US 12,430,147 B2
Hyperparameter tuning method, program trial system, and computer program
Shotaro Sano, Tokyo (JP); Toshihiko Yanase, Tokyo (JP); Takeru Ohta, Tokyo (JP); and Takuya Akiba, Tokyo (JP)
Assigned to Preferred Networks, Inc., Tokyo (JP)
Filed by Preferred Networks, Inc., Tokyo (JP)
Filed on Dec. 10, 2021, as Appl. No. 17/643,661.
Application 17/643,661 is a continuation of application No. PCT/JP2020/022428, filed on Jun. 5, 2020.
Claims priority of application No. 2019-109537 (JP), filed on Jun. 12, 2019.
Prior Publication US 2022/0100531 A1, Mar. 31, 2022
Int. Cl. G06F 9/44 (2018.01); G06F 9/448 (2018.01); G06N 20/00 (2019.01)
CPC G06F 9/4494 (2018.02) [G06N 20/00 (2019.01)] 28 Claims
OG exemplary drawing
 
1. A hyperparameter configuration device comprising:
at least one memory; and
at least one processor configured to:
acquire a program execution instruction written through a command-line interface, the program execution instruction including a name of a program to be trialed and parameter description data;
set a value of a hyperparameter of the program, based on the parameter description data;
acquire a result of a trial of the program, the trial of the program being executed with the value of the hyperparameter; and
set a next value of the hyperparameter of the program, based on the result of the trial,
wherein a behavior of a machine learning algorithm is controlled using the next value of the hyperparameter,
wherein the parameter description data is written through the command-line interface,
wherein the parameter description data includes a name of the hyperparameter, a distribution identifier of the hyperparameter, and a range specification value of the hyperparameter,
wherein the at least one processor sets the value of the hyperparameter based on the distribution identifier and the range specification value included in the parameter description data,
wherein the distribution identifier includes information of at least one of a type of a target to be selected as the value of the hyperparameter or a distribution of the target to be selected as the value of the hyperparameter, and
wherein the parameter description data is not written within the program.