CN109828751A - Integrated machine learning algorithm library and unified programming framework - Google Patents
Integrated machine learning algorithm library and unified programming framework Download PDFInfo
- Publication number
- CN109828751A CN109828751A CN201910116872.7A CN201910116872A CN109828751A CN 109828751 A CN109828751 A CN 109828751A CN 201910116872 A CN201910116872 A CN 201910116872A CN 109828751 A CN109828751 A CN 109828751A
- Authority
- CN
- China
- Prior art keywords
- model
- learning
- component
- algorithm
- machine learning
- 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
Abstract
The present invention relates to a kind of integrated machine learning algorithm libraries and unified programming framework, including model learning component, model modification component and learning strategy component;The model learning component utilizes the logic of machine learning algorithm and the batch data developing algorithm model of fixed size based on Batch Processing;The logic that the model modification component is updated specific algorithm model using dynamic data flow based on Timely Processing;The learning strategy component is built-in to judge that input data is the strategy of batch data or flow data, and two scheduling model learning object, model modification component components carry out the logic of dynamic learning.The present invention can overcome the single treatment mode disadvantage of conventional machines learning system, facilitate the application of machine learning.
Description
Technical field
The present invention relates to field of computer technology, and in particular to a kind of integrated machine learning algorithm library and unified programming
Frame.
Background technique
Our life has been come into machine learning extensively, and application in daily life is more and more frequent.In fact, machine
Study just with we: it sends a notification message to us, including microblogging hot spot, the dynamic that the people of concern has just sent out, mobile phone
The traffic route searched out on figure, Spam filtering, the even safety of bank information.In recent years, machine learning had been sent out
The variation of essence is given birth to, nowadays machine only needs least manual intervention that can learn.However, due to the place of machine learning
Reason method is more and process is complicated, needs to design and safeguard simultaneously two sets of codes of batch processing and stream process in practical application, without one
It covers unified programming model and adapts to various concrete scenes, be unfavorable for the large-scale application of machine learning.
Summary of the invention
In view of this, the purpose of the present invention is to provide a kind of integrated machine learning algorithm libraries and unified programming frame
Frame can overcome the single treatment mode disadvantage of conventional machines learning system, facilitate the application of machine learning.
To achieve the above object, the present invention adopts the following technical scheme:
A kind of integrated machine learning algorithm library and unified programming framework, including model learning component, model modification component and
Learning strategy component;The model learning component utilizes machine learning algorithm and fixed size based on Batch Processing
The logic of batch data developing algorithm model;The model modification component is based on Timely Processing using dynamic number
The logic that specific algorithm model is updated according to stream;The learning strategy component is built-in to judge that input data is batch data
Or the strategy of flow data, two scheduling model learning object, model modification component components carry out the logic of dynamic learning.
Further, the model learning component includes a set of data conversion adapter.
Further, data set is enterprising in Hadoop, Spark or Flink platform when in use for the model learning component
Line number Data preprocess and algorithm model building, in preprocessing process, using data conversion adapter between different platform seamless biography
It passs, realizes cross-platform model construction, recycle the flow datas processing platforms such as Spark Structured Streaming or Flink
Realize streaming algorithm model modification.
Further, the model modification component passes through in Spark Structured Streaming, Flink streaming meter
It calculates and is realized on frame, the specified computing engines executed when operation, the algorithms library of component internal is based on stream process mode and provides point
The machine learning algorithms model modification strategy such as class, recurrence, cluster.
Further, the learning strategy component passes through scheduling model learning object and model modification based on workflow engine
Component completes batch machine learning or streaming machine learning task.
Further, the learning strategy component is provided with monitor and scheduler.
A kind of real-time learning control method in integrated machine learning algorithm library and unified programming framework, including following step
It is rapid:
Step S1: the scheduler of learning strategy component sends start command to model modification component;
Step S2: model modification component begins listening for flow data source, blocks itself after accumulating predetermined amount data;
Step S3: learning strategy component dispatcher starts monitor, and monitor sends start command to model learning component and carries out
Study;
Step S4: when model learning component completes algorithm model training, notice monitor algorithm model training has been completed, has been monitored
Device notification model updates component activation itself and continues to monitor flow data source, utilizes the data more new model of arrival.
A kind of batch learning control method in integrated machine learning algorithm library and unified programming framework, including following step
It is rapid:
Step S1: the scheduler of learning strategy component sends starting snoop command to model modification component;
Step S2: model modification component begins listening for the model data source of model learning component;
Step S3: the scheduler of learning strategy component starts monitor, monitor to model learning component send start command into
Row study;
Step S4: when model modification component, which listens to algorithm model training, to be completed, notice monitor starting next round study;
Step S5: practising by default wheel mathematics, when monitor reaches the threshold value of study number, monitor notification model update group
Part study terminates, and notification model learning object integrates final algorithm model.
Compared with the prior art, the invention has the following beneficial effects:
The integrated machine learning algorithm library of the present invention can overcome the single of conventional machines learning system with unified programming framework
Processing mode disadvantage facilitates the application of machine learning.
Detailed description of the invention
Fig. 1 is data flow diagram of the invention;
Fig. 2 is real-time learning control assembly logic interaction figure of the present invention;
Fig. 3 is batch study control assembly logic interaction figure of the invention.
Specific embodiment
The present invention will be further described with reference to the accompanying drawings and embodiments.
Fig. 1 is please referred to, the present invention provides a kind of integrated machine learning algorithm library and unified programming framework, unified to compile
Batch Processing is converted a kind of Timely Processing that can terminate by journey frame, to the batch of fixed size
Data are taken to be sampled for several times, and sampling learns immediately and construct single algorithm model when completing, by integrating final algorithm
Model;Unified programming framework includes model learning component, model modification component and learning strategy component in specific implementation;It is described
Model learning component utilizes the batch data developing algorithm of machine learning algorithm and fixed size based on Batch Processing
The logic of model;The model modification component is based on Timely Processing using dynamic data flow to specific algorithm
The logic that model is updated;The learning strategy component is built-in to judge that input data is the plan of batch data or flow data
Slightly, two scheduling model learning object, model modification component components carry out the logic of dynamic learning.
In the present embodiment, the model learning component includes a set of data conversion adapter, and the model learning component exists
Data set is constructed in the enterprising line number Data preprocess of Hadoop, Spark or Flink platform and algorithm model when use, pretreated
Cheng Zhong, using data converter between different platform seamless delivery, realize cross-platform model construction, recycle Spark
The flow datas processing platforms such as Structured Streaming or Flink realize streaming algorithm model modification.
In the present embodiment, the model modification component passes through in Spark Structured Streaming, Flink streaming
It is realized on Computational frame, the specified computing engines executed when operation, the algorithms library of component internal is based on stream process mode and provides
The machine learning algorithms model modification strategies such as classification, recurrence, cluster.
Learning strategy component described in the present embodiment is provided with monitor and scheduler.
Learning strategy component be based on Workflow(workflow) by scheduling model learning object and model modification component come
It completes batch machine learning or streaming machine learning task, scheduling process is as follows:
Start, learning strategy component starts Learning Process(learning process), a mark Attribute(is set and is belonged to
Property) " whether can end mark " and start a self-timing Activity(activity), if in the activity time data source data
Unchanged, setting flag property " whether can end mark " is " can terminate ";Otherwise input data is the data flow of variation, setting
Flag property " whether can end mark " is " not terminating ".
If flag property " whether can end mark " is " can terminate ", starting Batch Learning Process(batch
Learn sub-process), a Notification(notice is sent to model modification component), notification model more New Parent monitors model
The model output source of learning object;Then start second automatic Activity, start monitor and flag property " study is set
Whether number is more than threshold value " it is "No", monitor notification model learning object carries out sampling study;When model modification component detection
A Message(message is sent to monitor when completing to model training), monitor sends flag property " study time after receiving
Whether number is more than threshold value " it is used as Respond(response), and notification model learning object carries out sampling study next time;Work as monitoring
When device detects that study number reaches the threshold value of setting, monitor setting " whether study number is more than threshold value " is "Yes", model
More New Parent stops after receiving response;Last monitor notification model learning object integrates the algorithm model of all generations, study
Process terminates.
If flag property " whether can end mark " is " can not terminate ", learning strategy component starts Real-time
Learning Process(real-time learning sub-process), a Notification(notice is sent to model modification component), prison
Listen flow data source;Then start second automatic Activity, start monitor and whether flag property " starting to learn " is set
For "No";A Message is sent to monitor after the certain data volume of model modification component accumulation, and is waited to be answered;Monitor
Notification model learning object carries out first time study after receiving message;When monitor detects that algorithm model training is completed, if
Set that " whether starting to learn " is "Yes" and returns to model modification component as Respond;After model modification component receives response
It constantly reads flow data and updates algorithm model.
Referring to Fig. 2, a kind of real-time learning control method in integrated machine learning algorithm library and unified programming framework, packet
Include following steps:
Step S1: the scheduler of learning strategy component sends start command to model modification component;
Step S2: model modification component begins listening for flow data source, blocks itself after accumulating predetermined amount data;
Step S3: learning strategy component dispatcher starts monitor, and monitor sends start command to model learning component and carries out
Study;
Step S4: when model learning component completes algorithm model training, notice monitor model training has been completed, and monitor is logical
Perception model updates component activation itself and continues to monitor flow data source, utilizes the data more New Algorithm Model of arrival.
Referring to Fig. 3, a kind of batch learning control method in integrated machine learning algorithm library and unified programming framework, packet
Include following steps:
Step S1: the scheduler of learning strategy component sends starting snoop command to model modification component;
Step S2: model modification component begins listening for the model data source of model learning component;
Step S3: the scheduler of learning strategy component starts monitor, monitor to model learning component send start command into
Row study;
Step S4: when model modification component, which listens to algorithm model training, to be completed, notice monitor starting next round study;
Step S5: practising by default wheel mathematics, when monitor reaches the threshold value of study number, monitor notification model update group
Part study terminates, and notification model learning object integrates final algorithm model.
The foregoing is merely presently preferred embodiments of the present invention, all equivalent changes done according to scope of the present invention patent with
Modification, is all covered by the present invention.
Claims (8)
1. a kind of integrated machine learning algorithm library and unified programming framework, it is characterised in that: including model learning component, mould
Type more New Parent and learning strategy component;The model learning component is based on Batch Processing and utilizes machine learning algorithm
With the logic of the batch data developing algorithm model of fixed size;The model modification component is based on Timely Processing
The logic that specific algorithm model is updated using dynamic data flow;The built-in judgement input number of the learning strategy component
According to the strategy for being batch data or flow data, two scheduling model learning object, model modification component components carry out dynamics
The logic of habit.
2. integrated machine learning algorithm library according to claim 1 and unified programming framework, it is characterised in that: described
Model learning component includes a set of data conversion adapter.
3. integrated machine learning algorithm library according to claim 2 and unified programming framework, it is characterised in that: described
Model learning component when in use data set in the enterprising line number Data preprocess of Hadoop, Spark or Flink platform and algorithm model
Building in preprocessing process, bumpless transfer and is transmitted data set between different platform using data conversion adapter, is realized across flat
Platform model construction recycles Spark Structured Streaming, Flink flow data processing platform to realize streaming algorithm
Model modification.
4. integrated machine learning algorithm library according to claim 1 and unified programming framework, it is characterised in that: described
By realizing on Spark Structured Streaming, Flink streaming computing frame, when operation, refers to model modification component
Surely the computing engines executed, the algorithms library of component internal are based on stream process mode and provide classification, recurrence, cluster machine learning calculation
Method model modification strategy.
5. integrated machine learning algorithm library according to claim 1 and unified programming framework, it is characterised in that: described
Learning strategy component completes batch engineering by scheduling model learning object and model modification component based on workflow engine
It practises and streaming machine learning task.
6. integrated machine learning algorithm library according to claim 1 and unified programming framework, it is characterised in that: described
Learning strategy component is provided with monitor and scheduler.
7. the real-time learning control method in a kind of integrated machine learning algorithm library and unified programming framework, which is characterized in that
The following steps are included:
Step S1: the scheduler of learning strategy component sends start command to model modification component;
Step S2: model modification component begins listening for flow data source, blocks itself after accumulating predetermined amount data;
Step S3: learning strategy component dispatcher starts monitor, and monitor sends start command to model learning component and carries out
Study;
Step S4: when model learning component completes algorithm model training, notice monitor algorithm model training has been completed, has been monitored
Device notification model updates component activation itself and continues to monitor flow data source, utilizes the data more New Algorithm Model of arrival.
8. the batch learning control method in a kind of integrated machine learning algorithm library and unified programming framework, which is characterized in that
The following steps are included:
Step S1: the scheduler of learning strategy component sends starting snoop command to model modification component;
Step S2: model modification component begins listening for the model data source of model learning component;
Step S3: the scheduler of learning strategy component starts monitor, monitor to model learning component send start command into
Row study;
Step S4: when model modification component, which listens to algorithm model training, to be completed, notice monitor starting next round study;
Step S5: practising by default wheel mathematics, when monitor reaches the threshold value of study number, monitor notification model update group
Part study terminates, and notification model learning object integrates final algorithm model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910116872.7A CN109828751A (en) | 2019-02-15 | 2019-02-15 | Integrated machine learning algorithm library and unified programming framework |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910116872.7A CN109828751A (en) | 2019-02-15 | 2019-02-15 | Integrated machine learning algorithm library and unified programming framework |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109828751A true CN109828751A (en) | 2019-05-31 |
Family
ID=66863688
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910116872.7A Pending CN109828751A (en) | 2019-02-15 | 2019-02-15 | Integrated machine learning algorithm library and unified programming framework |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109828751A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110806932A (en) * | 2019-10-30 | 2020-02-18 | 广东博智林机器人有限公司 | Algorithm scheduling method and device |
CN112162734A (en) * | 2020-10-23 | 2021-01-01 | 福州大学 | Integrated machine learning algorithm library and unified programming framework (for deep learning) |
CN113420883A (en) * | 2021-06-28 | 2021-09-21 | 山东浪潮科学研究院有限公司 | Method and device for quantum programming frame to adapt to quantum computer |
Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20030207278A1 (en) * | 2002-04-25 | 2003-11-06 | Javed Khan | Methods for analyzing high dimensional data for classifying, diagnosing, prognosticating, and/or predicting diseases and other biological states |
EP2524966A1 (en) * | 2011-05-18 | 2012-11-21 | Rheinische Friedrich-Wilhelms-Universität Bonn | Molecular analysis of tuberculosis |
CN104021194A (en) * | 2014-06-13 | 2014-09-03 | 浪潮(北京)电子信息产业有限公司 | Mixed type processing system and method oriented to industry big data diversity application |
US20140298343A1 (en) * | 2013-03-26 | 2014-10-02 | Xerox Corporation | Method and system for scheduling allocation of tasks |
CN105677752A (en) * | 2015-12-30 | 2016-06-15 | 深圳先进技术研究院 | Streaming computing and batch computing combined processing system and method |
US20170063902A1 (en) * | 2015-08-31 | 2017-03-02 | Splunk Inc. | Interface Having Selectable, Interactive Views For Evaluating Potential Network Compromise |
CN107169575A (en) * | 2017-06-27 | 2017-09-15 | 北京天机数测数据科技有限公司 | A kind of modeling and method for visualizing machine learning training pattern |
CN107968840A (en) * | 2017-12-15 | 2018-04-27 | 华北电力大学(保定) | A kind of extensive power equipment monitoring, alarming Real-time Data Processing Method and system |
CN108615401A (en) * | 2018-06-11 | 2018-10-02 | 西安电子科技大学 | The non-homogeneous light parking situation recognition methods in interior based on deep learning |
-
2019
- 2019-02-15 CN CN201910116872.7A patent/CN109828751A/en active Pending
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20030207278A1 (en) * | 2002-04-25 | 2003-11-06 | Javed Khan | Methods for analyzing high dimensional data for classifying, diagnosing, prognosticating, and/or predicting diseases and other biological states |
EP2524966A1 (en) * | 2011-05-18 | 2012-11-21 | Rheinische Friedrich-Wilhelms-Universität Bonn | Molecular analysis of tuberculosis |
US20140298343A1 (en) * | 2013-03-26 | 2014-10-02 | Xerox Corporation | Method and system for scheduling allocation of tasks |
CN104021194A (en) * | 2014-06-13 | 2014-09-03 | 浪潮(北京)电子信息产业有限公司 | Mixed type processing system and method oriented to industry big data diversity application |
US20170063902A1 (en) * | 2015-08-31 | 2017-03-02 | Splunk Inc. | Interface Having Selectable, Interactive Views For Evaluating Potential Network Compromise |
CN105677752A (en) * | 2015-12-30 | 2016-06-15 | 深圳先进技术研究院 | Streaming computing and batch computing combined processing system and method |
CN107169575A (en) * | 2017-06-27 | 2017-09-15 | 北京天机数测数据科技有限公司 | A kind of modeling and method for visualizing machine learning training pattern |
CN107968840A (en) * | 2017-12-15 | 2018-04-27 | 华北电力大学(保定) | A kind of extensive power equipment monitoring, alarming Real-time Data Processing Method and system |
CN108615401A (en) * | 2018-06-11 | 2018-10-02 | 西安电子科技大学 | The non-homogeneous light parking situation recognition methods in interior based on deep learning |
Non-Patent Citations (1)
Title |
---|
GUO HONGQING; SU PEIYONG; GUO WENZHONG; GUO KUN: "Component-based Assembling Tool and Runtime Engine for the Machine Learning Process", 《2018 INTERNATIONAL CONFERENCE ON CLOUD COMPUTING, BIG DATA AND BLOCKCHAIN (ICCBB)》 * |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110806932A (en) * | 2019-10-30 | 2020-02-18 | 广东博智林机器人有限公司 | Algorithm scheduling method and device |
CN112162734A (en) * | 2020-10-23 | 2021-01-01 | 福州大学 | Integrated machine learning algorithm library and unified programming framework (for deep learning) |
CN112162734B (en) * | 2020-10-23 | 2022-03-08 | 福州大学 | Deep learning-oriented model generation method |
CN113420883A (en) * | 2021-06-28 | 2021-09-21 | 山东浪潮科学研究院有限公司 | Method and device for quantum programming frame to adapt to quantum computer |
CN113420883B (en) * | 2021-06-28 | 2022-11-22 | 山东浪潮科学研究院有限公司 | Method and equipment for quantum programming frame to adapt to quantum computer |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109828751A (en) | Integrated machine learning algorithm library and unified programming framework | |
KR101510860B1 (en) | Service Method and Server for Providing Application Comprehended User Intention | |
CN105306272B (en) | Information system fault scenes formation gathering method and system | |
CN105425970A (en) | Human-machine interaction method and device, and robot | |
CN208257859U (en) | A kind of river real-time remote monitoring system | |
JP2008199621A5 (en) | ||
CN101362334A (en) | Intelligent robot and operation method thereof | |
CN104408892B (en) | Safety-protection system and emergency system and dispensing device and reception device and method | |
WO2020147380A1 (en) | Human-computer interaction method and apparatus, computing device, and computer-readable storage medium | |
EP1929708A1 (en) | A distributed monitoring method | |
CN106847285A (en) | A kind of robot and its audio recognition method | |
CN101945353A (en) | Method and system for remote adjustment of system resources | |
CN113329480B (en) | Intention state management method, network element and system | |
CN113111115A (en) | Data information management system and method thereof | |
CN108597499B (en) | Voice processing method and voice processing device | |
CN106557362A (en) | It is applied to the task immigration method and system of intelligent robot | |
CN106570011A (en) | Distributed crawler URL seed distribution method, dispatching node, and grabbing node | |
CN111294241B (en) | Configurable Internet of things equipment access method | |
CN110610099A (en) | Financial risk intelligent early warning and wind control system based on FPGA hardware acceleration | |
CN210327644U (en) | Wisdom building site management system based on thing networking | |
CN106953714B (en) | Lower computer feedback data acquisition method and device for intelligent robot | |
CN112788592B (en) | Data transmission processing method for adding wake-up time | |
CN112968783B (en) | Low-power-consumption processing method and device based on transmitted data | |
CN209214761U (en) | A kind of noise monitoring system | |
CN109493863A (en) | A kind of intelligence awakening method and device |
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 |
Application publication date: 20190531 |
|
RJ01 | Rejection of invention patent application after publication |