replace edge.

10 . The computer-implemented method of claim 1 , further comprising receiving a configuration specification that defines preferences, priorities and/or constraints to be used by the hierarchy pipeline traverser.

11 . The computer-implemented method of claim 10 , wherein the configuration specification defines preferences, priorities and/or constraints for at least one of the root node or the transformer nodes or for at least one of the edges.

12 . A computer-implemented method for developing a hierarchical machine-learning pipeline, comprising:
receiving a hierarchy specification, the hierarchy specification providing a root node, transformer nodes, and edges interconnecting the root node and the transformer nodes; receiving a set of estimators for the root node, and one or more transformer options for each of the transformer nodes; receiving a configuration specification, the configuration specification defining preferences, priorities and/or constraints to be used by a hierarchy pipeline traverser; obtaining a rank for each estimator in the root node; traversing, using the hierarchy pipeline traverser, a first child layer of the transformer nodes connected to the root node via one of the edges; determining a first ranked list of pathways with respect to the one or more transformer options selected for the first child layer and at least one selected estimator of the root node; traversing, using the hierarchy pipeline traverser, a second child layer of the transformer nodes connected to either the first child layer or the root node by one of the edges; and determining a second ranked list of pathways with respect to the one or more transformer options selected for the first child layer, the second child layer, and at least one selected estimator of the root node.

13 . The computer-implemented method of claim 12 , further comprising selecting a predefined number of estimators based on the rank for the at least one selected estimator of the root node.

14 . The computer-implemented method of claim 12 , further comprising receiving the set of estimators for the root node and the one or more transformer options for each of the transformer nodes from a knowledge base.

15 . The computer-implemented method of claim 12 , further comprising placing preferred ones of the transformer nodes in a layer of the hierarchy specification closer to the root node as compared to non-preferred ones of the transformer nodes.

16 . The computer-implemented method of claim 12 , further comprising placing the transformer nodes having a greater time and/or memory complexity in a layer of the hierarchy specification farther away from the root node.

17 . The computer-implemented method of claim 12 , further comprising providing annotations for each of the edges, wherein the annotations include an insert edge, an append edge, or a replace edge.

18 . A non-transitory computer readable storage medium tangibly embodying a computer readable program code having computer readable instructions that, when executed, causes a computer device to carry out a method for developing a hierarchical machine-learning pipeline, the method comprising:
receiving a hierarchy specification, the hierarchy specification providing a root node, transformer nodes, and edges interconnecting the transformer nodes; receiving a set of estimators for the root node, and one or more transformer options for each of the transformer nodes; obtaining a rank for each estimator in the root node; traversing, using a hierarchy pipeline traverser, a first child layer of the transformer nodes connected to the root node via one of the edges; and determining a first ranked list of pathways with respect to the one or more transformer options selected for the first child layer and at least one selected estimator of the root node.

19 . The non-transitory computer readable storage medium of claim 18 , the method further comprising providing annotations for each of the edges, wherein the annotations including an insert edge, an append edge, or a replace edge.

20 . The non-transitory computer readable storage medium of claim 18 , the method further comprising receiving a configuration specification, the configuration specification defining preferences, priorities and/or constraints to be used by the hierarchy pipeline traverser.

Assignees

• IBM

Inventors

• Patel Dhavalkumar C
• Jayaraman Srideepika
• Lin Shuxin
• Bhamidipaty Anuradha
• Kalagnanam Jayant R

Classifications

• G06N7/01

  Probabilistic graphical models, e.g. probabilistic networks · CPC title

• G06N5/01

  Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound · CPC title

• G06N5/022

  Knowledge engineering; Knowledge acquisition · CPC title

• G06N20/00 Primary

  Machine learning · CPC title