Performing hyperparameter tuning of models in a massively parallel database system

Patent metadata

Field	Value
Publication number	US-2021397975-A1
Application number	US-202017124200-A
Country	US
Kind code	A1
Filing date	Dec 16, 2020
Priority date	Jun 17, 2020
Publication date	Dec 23, 2021
Grant date	—

Abstract

Hyperparameter tuning for a machine learning model is performed in a massively parallel database system. A computer system comprised of a plurality of compute units executes a relational database management system (RDBMS), wherein the RDBMS manages a relational database comprised of one or more tables storing data. One or more of the compute units perform the hyperparameter tuning for the machine learning model, wherein the hyperparameters are control parameters used in construction of the model, and the tuning of the hyperparameters is implemented as an operation in the RDBMS that accepts training and scoring data for the model, constructs the model using the hyperparameters and the training data, and generates goodness metrics for the model using the scoring data.

First claim

What is claimed is:

1 . A computer-implemented apparatus, comprising:
(a) a relational database management system (RDBMS) executing in a computer system comprised of a plurality of compute units, wherein the RDBMS manages a relational database comprised of one or more tables storing data; (b) one or more of the compute units tuning hyperparameters for a machine learning model, wherein the hyperparameters are control parameters used in construction of the model, and the tuning of the hyperparameters is implemented as an operation in the RDBMS that accepts training and scoring data for the model, constructs the model using the hyperparameters and the training data, and generates goodness metrics for the model using the scoring data.

2 . The apparatus of claim 1 , wherein a search space for the hyperparameters is defined by one or more combinations of the hyperparameters, and the search space is partitioned across the compute units to parallelize the tuning of the hyperparameters.

3 . The apparatus of claim 2 , wherein:
an enumerator enumerates the combinations of hyperparameters in the search space based on an optimization algorithm; and a function evaluator computes an objective function for the combinations of the hyperparameters enumerated in the search space, wherein the objective function computes one or more goodness metrics for the model generated using one or more of the combinations of the hyperparameters enumerated in the search space, to identify an optimal one of the combinations of the hyperparameters.

4 . The apparatus of claim 3 , wherein a plurality of the compute units perform the function evaluator concurrently using the search space that is partitioned across the compute units.

5 . The apparatus of claim 3 , wherein the operation includes training and scoring functions used for computation of the objective function.

6 . The apparatus of claim 5 , wherein
the training data is used by the training function to train the model generated using the hyperparameters; the scoring data is used by the scoring function to score the model trained by the training function; and the goodness metrics are used to evaluate the model scored by the scoring function.

7 . The apparatus of claim 3 , wherein the operation includes one or more optimization algorithms for the enumerator, and the enumerator performs a selective enumeration of the combinations of the hyperparameters in the search space based on the optimization algorithm.

8 . The apparatus of claim 7 , wherein the enumerator repeats the selective enumeration of the combinations of the hyperparameters in the search space until a convergence is reached based on the optimization algorithm.

9 . The apparatus of claim 3 , wherein the operation includes one or more arguments for: a ratio of the training and scoring data split used for verification; a k-fold value for cross-validation of the training and scoring data; and the goodness metrics used for comparison of predicted and actual values for the training and scoring data used by the model.

10 . A computer-implemented method, comprising:
(a) executing a relational database management system (RDBMS) in a computer system comprised of a plurality of compute units, wherein the RDBMS manages a relational database comprised of one or more tables storing data; (b) tuning hyperparameters for a machine learning model in one or more of the compute units, wherein the hyperparameters are control parameters used in construction of the model, and the tuning of the hyperparameters is implemented as an operation in the RDBMS that accepts training and scoring data for the model, constructs the model using the hyperparameters and the training data, and generates goodness metrics for the model using the scoring data.

11 . The method of claim 10 , wherein a search space for the hyperparameters is defined by one or more combinations of the hyperparameters, and the search space is partitioned across the compute units to parallelize the tuning of the hyperparameters.

12 . The method of claim 11 , wherein:
an enumerator enumerates the combinations of hyperparameters in the search space based on an optimization algorithm; and a function evaluator computes an objective function for the combinations of the hyperparameters enumerated in the search space, wherein the objective function computes one or more goodness metrics for the model generated using one or more of the combinations of the hyperparameters enumerated in the search space, to identify an optimal one of the combinations of the hyperparameters.

13 . The method of claim 12 , wherein a plurality of the compute units perform the function evaluator concurrently using the search space that is partitioned across the compute units.

14 . The method of claim 12 , wherein the operation includes training and scoring functions used for computation of the objective function.

15 . The method of claim 14 , wherein
the training data is used by the training function to train the model generated using the hyperparameters; the scoring data is used by the scoring function to score the model trained by the training function; and the goodness metrics are used to evaluate the model scored by the scoring function.

16 . The method of claim 12 , wherein the operation includes one or more optimization algorithms for the enumerator, and the enumerator performs a selective enumeration of the combinations of the hyperparameters in the search space based on the optimization algorithm.

17 . The method of claim 16 , wherein the enumerator repeats the selective enumeration of the combinations of the hyperparameters in the search space until a convergence is reached based on the optimization algorithm.

18 . The method of claim 12 , wherein the operation includes one or more arguments for: a ratio of the training and scoring data split used for verification; a k-fold value for cross-validation of the training and scoring data; and the goodness metrics used for comparison of predicted and actual values for the training and scoring data used by the model.

19 . A computer program product, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a computer system to cause the computer system to perform a method, comprising:
(a) executing a relational database management system (RDBMS) in a computer system comprised of a plurality of compute units, wherein the RDBMS manages a relational database comprised of one or more tables storing data; (b) tuning hyperparameters for a machine learning model in one or more of the compute units, wherein the hyperparameters are control parameters used in construction of the model, and the tuning of the hyperparameters is implemented as an operation in the RDBMS that accepts training and scoring data for the model, constructs the model using the hyperparameters and the training data, and generates goodness metrics for the model using the scoring data.

20 . The computer program product of claim 19 , wherein a search space for the hyperparameters is defined by one or more combinations of the hyperparameters, and the search space is partitioned across the compute units to parallelize the tuning of the hyperparameters.

Assignees

• Teradata Us Inc

Inventors

• Al-Omari Awny Kayed
• Oblogin Maksym Sergiyovych
• Bouaziz Khaled
• Hanlon Michael James
• Siddiqui Kashif Abdullah

Classifications

• G06F16/284

  Relational databases · CPC title

• G06N20/00 Primary

  Machine learning · CPC title

• G06F16/25

  Integrating or interfacing systems involving database management systems · CPC title

• G06N3/126 Primary

  Evolutionary algorithms, e.g. genetic algorithms or genetic programming · CPC title