CN110046046A - A kind of distributed hyperparameter optimization system and method based on Mesos - Google Patents
A kind of distributed hyperparameter optimization system and method based on Mesos Download PDFInfo
- Publication number
- CN110046046A CN110046046A CN201910278557.4A CN201910278557A CN110046046A CN 110046046 A CN110046046 A CN 110046046A CN 201910278557 A CN201910278557 A CN 201910278557A CN 110046046 A CN110046046 A CN 110046046A
- Authority
- CN
- China
- Prior art keywords
- mesos
- hyperparameter optimization
- distributed
- application
- layer
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5005—Allocation of resources, e.g. of the central processing unit [CPU] to service a request
- G06F9/5027—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals
Landscapes
- Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of distributed hyperparameter optimization system and method based on Mesos, it is combined including computation layer described in computation layer and dispatch layer with distributed hyperparameter optimization algorithm, the computation layer is made of all kinds of optimization algorithms, for carrying out the sampling of distributed hyperparameter optimization and generating calculating task, the dispatch layer is the specific implementation of a Mesos operation frame, is mainly responsible for resource allocation and executes calculating task.A kind of distributed hyperparameter optimization system and method based on Mesos proposed by the invention can satisfy multi-tenant and use under mixed portion's scene of High-Performance Computing Cluster, improve the efficiency of hyperparameter optimization, the scope of application of the strategy is more extensive, invasive low to already existing Mesos group system.
Description
Technical field
The present invention relates to scheduler technical field more particularly to a kind of distributed hyperparameter optimization systems based on Mesos
And method.
Background technique
Scheduler is the ability that can be provided timing and excite task, the ability of resource management, the dependence of maintenance task
With the system of execution sequence, some scheduling systems are also integrated with the tool of Mission Monitor and various measure of criterions, cluster resource tune
Degree is the system of the resource management and scheduling in distributed system, and main to provide the ability of resource management, parameter optimization is to reach
To a kind of method of design object, by parameterizing design object, using optimization method, continuous adjusted design variable makes
Design result is obtained constantly close to the target value of parametrization, and the concept of hyperparameter optimization is generally used in present artificial intelligence
In learning type model, it is generally the case that some parameters for needing artificially to specify, these parameters are exactly hyper parameter, super for these
The optimization of parameter refers in the case where determining that model and parameter combination determine, sets the optimization range of each parameter, then right
These parameters are optimized to reach a satisfied training effect.
The realization of hyperparameter optimization at present, there are probelem in two aspects: be on the one hand hyperparameter optimization sampling and hold
It is serial implementation in capable process, the machine learning for nowadays artificial intelligence field, the hyper parameter in deep learning algorithm is excellent
In change, need to spend a large amount of resource and time due to carrying out primary parameter outcome evaluation, the serial drawback of hyperparameter optimization is just
It exposes out;On the other hand, be deficient in resources scheduling mechanism during super several optimizations, does not support rent in current realization more
Family, platform model submit hyperparameter optimization operation, this is disagreed for now a large amount of using the situation of large-scale cluster resource, lead
Applying, family is possible to carry out hyperparameter optimization using extremely limited resource, can not utilize cluster resource very well, additionally
Have one have to put forward a bit, currently existing scheme is mainly by cloud service provider Google etc. based on oneself system architecture
It is researched and developed, the scope of application is very narrow.
Mesos is a cluster resource scheduling system, in view of the above-mentioned problems, main target of the present invention and content concentrate on setting
It counts and realizes a kind of distributed hyperparameter optimization system and method based on Mesos, allow hyperparameter optimization operation in large size
It runs on distributed type assemblies, is asked with the resource contention under solving the use of a large amount of computing resources and multi-tenant scene of hyperparameter optimization
Topic.
Summary of the invention
The purpose of the present invention is to solve disadvantages existing in the prior art, and a kind of point based on Mesos proposed
Cloth hyperparameter optimization system and method.
To achieve the goals above, present invention employs following technical solutions:
A kind of distributed hyperparameter optimization system based on Mesos, including computation layer described in computation layer and dispatch layer with point
The hyperparameter optimization algorithm of cloth combines, and the computation layer is made of all kinds of optimization algorithms, for carrying out distributed hyper parameter
The sampling and generation calculating task of optimization, the dispatch layer is the specific implementation of a Mesos operation frame, is mainly responsible for resource
Distribution and execution calculating task;
The computation layer is formed by main application and from application, and using client/server, the main application is responsible for running hyper parameter
Optimization algorithm, it is described from application be responsible for operation intelligent algorithm training program, the dispatch layer by client, scheduler, hold
Row device is constituted, and for users to use, the scheduler is responsible for the scheduling of system resource to the client, and the actuator is responsible for calculating
The execution of layer task, is communicatively coupled between each section by network.
Preferably, only have a main application in the computation layer operational process, have one or more according to user demand
It is a to be calculated from application.
A kind of distributed hyperparameter optimization method based on Mesos, the described method comprises the following steps:
(1) the specified artificial intelligence program for needing to optimize of user, and the hyper parameter for needing to optimize is configured, it is submitted by system
Task gives Mesos cluster;
(2) the integrated dispatching algorithm of dispatch layer is according to the resource requirement of user's submission task and the operation conditions of task
It distributes cluster resource, starts computation layer various components;
(3) computation layer starts the main configuration progress hyperparameter optimization applied according to user first, and new to dispatch layer application
Resource run from application, be responsible for being calculated according to the parameter that main application provides from application, and calculated result is fed back to
Main application judges whether the stopping requirement for meeting user;
(4) finally when main application, which reaches user, to be stopped requiring, hyperparameter optimization result is returned into user, operation terminates.
Compared with the prior art, the beneficial effects of the present invention are:
A kind of distributed hyperparameter optimization system and method based on Mesos proposed by the invention can satisfy multi-tenant
It is used under mixed portion's scene of High-Performance Computing Cluster, improves the efficiency of hyperparameter optimization, the scope of application of the strategy is more extensive, right
Already existing Mesos group system is invasive low.
Detailed description of the invention
Fig. 1 is a kind of calculating layer frame of the distributed hyperparameter optimization system and method based on Mesos proposed by the present invention
Structure schematic diagram.
Fig. 2 is a kind of scheduling layer frame of the distributed hyperparameter optimization system and method based on Mesos proposed by the present invention
Structure schematic diagram.
Fig. 3 is a kind of operational method of the distributed hyperparameter optimization system and method based on Mesos proposed by the present invention
Flow diagram.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.
In the description of the present invention, it is to be understood that, term " on ", "lower", "front", "rear", "left", "right", "top",
The orientation or positional relationship of the instructions such as "bottom", "inner", "outside" is to be based on the orientation or positional relationship shown in the drawings, merely to just
In description the present invention and simplify description, rather than the device or element of indication or suggestion meaning must have a particular orientation, with
Specific orientation construction and operation, therefore be not considered as limiting the invention.
Referring to Fig.1-3, a kind of distributed hyperparameter optimization system based on Mesos, including computation layer and dispatch layer calculate
Layer is combined with distributed hyperparameter optimization algorithm, and computation layer is made of all kinds of optimization algorithms, for carrying out distributed super ginseng
The sampling and generation calculating task of number optimization, dispatch layer is the specific implementation of a Mesos operation frame, is mainly responsible for resource point
Match and execute calculating task;
Computation layer is formed by main application and from application, and using client/server, main application is responsible for running hyperparameter optimization algorithm,
It is responsible for operation intelligent algorithm training program from application, dispatch layer is made of client, scheduler, actuator, and client supplies
User uses, and scheduler is responsible for the scheduling of system resource, and actuator is responsible for the execution of computation layer task, passes through net between each section
Network is communicatively coupled.
Only have a main application in computation layer operational process, one or more is had according to user demand and is carried out from application
It calculates.
A kind of distributed hyperparameter optimization method based on Mesos, method the following steps are included:
(1) the specified artificial intelligence program for needing to optimize of user, and the hyper parameter for needing to optimize is configured, it is submitted by system
Task gives Mesos cluster;
(2) the integrated dispatching algorithm of dispatch layer is according to the resource requirement of user's submission task and the operation conditions of task
It distributes cluster resource, starts computation layer various components;
(3) computation layer starts the main configuration progress hyperparameter optimization applied according to user first, and new to dispatch layer application
Resource run from application, be responsible for being calculated according to the parameter that main application provides from application, and calculated result is fed back to
Main application judges whether the stopping requirement for meeting user;
(4) finally when main application, which reaches user, to be stopped requiring, hyperparameter optimization result is returned into user, operation terminates.
The foregoing is only a preferred embodiment of the present invention, but scope of protection of the present invention is not limited thereto,
Anyone skilled in the art in the technical scope disclosed by the present invention, according to the technique and scheme of the present invention and its
Inventive concept is subject to equivalent substitution or change, should be covered by the protection scope of the present invention.
Claims (3)
1. a kind of distributed hyperparameter optimization system based on Mesos, including computation layer and dispatch layer, which is characterized in that described
Computation layer is combined with distributed hyperparameter optimization algorithm, and the computation layer is made of all kinds of optimization algorithms, for being divided
The sampling and generation calculating task of cloth hyperparameter optimization, the dispatch layer is the specific implementation of a Mesos operation frame, main
It is responsible for resource allocation and executes calculating task;
The computation layer is formed by main application and from application, and using client/server, the main application is responsible for running hyperparameter optimization
Algorithm, described to be responsible for operation intelligent algorithm training program from application, the dispatch layer is by client, scheduler, actuator
It constitutes, for users to use, the scheduler is responsible for the scheduling of system resource to the client, and the actuator is responsible for computation layer and is appointed
The execution of business is communicatively coupled between each section by network.
2. a kind of distributed hyperparameter optimization system based on Mesos according to claim 1, which is characterized in that described
Only have a main application in computation layer operational process, one or more is had according to user demand and is calculated from application.
3. a kind of distributed hyperparameter optimization method based on Mesos, the described method comprises the following steps:
(1) the specified artificial intelligence program for needing to optimize of user, and the hyper parameter for needing to optimize is configured, task is submitted by system
Give Mesos cluster;
(2) the integrated dispatching algorithm of dispatch layer is its point according to the resource requirement of user's submission task and the operation conditions of task
With cluster resource, start computation layer various components;
(3) computation layer starts the main configuration progress hyperparameter optimization applied according to user, and the money new to dispatch layer application first
Source is run from application, is responsible for being calculated according to the parameter that main application provides from application, and calculated result is fed back to lead and is answered
With the stopping requirement for judging whether to meet user;
(4) finally when main application, which reaches user, to be stopped requiring, hyperparameter optimization result is returned into user, operation terminates.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910278557.4A CN110046046A (en) | 2019-04-09 | 2019-04-09 | A kind of distributed hyperparameter optimization system and method based on Mesos |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910278557.4A CN110046046A (en) | 2019-04-09 | 2019-04-09 | A kind of distributed hyperparameter optimization system and method based on Mesos |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110046046A true CN110046046A (en) | 2019-07-23 |
Family
ID=67276384
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910278557.4A Pending CN110046046A (en) | 2019-04-09 | 2019-04-09 | A kind of distributed hyperparameter optimization system and method based on Mesos |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110046046A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113608722A (en) * | 2021-07-31 | 2021-11-05 | 云南电网有限责任公司信息中心 | Algorithm packaging method based on distributed technology |
CN113712511A (en) * | 2021-09-03 | 2021-11-30 | 湖北理工学院 | Stable mode discrimination method for brain imaging fusion features |
US12067420B2 (en) | 2020-10-22 | 2024-08-20 | Hewlett Packard Enterprise Development Lp | Deep learning autotuning task optimization |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20170330105A1 (en) * | 2016-05-13 | 2017-11-16 | Cognitive Scale, Inc. | Ranking of Parse Options Using Machine Learning |
CN107463356A (en) * | 2017-08-17 | 2017-12-12 | 北京云纵信息技术有限公司 | The execution method and apparatus of flow of task |
CN108984257A (en) * | 2018-07-06 | 2018-12-11 | 无锡雪浪数制科技有限公司 | A kind of machine learning platform for supporting custom algorithm component |
-
2019
- 2019-04-09 CN CN201910278557.4A patent/CN110046046A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20170330105A1 (en) * | 2016-05-13 | 2017-11-16 | Cognitive Scale, Inc. | Ranking of Parse Options Using Machine Learning |
CN107463356A (en) * | 2017-08-17 | 2017-12-12 | 北京云纵信息技术有限公司 | The execution method and apparatus of flow of task |
CN108984257A (en) * | 2018-07-06 | 2018-12-11 | 无锡雪浪数制科技有限公司 | A kind of machine learning platform for supporting custom algorithm component |
Non-Patent Citations (1)
Title |
---|
李铄,陆忠华,孙永泽: "基于Mesos的分布式参数优化调度策略及系统设计", 《科研信息化技术与应用》 * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US12067420B2 (en) | 2020-10-22 | 2024-08-20 | Hewlett Packard Enterprise Development Lp | Deep learning autotuning task optimization |
CN113608722A (en) * | 2021-07-31 | 2021-11-05 | 云南电网有限责任公司信息中心 | Algorithm packaging method based on distributed technology |
CN113712511A (en) * | 2021-09-03 | 2021-11-30 | 湖北理工学院 | Stable mode discrimination method for brain imaging fusion features |
CN113712511B (en) * | 2021-09-03 | 2023-05-30 | 湖北理工学院 | Stable mode discrimination method for brain imaging fusion characteristics |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110046046A (en) | A kind of distributed hyperparameter optimization system and method based on Mesos | |
CN106776005A (en) | A kind of resource management system and method towards containerization application | |
Chen et al. | Optimal deadline scheduling with commitment | |
CN104636197B (en) | A kind of evaluation method of data center's virtual machine (vm) migration scheduling strategy | |
CN107045455A (en) | A kind of Docker Swarm cluster resource method for optimizing scheduling based on load estimation | |
CN107508901A (en) | Distributed data processing method, apparatus, server and system | |
CN110018817A (en) | The distributed operation method and device of data, storage medium and processor | |
CN107330516A (en) | Model parameter training method, apparatus and system | |
CN103699446A (en) | Quantum-behaved particle swarm optimization (QPSO) algorithm based multi-objective dynamic workflow scheduling method | |
CN109857534A (en) | A kind of intelligent task scheduling strategy training method based on Policy-Gradient Reinforcement Learning | |
CN108985709A (en) | Workflow management method towards more satellite data centers collaboration Remote Sensing Products production | |
CN110060765A (en) | A kind of standardization cloud radiotherapy planning method, storage medium and system | |
CN103164190A (en) | Rapid parallelization method of totally-distributed type watershed eco-hydrology model | |
CN107239337A (en) | The distribution of virtual resources and dispatching method and system | |
CN101661520A (en) | Synergetic design method for mechanical and electrical products | |
CN106570679A (en) | Participant capability based excitation method for crowdsourcing task of group | |
CN104965762B (en) | A kind of scheduling system towards hybrid task | |
CN101639788A (en) | Multi-core parallel method for continuous system simulation based on TBB threading building blocks | |
CN110347489A (en) | A kind of method for stream processing that the multicenter data collaborative based on Spark calculates | |
CN115085202A (en) | Power grid multi-region intelligent power collaborative optimization method, device, equipment and medium | |
Balla et al. | Reliability-aware: task scheduling in cloud computing using multi-agent reinforcement learning algorithm and neural fitted Q. | |
CN112488542A (en) | Intelligent building site material scheduling method and system based on machine learning | |
CN105528250B (en) | The evaluation and test of Multi-core computer system certainty and control method | |
CN112967148B (en) | Block chain consensus mechanism for intelligent Internet of things computing service | |
CN105426247B (en) | A kind of HLA federal members programming dispatching method |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190723 |