CN110286587A - Method, server and the computer readable storage medium of implementation model iteration control - Google Patents

Method, server and the computer readable storage medium of implementation model iteration control Download PDF

Info

Publication number
CN110286587A
CN110286587A CN201910422133.0A CN201910422133A CN110286587A CN 110286587 A CN110286587 A CN 110286587A CN 201910422133 A CN201910422133 A CN 201910422133A CN 110286587 A CN110286587 A CN 110286587A
Authority
CN
China
Prior art keywords
model
line
time section
parameter
line time
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.)
Granted
Application number
CN201910422133.0A
Other languages
Chinese (zh)
Other versions
CN110286587B (en
Inventor
朱猛
范瑞
朱昀
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Ping An Bank Co Ltd
Original Assignee
Ping An Bank Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Ping An Bank Co Ltd filed Critical Ping An Bank Co Ltd
Priority to CN201910422133.0A priority Critical patent/CN110286587B/en
Publication of CN110286587A publication Critical patent/CN110286587A/en
Application granted granted Critical
Publication of CN110286587B publication Critical patent/CN110286587B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
    • G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
    • G05B13/04—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators
    • G05B13/042—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators in which a parameter or coefficient is automatically adjusted to optimise the performance

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Computation (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Automation & Control Theory (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Stored Programmes (AREA)

Abstract

The present invention relates to intelligent Decision Technology fields, disclose method, server and the storage medium of a kind of implementation model iteration control.This method comprises: server is after the iterative operation of Boot Model, obtain the preset kind model running parameter of current upper line model, and preset kind model running parameter corresponding with pre-stored each model on-line time section carries out parameter matching, calculate the corresponding model running parameter differences value of each model on-line time section, determine that upper line model corresponding with the currently upper most matched model on-line time section of line model is and the current upper most matched line model in history of line model, the current online model of iteration is analysed whether according to predetermined parser, to the current online model of iteration, it is then that current online model is offline, and the line model in history of online determination.This method combines real situation, has evaded manually-operated deviation, improves the convenience and accuracy of model iteration.

Description

Method, server and the computer readable storage medium of implementation model iteration control
Technical field
The present invention relates to intelligent Decision Technology field more particularly to a kind of methods of implementation model iteration control, server And computer readable storage medium.
Background technique
The common practice of model iteration is to do model training under offline data mode with single model in the industry at present, Artificial judgment or setting system thresholds Controlling model iteration, training process low efficiency, and selecting the model of iteration is all root Judge according to artificial experience, do not account for real contextual parameter, so that model iteration is often difficult to get a desired effect.In addition, For the process of model iteration usually under manual control, the response timeliness of iteration is unable to satisfy needs.Therefore, how to realize more Convenient, more accurate model iteration has become a technical problem urgently to be resolved.
Summary of the invention
In view of the foregoing, it is necessary to propose the method for implementation model iteration control a kind of, server and computer-readable Storage medium, primary purpose be to combine real situation, evades manually-operated deviation, improve model iteration convenience and Accuracy.
To achieve the above object, the present invention proposes that a kind of server of implementation model iteration control, the server include depositing Reservoir and processor are stored with model iteration control program on the memory, and the model iteration control program is by the place Reason device realizes following steps when executing:
The iterative operation of S11, start by set date model, alternatively, after receiving the iterative instruction for model, Boot Model Iterative operation；
S12, after the iterative operation of Boot Model, obtain the preset kind model running parameter of current upper line model, root It is closed according to the mapping of pre-stored model on-line time section, the preset kind model running parameter of online model, upper line model Coefficient evidence, the preset kind model running parameter that will acquire preset kind model corresponding with each model on-line time section respectively Operating parameter carries out parameter matching, obtains the corresponding match parameter group of each model on-line time section, calculates separately out each Parameter difference or parameter difference absolute value with parameter group；
S13, respectively by the parameter difference or parameter difference of the corresponding each match parameter group of each model on-line time section It is worth absolute value to substitute into predetermined calculation formula, calculates separately out the corresponding model running ginseng of each model on-line time section Number difference value；
S14, according to the corresponding model running parameter differences value of calculated each model on-line time section, determine and institute The currently upper most matched model on-line time section of line model is stated, the corresponding upper line model of model on-line time section determined is true It is set to and the currently upper most matched line model in history of line model；
S15, it is analysed whether to use the determining iteration of line model in history currently online according to predetermined parser Model, it is to current online model described in iteration, then the current online model is offline, and the online determination Line model in history.
Optionally, the predetermined calculation formula are as follows: f (Ci)=a1Ci1+a2Ci2+……+amCim, in formula, f (Ci) Represent the corresponding model running parameter differences value of the i-th on-line time section, Ci1For first kind match parameter group parameter difference or Parameter difference absolute value, a1It is corresponding for the parameter difference or parameter difference absolute value of predetermined first kind match parameter group Weighted value, Ci2For the parameter difference or parameter difference absolute value of the second class match parameter group, a2It is predetermined second The parameter difference or the corresponding weighted value of parameter difference absolute value of class match parameter group, CimFor the ginseng of m class match parameter group Number difference or parameter difference absolute value, amFor the parameter difference or parameter difference of predetermined m class match parameter group The corresponding weighted value of absolute value, i, m are positive integer.
Optionally, described according to the corresponding model running parameter differences value of calculated each model on-line time section, really It makes and the currently upper most matched model on-line time section of line model, comprising:
If the corresponding model running parameter differences value of all model on-line time sections is all larger than or is equal to preset threshold, Determining does not terminate with the current above most matched model on-line time section of line model, the process；Or
If the corresponding model running parameter differences value of an only model on-line time section is less than preset threshold, it is determined that should The most matched model on-line time section of line model is currently gone up described in model on-line time Duan Weiyu；Or
It, will be multiple if there is the corresponding model running parameter differences value of multiple model on-line time sections to be less than preset threshold In the corresponding model running parameter differences value of model on-line time section on the corresponding model of the smallest model running parameter differences value The line period currently goes up the most matched model on-line time section of line model as with described.
Optionally, described to analyse whether to be worked as with the determining iteration of line model in history according to predetermined parser Preceding online model, comprising:
It is trained using in history line model of the predetermined model training data to the determination, generation trains Line model in history；
Using predetermined model verify data, the trained line model in history and described current is separately verified The accuracy rate of upper line model；
If the accuracy rate of the trained line model in history is higher, it is determined that with the line model in history of the determination Current online model described in iteration.
To achieve the above object, the present invention also proposes that a kind of server of implementation model iteration control, the server include Memory and processor are stored with model iteration control program on the memory, and the model iteration control program is described Processor realizes following steps when executing:
The iterative operation of S21, start by set date model, alternatively, after receiving the iterative instruction for model, Boot Model Iterative operation；
S22, after the iterative operation of Boot Model, select application environment one by one, after selecting an application environment, obtain The preset kind model running parameter of the current upper line model of the running environment, it is corresponding pre-stored according to the application environment The mapping relations data of model on-line time section, the preset kind model running parameter of online model, upper line model, will acquire Preset kind model running parameter corresponding with each model on-line time section preset kind model running parameter carries out respectively Parameter matching, obtains the corresponding match parameter group of each model on-line time section, calculates separately out the ginseng of each match parameter group Number difference or parameter difference absolute value；
S23, respectively by the parameter difference or parameter difference of the corresponding each match parameter group of each model on-line time section It is worth absolute value to substitute into predetermined calculation formula, calculates separately out the corresponding model running ginseng of each model on-line time section Number difference value；
S24, according to the corresponding model running parameter differences value of calculated each model on-line time section, determine and institute The currently upper most matched model on-line time section of line model is stated, the corresponding upper line model of model on-line time section determined is true It is set to and the currently upper most matched line model in history of line model；
S25, it is analysed whether to use the determining iteration of line model in history currently online according to predetermined parser Model, it is to current online model described in iteration, then the current online model is offline, and the online determination Line model in history.
Optionally, described according to the corresponding model running parameter differences value of calculated each model on-line time section, really It makes and the currently upper most matched model on-line time section of line model, comprising:
If the corresponding model running parameter differences value of all model on-line time sections is all larger than or is equal to preset threshold, Determining does not terminate with the current above most matched model on-line time section of line model, the process；Or
If the corresponding model running parameter differences value of an only model on-line time section is less than preset threshold, it is determined that should The most matched model on-line time section of line model is currently gone up described in model on-line time Duan Weiyu；Or
It, will be multiple if there is the corresponding model running parameter differences value of multiple model on-line time sections to be less than preset threshold In the corresponding model running parameter differences value of model on-line time section on the corresponding model of the smallest model running parameter differences value The line period currently goes up the most matched model on-line time section of line model as with described.
Optionally, described to analyse whether to be worked as with the determining iteration of line model in history according to predetermined parser Preceding online model, comprising:
It is trained using in history line model of the predetermined model training data to the determination, generation trains Line model in history；
Using predetermined model verify data, the trained line model in history and described current is separately verified The accuracy rate of upper line model；
If the accuracy rate of the trained line model in history is higher, it is determined that with the line model in history of the determination Current online model described in iteration.
To achieve the above object, a kind of method that the present invention proposes implementation model iteration control, this method comprises:
The iterative operation of S11, server start by set date model, alternatively, server is receiving the iterative instruction for model Afterwards, the iterative operation of Boot Model；
S12, after the iterative operation of Boot Model, the server obtains the preset kind model fortune of current upper line model Row parameter, model on-line time section according to the pre-stored data, the preset kind model running parameter of online model, upper line model Mapping relations data, the preset kind model running parameter that will acquire is corresponding with each model on-line time section default respectively Type model operating parameter carries out parameter matching, obtains the corresponding match parameter group of each model on-line time section, calculates separately The parameter difference or parameter difference absolute value of each match parameter group out；
S13, the server are respectively by the parameter difference of the corresponding each match parameter group of each model on-line time section Or parameter difference absolute value substitutes into predetermined calculation formula, and it is corresponding to calculate separately out each model on-line time section Model running parameter differences value；
S14, the server according to the corresponding model running parameter differences value of calculated each model on-line time section, Determine with the currently upper most matched model on-line time section of line model, the model on-line time section determined is corresponding Upper line model is determined as and the currently upper most matched line model in history of line model；
S15, the server analyse whether to be changed with determining line model in history according to predetermined parser Generation current online model, it is to current online model described in iteration, then the current online model is offline and online The line model in history of the determination.
A kind of method that the present invention also proposes implementation model iteration control, this method comprises:
The iterative operation of S21, server start by set date model, alternatively, server is receiving the iterative instruction for model Afterwards, the iterative operation of Boot Model；
S22, after the iterative operation of Boot Model, the server selects application environment one by one, selection one application After environment, the preset kind model running parameter of the current upper line model of the running environment is obtained, it is corresponding according to the application environment Pre-stored model on-line time section, the preset kind model running parameter of online model, upper line model mapping close Coefficient evidence, the preset kind model running parameter that will acquire preset kind model corresponding with each model on-line time section respectively Operating parameter carries out parameter matching, obtains the corresponding match parameter group of each model on-line time section, calculates separately out each Parameter difference or parameter difference absolute value with parameter group；
S23, the server are respectively by the parameter difference of the corresponding each match parameter group of each model on-line time section Or parameter difference absolute value substitutes into predetermined calculation formula, and it is corresponding to calculate separately out each model on-line time section Model running parameter differences value；
S24, the server according to the corresponding model running parameter differences value of calculated each model on-line time section, Determine with the currently upper most matched model on-line time section of line model, the model on-line time section determined is corresponding Upper line model is determined as and the currently upper most matched line model in history of line model；
S25, the server analyse whether to be changed with determining line model in history according to predetermined parser Generation current online model, it is to current online model described in iteration, then the current online model is offline and online The line model in history of the determination.
The present invention also proposes a kind of computer readable storage medium, and storage model changes on the computer readable storage medium Generation control program, the model iteration control program can be executed by one or more processor, to realize any of the above-described institute The step of stating the method for implementation model iteration control.
Compared with the prior art, the present invention passes through the preset kind model running parameter of the current upper line model that will acquire and each The corresponding preset kind model running parameter of a model on-line time section carries out parameter matching, calculates each model on-line time The corresponding model running parameter differences value of section, and then the determining and current upper most matched model of line model, effectively combine reality Situation has evaded manually-operated deviation, improves the convenience and accuracy of model iteration.
Detailed description of the invention
Fig. 1 is the hardware structure diagram of one embodiment of server of implementation model iteration control of the present invention.
Fig. 2 is the functional block diagram of 10 1 embodiment of model iteration control program in Fig. 1.
Fig. 3 is the flow chart of the method first embodiment of implementation model iteration control of the present invention.
Fig. 4 is the flow chart of the method second embodiment of implementation model iteration control of the present invention.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right The present invention is further elaborated.It should be appreciated that described herein, specific examples are only used to explain the present invention, not For limiting the present invention.Based on the embodiments of the present invention, those of ordinary skill in the art are not before making creative work Every other embodiment obtained is put, shall fall within the protection scope of the present invention.
It should be noted that the description for being related to " first ", " second " etc. in the present invention is used for description purposes only, and cannot It is interpreted as its relative importance of indication or suggestion or implicitly indicates the quantity of indicated technical characteristic.Define as a result, " the One ", the feature of " second " can explicitly or implicitly include at least one of the features.In addition, the skill between each embodiment Art scheme can be combined with each other, but must be based on can be realized by those of ordinary skill in the art, when technical solution Will be understood that the combination of this technical solution is not present in conjunction with there is conflicting or cannot achieve when, also not the present invention claims Protection scope within.
As shown in Figure 1, the hardware structure diagram of one embodiment of server for implementation model iteration control of the present invention.In this reality It applies in example, server 1 includes memory 11 and processor 12, is stored with model iteration control program 10, institute in the memory 11 Stating model iteration control program 10 can be executed by the processor 12.
Memory 11 includes the readable storage medium storing program for executing of memory and at least one type.The operation for inside saving as server 1 provides Caching；Readable storage medium storing program for executing can for such as flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory), with Machine access memory (RAM), static random-access memory (SRAM), read-only memory (ROM), electrically erasable are read-only The non-volatile memories of memory (EEPROM), programmable read only memory (PROM), magnetic storage, disk, CD etc. are situated between Matter.In some embodiments, readable storage medium storing program for executing can be the internal storage unit of server 1, such as the server 1 is hard Disk；In further embodiments, which is also possible to the External memory equipment of server 1, such as services The plug-in type hard disk being equipped on device 1, intelligent memory card (Smart Media Card, SMC), secure digital (Secure Digital, SD) card, flash card (Flash Card) etc..In the present embodiment, the readable storage medium storing program for executing of memory 11 is commonly used in Storage is installed on the model iteration in the operating system and types of applications software of server 1, such as storage one embodiment of the invention Control the code etc. of program 10.It has exported or will export each in addition, memory 11 can be also used for temporarily storing Class data.
Processor 12 can be in some embodiments central processing unit (Central Processing Unit, CPU), Controller, microcontroller, microprocessor or other data processing chips.The processor 12 is commonly used in the control server 1 Overall operation, such as execute with other equipment carry out data interaction or communication it is relevant control and handle etc..The present embodiment In, the processor 12 is for running the program code stored in the memory 11 or processing data, such as moving model Iteration control program 10 etc..
Optionally, the server 1 can also include user interface, user interface may include display (Display), Input unit such as keyboard (Keyboard), optional user interface can also include standard wireline interface and wireless interface.It can Choosing, in some embodiments, display can be light-emitting diode display, liquid crystal display, touch-control liquid crystal display and OLED (Organic Light-Emitting Diode, Organic Light Emitting Diode) touches device etc..Wherein, display can also be appropriate Referred to as display screen or display unit, for showing the information handled in the server 1 and for showing visual user circle Face.
Fig. 1 illustrates only the server 1 with component 11-12 and model iteration control program 10, those skilled in the art Member may include less or more than illustrating it is understood that structure shown in fig. 1 does not constitute the restriction to server 1 More components perhaps combines certain components or different component layouts.
In the first embodiment of the present invention, it is realized when the model iteration control program 10 is executed by the processor 12 Following steps:
The iterative operation of S11, start by set date model, alternatively, after receiving the iterative instruction for model, Boot Model Iterative operation；
The model can be convolutional neural networks model, supporting vector machine model, Random Forest model etc., the iteration The purpose of operation is used for Optimized model, and the iterative operation includes but is not limited to adjust version, adjustment model parameter etc..
S12, after the iterative operation of Boot Model, obtain the preset kind model running parameter of current upper line model, root It is closed according to the mapping of pre-stored model on-line time section, the preset kind model running parameter of online model, upper line model Coefficient evidence, the preset kind model running parameter that will acquire preset kind model corresponding with each model on-line time section respectively Operating parameter carries out parameter matching, obtains the corresponding match parameter group of each model on-line time section, calculates separately out each Parameter difference or parameter difference absolute value with parameter group；
Specifically, be stored in advance the online model of each on-line time section built according to time series, upper line model it is pre- If the mapping relations data set of Type model operating parameter, the preset kind model running parameter can be preset time internal model Type analysis accuracy rate (for example, model analysis accuracy rate in nearest 24 hours), unit time model are averaged call number (example Such as, model per minute is averaged call number) etc..For example, the corresponding volume of the 1st model on-line time section in pre-stored data set Model analysis accuracy rate is Y in nearest 24 hours of product neural network model11, unit time model be averaged call number be Y12, Model analysis accuracy rate is Y in nearest 24 hours of the corresponding supporting vector machine model of 2nd model on-line time section21, unit when Between model be averaged call number be Y22.After the iterative operation of Boot Model, the corresponding current online mould of current slot n is obtained Model analysis accuracy rate Y in nearest 24 hours of type convolutional neural networks modeln1It is averaged call number with unit time model Yn2, after carrying out parameter matching, model analysis in nearest 24 hours of the corresponding convolution time network model of the 1st model on-line time section Accuracy rate match parameter group is (Yn1, Y11), unit time model call number parameter group be (Yn2, Y12), when the 2nd model is online Between in the corresponding supporting vector machine model of section nearest 24 hours model analysis accuracy rate match parameter group be (Yn1, Y21), unit when Between model call number parameter group be (Yn2, Y22), the corresponding convolution time network model nearest 24 of the 1st model on-line time section Model analysis accuracy rate parameter difference is Y in hourn1Subtract Y11Difference.
S13, respectively by the parameter difference or parameter difference of the corresponding each match parameter group of each model on-line time section It is worth absolute value to substitute into predetermined calculation formula, calculates separately out the corresponding model running ginseng of each model on-line time section Number difference value；
Optionally, the predetermined calculation formula are as follows: f (Ci)=a1Ci1+a2Ci2+……+amCim, in formula, f (Ci) Represent the corresponding model running parameter differences value of the i-th on-line time section, Ci1For first kind match parameter group parameter difference or Parameter difference absolute value, a1It is corresponding for the parameter difference or parameter difference absolute value of predetermined first kind match parameter group Weighted value, Ci2For the parameter difference or parameter difference absolute value of the second class match parameter group, a2It is predetermined second The parameter difference or the corresponding weighted value of parameter difference absolute value of class match parameter group, CimFor the ginseng of m class match parameter group Number difference or parameter difference absolute value, amFor the parameter difference or parameter difference of predetermined m class match parameter group The corresponding weighted value of absolute value, i, m are positive integer.
Only to choose two class preset kind model running parameters (for example, model analysis accuracy rate, every point in nearest 24 hours Clock is averaged call number) for, model point in nearest 24 hours of the corresponding convolution time network model of the 1st model on-line time section Analyse accuracy rate parameter difference C11Are as follows: C11=Yn1-Y11, the corresponding convolution time network model of the 1st model on-line time section is per minute Average call number parameter difference C12Are as follows: C12=Yn2-Y12, the corresponding model running parameter differences value of the 1st model on-line time section f(C1)=a1C11+a2C12=a1(Yn1-Y11)+a2(Yn2-Y12), a1It is accurate for model analysis in predetermined nearest 24 hours The corresponding weighted value of rate parameter difference, a2For the predetermined corresponding weighted value of call number parameter difference that is averaged per minute.
S14, according to the corresponding model running parameter differences value of calculated each model on-line time section, determine and institute The currently upper most matched model on-line time section of line model is stated, the corresponding upper line model of model on-line time section determined is true It is set to and the currently upper most matched line model in history of line model；
Optionally, described according to the corresponding model running parameter differences value of calculated each model on-line time section, really It makes and the currently upper most matched model on-line time section of line model, comprising:
If the corresponding model running parameter differences value of all model on-line time sections is all larger than or is equal to preset threshold, Determining does not terminate with the current above most matched model on-line time section of line model, the process；Or
If the corresponding model running parameter differences value of an only model on-line time section is less than preset threshold, it is determined that should The most matched model on-line time section of line model is currently gone up described in model on-line time Duan Weiyu；Or
It, will be multiple if there is the corresponding model running parameter differences value of multiple model on-line time sections to be less than preset threshold In the corresponding model running parameter differences value of model on-line time section on the corresponding model of the smallest model running parameter differences value The line period currently goes up the most matched model on-line time section of line model as with described.
For example, setting preset threshold as 10, if the corresponding model running parameter differences value of all model on-line time sections is equal More than or equal to 10, then the most matched model on-line time section of line model is not gone up with current；If only the 2nd model on-line time The corresponding model running parameter differences value of section is less than 10, it is determined that currently upper line model is most by the 2nd model on-line time Duan Weiyu The corresponding supporting vector machine model of 2nd model on-line time section is determined as and current online mould by the model on-line time section matched The most matched line model in history of type；If the corresponding model running parameter differences value of the 1st model on-line time section is the 6, the 2nd model The corresponding model running parameter differences value of on-line time section is 3, other corresponding model running parameter differences of model on-line time section Different value is all larger than 10, then the corresponding 2nd model on-line time Duan Weiyu of preference pattern operating parameter difference value the smallest 3 it is current on The most matched model on-line time section of line model；If the corresponding model running parameter differences value of the 1st model on-line time section is 6, The corresponding model running parameter differences value of 2nd model on-line time section is the corresponding model running of the 3, the 5th model on-line time section Parameter differences value is also 3, other corresponding model running parameter differences values of model on-line time section are all larger than 10, then transports in model A work is randomly choosed in the corresponding 2nd model on-line time section of row parameter differences value the smallest 3 and the 5th model on-line time section To go up the most matched model on-line time section of line model with current.
S15, it is analysed whether to use the determining iteration of line model in history currently online according to predetermined parser Model, it is to current online model described in iteration, then the current online model is offline, and the online determination Line model in history.
Optionally, described to analyse whether to be worked as with the determining iteration of line model in history according to predetermined parser Preceding online model, comprising:
It is trained using in history line model of the predetermined model training data to the determination, generation trains Line model in history；
Using predetermined model verify data, the trained line model in history and described current is separately verified The accuracy rate of upper line model；
If the accuracy rate of the trained line model in history is higher, it is determined that with the line model in history of the determination Current online model described in iteration.
In the second embodiment of the present invention, the model iteration control program 10 is realized when being executed by the processor 12 Following steps:
The iterative operation of S21, start by set date model, alternatively, after receiving the iterative instruction for model, Boot Model Iterative operation；
The model can be convolutional neural networks model, supporting vector machine model, Random Forest model etc., the iteration The purpose of operation is used for Optimized model, and the iterative operation includes but is not limited to adjust version, adjustment model parameter etc..
S22, after the iterative operation of Boot Model, select application environment one by one, after selecting an application environment, obtain The preset kind model running parameter of the current upper line model of the running environment, it is corresponding pre-stored according to the application environment The mapping relations data of model on-line time section, the preset kind model running parameter of online model, upper line model, will acquire Preset kind model running parameter corresponding with each model on-line time section preset kind model running parameter carries out respectively Parameter matching, obtains the corresponding match parameter group of each model on-line time section, calculates separately out the ginseng of each match parameter group Number difference or parameter difference absolute value；
Specifically, be stored in advance the online model of each application environment, each on-line time section built according to time series, on The mapping relations data set of the preset kind model running parameter of line model, the application environment include recognition of face, Yong Hufen Class, big data analysis etc., the preset kind model running parameter can be model analysis accuracy rate in preset time (for example, Model analysis accuracy rate in nearest 24 hours), unit time model be averaged call number (for example, per minute model averagely adjust With number) etc..For example, in pre-stored data set face recognition application environment the corresponding volume of the 1st model on-line time section Model analysis accuracy rate is Y in nearest 24 hours of product neural network model11, unit time model be averaged call number be Y12, Model analysis in nearest 24 hours of the corresponding supporting vector machine model of the 2nd model on-line time section of face recognition application environment Accuracy rate is Y21, unit time model be averaged call number be Y22.After the iterative operation of Boot Model, selection application one by one Environment, if having selected face recognition application environment, the current slot n for obtaining face recognition application environment is corresponding current online Model analysis accuracy rate Y in nearest 24 hours of model convolutional neural networks modeln1It is averagely called with unit time model secondary Number Yn2, after carrying out parameter matching, the corresponding convolution time network model of the 1st model on-line time section of face recognition application environment Model analysis accuracy rate match parameter group is (Y in nearest 24 hoursn1, Y11), unit time model call number parameter group be (Yn2, Y12), in the corresponding supporting vector machine model of the 2nd model on-line time section nearest 24 hours of face recognition application environment Model analysis accuracy rate match parameter group is (Yn1, Y21), unit time model call number parameter group be (Yn2, Y22), face is known Model analysis accuracy rate in the corresponding convolution time network model of the 1st model on-line time section nearest 24 hours of other application environment Parameter difference is Yn1Subtract Y11Difference.
S23, respectively by the parameter difference or parameter difference of the corresponding each match parameter group of each model on-line time section It is worth absolute value to substitute into predetermined calculation formula, calculates separately out the corresponding model running ginseng of each model on-line time section Number difference value；
Optionally, the predetermined calculation formula are as follows: f (Ci)=a1Ci1+a2Ci2+……+amCim, in formula, f (Ci) Represent the corresponding model running parameter differences value of the i-th on-line time section, Ci1For first kind match parameter group parameter difference or Parameter difference absolute value, a1It is corresponding for the parameter difference or parameter difference absolute value of predetermined first kind match parameter group Weighted value, Ci2For the parameter difference or parameter difference absolute value of the second class match parameter group, a2It is predetermined second The parameter difference or the corresponding weighted value of parameter difference absolute value of class match parameter group, CimFor the ginseng of m class match parameter group Number difference or parameter difference absolute value, amFor the parameter difference or parameter difference of predetermined m class match parameter group The corresponding weighted value of absolute value, i, m are positive integer.
If application environment is recognition of face, only to choose two class preset kind model running parameters (for example, nearest 24 hours Interior model analysis accuracy rate, be averaged call number per minute) for, the 1st model on-line time section of face recognition application environment Model analysis accuracy rate parameter difference C in nearest 24 hours of corresponding convolution time network model11Are as follows: C11=Yn1-Y11, face Identify application environment the corresponding convolution time network model of the 1st model on-line time section per minute be averaged call number parameter difference Value C12Are as follows: C12=Yn2-Y12, the corresponding model running parameter differences value of the 1st model on-line time section of face recognition application environment f(C1)=a1C11+a2C12=a1(Yn1-Y11)+a2(Yn2-Y12), a1It is accurate for model analysis in predetermined nearest 24 hours The corresponding weighted value of rate parameter difference, a2For the predetermined corresponding weighted value of call number parameter difference that is averaged per minute.
S24, according to the corresponding model running parameter differences value of calculated each model on-line time section, determine and institute The currently upper most matched model on-line time section of line model is stated, the corresponding upper line model of model on-line time section determined is true It is set to and the currently upper most matched line model in history of line model；
Optionally, described according to the corresponding model running parameter differences value of calculated each model on-line time section, really It makes and the currently upper most matched model on-line time section of line model, comprising:
If the corresponding model running parameter differences value of all model on-line time sections is all larger than or is equal to preset threshold, Determining does not terminate with the current above most matched model on-line time section of line model, the process；Or
If the corresponding model running parameter differences value of an only model on-line time section is less than preset threshold, it is determined that should The most matched model on-line time section of line model is currently gone up described in model on-line time Duan Weiyu；Or
It, will be multiple if there is the corresponding model running parameter differences value of multiple model on-line time sections to be less than preset threshold In the corresponding model running parameter differences value of model on-line time section on the corresponding model of the smallest model running parameter differences value The line period currently goes up the most matched model on-line time section of line model as with described.
For example, setting preset threshold as 10, if the corresponding mould of all model on-line time sections of face recognition application environment Type operating parameter difference value is all larger than or is equal to 10, then does not go up the most matched model on-line time section of line model with current； If only the corresponding model running parameter differences value of the 2nd model on-line time section of face recognition application environment is less than 10, it is determined that 2nd model on-line time Duan Weiyu of face recognition application environment currently goes up the most matched model on-line time section of line model, will The corresponding supporting vector machine model of the 2nd model on-line time section of face recognition application environment is determined as with currently upper line model most Matched line model in history；If the corresponding model running parameter difference of the 1st model on-line time section of face recognition application environment Different value is 6, and the corresponding model running parameter differences value of the 2nd model on-line time section of face recognition application environment is 3, and face is known Other corresponding model running parameter differences values of model on-line time section of other application environment are all larger than 10, then select recognition of face The corresponding 2nd model on-line time Duan Weiyu of model running parameter differences value the smallest 3 of application environment currently goes up line model most Matched model on-line time section；If the corresponding model running parameter of the 1st model on-line time section of face recognition application environment Difference value is 6, and the corresponding model running parameter differences value of the 2nd model on-line time section of face recognition application environment is 3, face The corresponding model running parameter differences value of the 5th model on-line time section for identifying application environment is also 3, face recognition application environment Other corresponding model running parameter differences values of model on-line time section be all larger than 10, then in the mould of face recognition application environment One is randomly choosed in the corresponding 2nd model on-line time section of type operating parameter difference value the smallest 3 and the 5th model on-line time section It is a as with the current upper most matched model on-line time section of line model.
S25, it is analysed whether to use the determining iteration of line model in history currently online according to predetermined parser Model, it is to current online model described in iteration, then the current online model is offline, and the online determination Line model in history.
Optionally, described to analyse whether to be worked as with the determining iteration of line model in history according to predetermined parser Preceding online model, comprising:
It is trained using in history line model of the predetermined model training data to the determination, generation trains Line model in history；
Using predetermined model verify data, the trained line model in history and described current is separately verified The accuracy rate of upper line model；
If the accuracy rate of the trained line model in history is higher, it is determined that with the line model in history of the determination Current online model described in iteration.
As shown in Fig. 2, for the functional block diagram of 10 1 embodiment of model iteration control program in Fig. 1.In the present embodiment In, model iteration control program 10 includes computing module 110 and execution module 120.
In the first embodiment of the present invention:
The computing module 110, for the iterative operation of start by set date model, alternatively, receiving the iteration for model After instruction, the iterative operation of Boot Model；After the iterative operation of Boot Model, the preset kind mould of current upper line model is obtained Type operating parameter, model on-line time section according to the pre-stored data, online model, upper line model preset kind model running The mapping relations data of parameter, the preset kind model running parameter that will acquire are corresponding with each model on-line time section respectively Preset kind model running parameter carries out parameter matching, obtains the corresponding match parameter group of each model on-line time section, respectively Calculate the parameter difference or parameter difference absolute value of each match parameter group；It is respectively that each model on-line time section is corresponding Each match parameter group parameter difference or parameter difference absolute value substitute into predetermined calculation formula, calculate separately The corresponding model running parameter differences value of each model on-line time section out.
The model can be convolutional neural networks model, supporting vector machine model, Random Forest model etc., the iteration The purpose of operation is used for Optimized model, and the iterative operation includes but is not limited to adjust version, adjustment model parameter etc..
Specifically, be stored in advance the online model of each on-line time section built according to time series, upper line model it is pre- If the mapping relations data set of Type model operating parameter, the preset kind model running parameter can be preset time internal model Type analysis accuracy rate (for example, model analysis accuracy rate in nearest 24 hours), unit time model are averaged call number (example Such as, model per minute is averaged call number) etc..For example, the corresponding volume of the 1st model on-line time section in pre-stored data set Model analysis accuracy rate is Y in nearest 24 hours of product neural network model11, unit time model be averaged call number be Y12, Model analysis accuracy rate is Y in nearest 24 hours of the corresponding supporting vector machine model of 2nd model on-line time section21, unit when Between model be averaged call number be Y22.After the iterative operation of Boot Model, the corresponding current online mould of current slot n is obtained Model analysis accuracy rate Y in nearest 24 hours of type convolutional neural networks modeln1It is averaged call number with unit time model Yn2, after carrying out parameter matching, model analysis in nearest 24 hours of the corresponding convolution time network model of the 1st model on-line time section Accuracy rate match parameter group is (Yn1, Y11), unit time model call number parameter group be (Yn2, Y12), when the 2nd model is online Between in the corresponding supporting vector machine model of section nearest 24 hours model analysis accuracy rate match parameter group be (Yn1, Y21), unit when Between model call number parameter group be (Yn2, Y22), the corresponding convolution time network model nearest 24 of the 1st model on-line time section Model analysis accuracy rate parameter difference is Y in hourn1Subtract Y11Difference.Then each model on-line time section is corresponding The parameter difference or parameter difference absolute value of each match parameter group substitute into predetermined calculation formula, can calculate The corresponding model running parameter differences value of each model on-line time section.
Optionally, the predetermined calculation formula are as follows: f (Ci)=a1Ci1+a2Ci2+……+amCim, in formula, f (Ci) Represent the corresponding model running parameter differences value of the i-th on-line time section, Ci1For first kind match parameter group parameter difference or Parameter difference absolute value, a1It is corresponding for the parameter difference or parameter difference absolute value of predetermined first kind match parameter group Weighted value, Ci2For the parameter difference or parameter difference absolute value of the second class match parameter group, a2It is predetermined second The parameter difference or the corresponding weighted value of parameter difference absolute value of class match parameter group, CimFor the ginseng of m class match parameter group Number difference or parameter difference absolute value, amFor the parameter difference or parameter difference of predetermined m class match parameter group The corresponding weighted value of absolute value, i, m are positive integer.
Only to choose two class preset kind model running parameters (for example, model analysis accuracy rate, every point in nearest 24 hours Clock is averaged call number) for, model point in nearest 24 hours of the corresponding convolution time network model of the 1st model on-line time section Analyse accuracy rate parameter difference C11Are as follows: C11=Yn1-Y11, the corresponding convolution time network model of the 1st model on-line time section is per minute Average call number parameter difference C12Are as follows: C12=Yn2-Y12, the corresponding model running parameter differences value of the 1st model on-line time section f(C1)=a1C11+a2C12=a1(Yn1-Y11)+a2(Yn2-Y12), a1It is accurate for model analysis in predetermined nearest 24 hours The corresponding weighted value of rate parameter difference, a2For the predetermined corresponding weighted value of call number parameter difference that is averaged per minute.
The execution module 120, for according to the corresponding model running parameter of calculated each model on-line time section Difference value is determined with the currently upper most matched model on-line time section of line model, the model on-line time that will be determined The corresponding upper line model of section is determined as and the currently upper most matched line model in history of line model；According to predetermined point Analysis algorithm analyses whether the model currently online with the determining iteration of line model in history, to current online described in iteration Model, then it is the current online model is offline, and the line model in history of the online determination.
Optionally, the execution module 120 is joined according to the corresponding model running of calculated each model on-line time section Number difference value is determined and the currently upper most matched model on-line time section of line model, comprising:
If the corresponding model running parameter differences value of all model on-line time sections is all larger than or is equal to preset threshold, Determining does not terminate with the current above most matched model on-line time section of line model, the process；Or
If the corresponding model running parameter differences value of an only model on-line time section is less than preset threshold, it is determined that should The most matched model on-line time section of line model is currently gone up described in model on-line time Duan Weiyu；Or
It, will be multiple if there is the corresponding model running parameter differences value of multiple model on-line time sections to be less than preset threshold In the corresponding model running parameter differences value of model on-line time section on the corresponding model of the smallest model running parameter differences value The line period currently goes up the most matched model on-line time section of line model as with described.
For example, setting preset threshold as 10, if the corresponding model running parameter differences value of all model on-line time sections is equal More than or equal to 10, then the most matched model on-line time section of line model is not gone up with current；If only the 2nd model on-line time The corresponding model running parameter differences value of section is less than 10, it is determined that currently upper line model is most by the 2nd model on-line time Duan Weiyu The corresponding supporting vector machine model of 2nd model on-line time section is determined as and current online mould by the model on-line time section matched The most matched line model in history of type；If the corresponding model running parameter differences value of the 1st model on-line time section is the 6, the 2nd model The corresponding model running parameter differences value of on-line time section is 3, other corresponding model running parameter differences of model on-line time section Different value is all larger than 10, then the corresponding 2nd model on-line time Duan Weiyu of preference pattern operating parameter difference value the smallest 3 it is current on The most matched model on-line time section of line model；If the corresponding model running parameter differences value of the 1st model on-line time section is 6, The corresponding model running parameter differences value of 2nd model on-line time section is the corresponding model running of the 3, the 5th model on-line time section Parameter differences value is also 3, other corresponding model running parameter differences values of model on-line time section are all larger than 10, then transports in model A work is randomly choosed in the corresponding 2nd model on-line time section of row parameter differences value the smallest 3 and the 5th model on-line time section To go up the most matched model on-line time section of line model with current.
Optionally, the execution module 120 is analysed whether with determination in history according to predetermined parser The current online model of line model iteration, comprising:
It is trained using in history line model of the predetermined model training data to the determination, generation trains Line model in history；
Using predetermined model verify data, the trained line model in history and described current is separately verified The accuracy rate of upper line model；
If the accuracy rate of the trained line model in history is higher, it is determined that with the line model in history of the determination Current online model described in iteration.
In the second embodiment of the present invention:
The computing module 110, for the iterative operation of start by set date model, alternatively, receiving the iteration for model After instruction, the iterative operation of Boot Model；After the iterative operation of Boot Model, application environment is selected one by one, in selection one After application environment, the preset kind model running parameter of the current upper line model of the running environment is obtained, according to the application environment Corresponding pre-stored model on-line time section, the preset kind model running parameter of online model, upper line model are reflected Relation data is penetrated, the preset kind model running parameter that will acquire preset kind corresponding with each model on-line time section respectively Model running parameter carries out parameter matching, obtains the corresponding match parameter group of each model on-line time section, calculates separately out each The parameter difference or parameter difference absolute value of a match parameter group；It is respectively that each model on-line time section is each corresponding Parameter difference or parameter difference absolute value with parameter group substitute into predetermined calculation formula, calculate separately out each mould The corresponding model running parameter differences value of type on-line time section.
The model can be convolutional neural networks model, supporting vector machine model, Random Forest model etc., the iteration The purpose of operation is used for Optimized model, and the iterative operation includes but is not limited to adjust version, adjustment model parameter etc..
Specifically, be stored in advance the online model of each application environment, each on-line time section built according to time series, on The mapping relations data set of the preset kind model running parameter of line model, the application environment include recognition of face, Yong Hufen Class, big data analysis etc., the preset kind model running parameter can be model analysis accuracy rate in preset time (for example, Model analysis accuracy rate in nearest 24 hours), unit time model be averaged call number (for example, per minute model averagely adjust With number) etc..For example, in pre-stored data set face recognition application environment the corresponding volume of the 1st model on-line time section Model analysis accuracy rate is Y in nearest 24 hours of product neural network model11, unit time model be averaged call number be Y12, Model analysis in nearest 24 hours of the corresponding supporting vector machine model of the 2nd model on-line time section of face recognition application environment Accuracy rate is Y21, unit time model be averaged call number be Y22.After the iterative operation of Boot Model, selection application one by one Environment, if having selected face recognition application environment, the current slot n for obtaining face recognition application environment is corresponding current online Model analysis accuracy rate Y in nearest 24 hours of model convolutional neural networks modeln1It is averagely called with unit time model secondary Number Yn2, after carrying out parameter matching, the corresponding convolution time network model of the 1st model on-line time section of face recognition application environment Model analysis accuracy rate match parameter group is (Y in nearest 24 hoursn1, Y11), unit time model call number parameter group be (Yn2, Y12), in the corresponding supporting vector machine model of the 2nd model on-line time section nearest 24 hours of face recognition application environment Model analysis accuracy rate match parameter group is (Yn1, Y21), unit time model call number parameter group be (Yn2, Y22), face is known Model analysis accuracy rate in the corresponding convolution time network model of the 1st model on-line time section nearest 24 hours of other application environment Parameter difference is Yn1Subtract Y11Difference.Then by the parameter of the corresponding each match parameter group of each model on-line time section Difference or parameter difference absolute value substitute into predetermined calculation formula, and it is right can to calculate each model on-line time section The model running parameter differences value answered.
Optionally, the predetermined calculation formula are as follows: f (Ci)=a1Ci1+a2Ci2+……+amCim, in formula, f (Ci) Represent the corresponding model running parameter differences value of the i-th on-line time section, Ci1For first kind match parameter group parameter difference or Parameter difference absolute value, a1It is corresponding for the parameter difference or parameter difference absolute value of predetermined first kind match parameter group Weighted value, Ci2For the parameter difference or parameter difference absolute value of the second class match parameter group, a2It is predetermined second The parameter difference or the corresponding weighted value of parameter difference absolute value of class match parameter group, CimFor the ginseng of m class match parameter group Number difference or parameter difference absolute value, amFor the parameter difference or parameter difference of predetermined m class match parameter group The corresponding weighted value of absolute value, i, m are positive integer.
If application environment is recognition of face, only to choose two class preset kind model running parameters (for example, nearest 24 hours Interior model analysis accuracy rate, be averaged call number per minute) for, the 1st model on-line time section of face recognition application environment Model analysis accuracy rate parameter difference C in nearest 24 hours of corresponding convolution time network model11Are as follows: C11=Yn1-Y11, face Identify application environment the corresponding convolution time network model of the 1st model on-line time section per minute be averaged call number parameter difference Value C12Are as follows: C12=Yn2-Y12, the corresponding model running parameter differences value of the 1st model on-line time section of face recognition application environment f(C1)=a1C11+a2C12=a1(Yn1-Y11)+a2(Yn2-Y12), a1It is accurate for model analysis in predetermined nearest 24 hours The corresponding weighted value of rate parameter difference, a2For the predetermined corresponding weighted value of call number parameter difference that is averaged per minute.
The execution module 120, for according to the corresponding model running parameter of calculated each model on-line time section Difference value is determined with the currently upper most matched model on-line time section of line model, the model on-line time that will be determined The corresponding upper line model of section is determined as and the currently upper most matched line model in history of line model；According to predetermined point Analysis algorithm analyses whether the model currently online with the determining iteration of line model in history, to current online described in iteration Model, then it is the current online model is offline, and the line model in history of the online determination.
Optionally, the execution module 120 is joined according to the corresponding model running of calculated each model on-line time section Number difference value is determined and the currently upper most matched model on-line time section of line model, comprising:
If the corresponding model running parameter differences value of all model on-line time sections is all larger than or is equal to preset threshold, Determining does not terminate with the current above most matched model on-line time section of line model, the process；Or
If the corresponding model running parameter differences value of an only model on-line time section is less than preset threshold, it is determined that should The most matched model on-line time section of line model is currently gone up described in model on-line time Duan Weiyu；Or
It, will be multiple if there is the corresponding model running parameter differences value of multiple model on-line time sections to be less than preset threshold In the corresponding model running parameter differences value of model on-line time section on the corresponding model of the smallest model running parameter differences value The line period currently goes up the most matched model on-line time section of line model as with described.
For example, setting preset threshold as 10, if the corresponding mould of all model on-line time sections of face recognition application environment Type operating parameter difference value is all larger than or is equal to 10, then does not go up the most matched model on-line time section of line model with current； If only the corresponding model running parameter differences value of the 2nd model on-line time section of face recognition application environment is less than 10, it is determined that 2nd model on-line time Duan Weiyu of face recognition application environment currently goes up the most matched model on-line time section of line model, will The corresponding supporting vector machine model of the 2nd model on-line time section of face recognition application environment is determined as with currently upper line model most Matched line model in history；If the corresponding model running parameter difference of the 1st model on-line time section of face recognition application environment Different value is 6, and the corresponding model running parameter differences value of the 2nd model on-line time section of face recognition application environment is 3, and face is known Other corresponding model running parameter differences values of model on-line time section of other application environment are all larger than 10, then select recognition of face The corresponding 2nd model on-line time Duan Weiyu of model running parameter differences value the smallest 3 of application environment currently goes up line model most Matched model on-line time section；If the corresponding model running parameter of the 1st model on-line time section of face recognition application environment Difference value is 6, and the corresponding model running parameter differences value of the 2nd model on-line time section of face recognition application environment is 3, face The corresponding model running parameter differences value of the 5th model on-line time section for identifying application environment is also 3, face recognition application environment Other corresponding model running parameter differences values of model on-line time section be all larger than 10, then in the mould of face recognition application environment One is randomly choosed in the corresponding 2nd model on-line time section of type operating parameter difference value the smallest 3 and the 5th model on-line time section It is a as with the current upper most matched model on-line time section of line model.
Optionally, the execution module 120 is analysed whether with determination in history according to predetermined parser The current online model of line model iteration, comprising:
It is trained using in history line model of the predetermined model training data to the determination, generation trains Line model in history；
Using predetermined model verify data, the trained line model in history and described current is separately verified The accuracy rate of upper line model；
If the accuracy rate of the trained line model in history is higher, it is determined that with the line model in history of the determination Current online model described in iteration.
As shown in figure 3, the flow chart of the method first embodiment of model iteration control to realize the present invention.
The iterative operation of S11,1 start by set date model of server, alternatively, server 1 refers to receiving the iteration for model After order, the iterative operation of Boot Model；
The model can be convolutional neural networks model, supporting vector machine model, Random Forest model etc., the iteration The purpose of operation is used for Optimized model, and the iterative operation includes but is not limited to adjust version, adjustment model parameter etc..
S12, after the iterative operation of Boot Model, the server 1 obtains the preset kind model of current upper line model Operating parameter, the preset kind model running ginseng of model on-line time section according to the pre-stored data, online model, upper line model Several mapping relations data, the preset kind model running parameter that will acquire are corresponding pre- with each model on-line time section respectively If Type model operating parameter carries out parameter matching, the corresponding match parameter group of each model on-line time section is obtained, is counted respectively Calculate the parameter difference or parameter difference absolute value of each match parameter group；
Specifically, be stored in advance the online model of each on-line time section built according to time series, upper line model it is pre- If the mapping relations data set of Type model operating parameter, the preset kind model running parameter can be preset time internal model Type analysis accuracy rate (for example, model analysis accuracy rate in nearest 24 hours), unit time model are averaged call number (example Such as, model per minute is averaged call number) etc..For example, the corresponding volume of the 1st model on-line time section in pre-stored data set Model analysis accuracy rate is Y in nearest 24 hours of product neural network model11, unit time model be averaged call number be Y12, Model analysis accuracy rate is Y in nearest 24 hours of the corresponding supporting vector machine model of 2nd model on-line time section21, unit when Between model be averaged call number be Y22.After the iterative operation of Boot Model, server 1 obtains that current slot n is corresponding works as Model analysis accuracy rate Y in nearest 24 hours of preceding upper line model convolutional neural networks modeln1It is average with unit time model Call number Yn2, after carrying out parameter matching, in the corresponding convolution time network model of the 1st model on-line time section nearest 24 hours Model analysis accuracy rate match parameter group is (Yn1, Y11), unit time model call number parameter group be (Yn2, Y12), the 2nd mould Model analysis accuracy rate match parameter group is (Y in the corresponding supporting vector machine model of type on-line time section nearest 24 hoursn1, Y21), unit time model call number parameter group be (Yn2, Y22), the corresponding convolution time network of the 1st model on-line time section Model analysis accuracy rate parameter difference is Y in model nearest 24 hoursn1Subtract Y11Difference.
S13, the server 1 are respectively by the parameter difference of the corresponding each match parameter group of each model on-line time section Or parameter difference absolute value substitutes into predetermined calculation formula, and it is corresponding to calculate separately out each model on-line time section Model running parameter differences value；
Optionally, the predetermined calculation formula are as follows: f (Ci)=a1Ci1+a2Ci2+……+amCim, in formula, f (Ci) Represent the corresponding model running parameter differences value of the i-th on-line time section, Ci1For first kind match parameter group parameter difference or Parameter difference absolute value, a1It is corresponding for the parameter difference or parameter difference absolute value of predetermined first kind match parameter group Weighted value, Ci2For the parameter difference or parameter difference absolute value of the second class match parameter group, a2It is predetermined second The parameter difference or the corresponding weighted value of parameter difference absolute value of class match parameter group, CimFor the ginseng of m class match parameter group Number difference or parameter difference absolute value, amFor the parameter difference or parameter difference of predetermined m class match parameter group The corresponding weighted value of absolute value, i, m are positive integer.
Only to choose two class preset kind model running parameters (for example, model analysis accuracy rate, every point in nearest 24 hours Clock is averaged call number) for, model point in nearest 24 hours of the corresponding convolution time network model of the 1st model on-line time section Analyse accuracy rate parameter difference C11Are as follows: C11=Yn1-Y11, the corresponding convolution time network model of the 1st model on-line time section is per minute Average call number parameter difference C12Are as follows: C12=Yn2-Y12, the corresponding model running parameter differences value of the 1st model on-line time section f(C1)=a1C11+a2C12=a1(Yn1-Y11)+a2(Yn2-Y12), a1It is accurate for model analysis in predetermined nearest 24 hours The corresponding weighted value of rate parameter difference, a2For the predetermined corresponding weighted value of call number parameter difference that is averaged per minute.
S14, the server 1 are according to the corresponding model running parameter differences of calculated each model on-line time section Value, determine with the currently upper most matched model on-line time section of line model, the model on-line time section determined is right The upper line model answered is determined as and the currently upper most matched line model in history of line model；
Optionally, the server 1 is according to the corresponding model running parameter difference of calculated each model on-line time section Different value is determined and the currently upper most matched model on-line time section of line model, comprising:
If the corresponding model running parameter differences value of all model on-line time sections is all larger than or is equal to preset threshold, Determining does not terminate with the current above most matched model on-line time section of line model, the process；Or
If the corresponding model running parameter differences value of an only model on-line time section is less than preset threshold, it is determined that should The most matched model on-line time section of line model is currently gone up described in model on-line time Duan Weiyu；Or
It, will be multiple if there is the corresponding model running parameter differences value of multiple model on-line time sections to be less than preset threshold In the corresponding model running parameter differences value of model on-line time section on the corresponding model of the smallest model running parameter differences value The line period currently goes up the most matched model on-line time section of line model as with described.
For example, setting preset threshold as 10, if the corresponding model running parameter differences value of all model on-line time sections is equal More than or equal to 10, then the most matched model on-line time section of line model is not gone up with current；If only the 2nd model on-line time The corresponding model running parameter differences value of section is less than 10, it is determined that currently upper line model is most by the 2nd model on-line time Duan Weiyu The corresponding supporting vector machine model of 2nd model on-line time section is determined as and current online mould by the model on-line time section matched The most matched line model in history of type；If the corresponding model running parameter differences value of the 1st model on-line time section is the 6, the 2nd model The corresponding model running parameter differences value of on-line time section is 3, other corresponding model running parameter differences of model on-line time section Different value is all larger than 10, then the corresponding 2nd model on-line time Duan Weiyu of preference pattern operating parameter difference value the smallest 3 it is current on The most matched model on-line time section of line model；If the corresponding model running parameter differences value of the 1st model on-line time section is 6, The corresponding model running parameter differences value of 2nd model on-line time section is the corresponding model running of the 3, the 5th model on-line time section Parameter differences value is also 3, other corresponding model running parameter differences values of model on-line time section are all larger than 10, then transports in model A work is randomly choosed in the corresponding 2nd model on-line time section of row parameter differences value the smallest 3 and the 5th model on-line time section To go up the most matched model on-line time section of line model with current.
S15, the server 1 analyse whether to be changed with determining line model in history according to predetermined parser Generation current online model, it is to current online model described in iteration, then the current online model is offline and online The line model in history of the determination.
Optionally, the server 1 is analysed whether according to predetermined parser with the determining online mould of history The current online model of type iteration, comprising:
It is trained using in history line model of the predetermined model training data to the determination, generation trains Line model in history；
Using predetermined model verify data, the trained line model in history and described current is separately verified The accuracy rate of upper line model；
If the accuracy rate of the trained line model in history is higher, it is determined that with the line model in history of the determination Current online model described in iteration.
As shown in figure 4, the flow chart of the method second embodiment for implementation model iteration control of the present invention.
The iterative operation of S21,1 start by set date model of server, alternatively, server 1 refers to receiving the iteration for model After order, the iterative operation of Boot Model；
The model can be convolutional neural networks model, supporting vector machine model, Random Forest model etc., the iteration The purpose of operation is used for Optimized model, and the iterative operation includes but is not limited to adjust version, adjustment model parameter etc..
S22, after the iterative operation of Boot Model, the server 1 selects application environment one by one, selection one application After environment, the preset kind model running parameter of the current upper line model of the running environment is obtained, it is corresponding according to the application environment Pre-stored model on-line time section, the preset kind model running parameter of online model, upper line model mapping close Coefficient evidence, the preset kind model running parameter that will acquire preset kind model corresponding with each model on-line time section respectively Operating parameter carries out parameter matching, obtains the corresponding match parameter group of each model on-line time section, calculates separately out each Parameter difference or parameter difference absolute value with parameter group；
Specifically, be stored in advance the online model of each application environment, each on-line time section built according to time series, on The mapping relations data set of the preset kind model running parameter of line model, the application environment include recognition of face, Yong Hufen Class, big data analysis etc., the preset kind model running parameter can be model analysis accuracy rate in preset time (for example, Model analysis accuracy rate in nearest 24 hours), unit time model be averaged call number (for example, per minute model averagely adjust With number) etc..For example, in pre-stored data set face recognition application environment the corresponding volume of the 1st model on-line time section Model analysis accuracy rate is Y in nearest 24 hours of product neural network model11, unit time model be averaged call number be Y12, Model analysis in nearest 24 hours of the corresponding supporting vector machine model of the 2nd model on-line time section of face recognition application environment Accuracy rate is Y21, unit time model be averaged call number be Y22.After the iterative operation of Boot Model, server 1 selects one by one Application environment is selected, if having selected face recognition application environment, the current slot n of acquisition face recognition application environment is corresponding to be worked as Model analysis accuracy rate Y in nearest 24 hours of preceding upper line model convolutional neural networks modeln1It is average with unit time model Call number Yn2, after carrying out parameter matching, the corresponding convolution time net of the 1st model on-line time section of face recognition application environment Model analysis accuracy rate match parameter group is (Y in network model nearest 24 hoursn1, Y11), unit time model call number parameter Group is (Yn2, Y12), the corresponding supporting vector machine model nearest 24 of the 2nd model on-line time section of face recognition application environment is small When interior model analysis accuracy rate match parameter group be (Yn1, Y21), unit time model call number parameter group be (Yn2, Y22), people Face identifies that model analysis is quasi- in the corresponding convolution time network model of the 1st model on-line time section nearest 24 hours of application environment True rate parameter difference is Yn1Subtract Y11Difference.
S23, the server 1 are respectively by the parameter difference of the corresponding each match parameter group of each model on-line time section Or parameter difference absolute value substitutes into predetermined calculation formula, and it is corresponding to calculate separately out each model on-line time section Model running parameter differences value；
Optionally, the predetermined calculation formula are as follows: f (Ci)=a1Ci1+a2Ci2+……+amCim, in formula, f (Ci) Represent the corresponding model running parameter differences value of the i-th on-line time section, Ci1For first kind match parameter group parameter difference or Parameter difference absolute value, a1It is corresponding for the parameter difference or parameter difference absolute value of predetermined first kind match parameter group Weighted value, Ci2For the parameter difference or parameter difference absolute value of the second class match parameter group, a2It is predetermined second The parameter difference or the corresponding weighted value of parameter difference absolute value of class match parameter group, CimFor the ginseng of m class match parameter group Number difference or parameter difference absolute value, amFor the parameter difference or parameter difference of predetermined m class match parameter group The corresponding weighted value of absolute value, i, m are positive integer.
If application environment is recognition of face, only to choose two class preset kind model running parameters (for example, nearest 24 hours Interior model analysis accuracy rate, be averaged call number per minute) for, the 1st model on-line time section of face recognition application environment Model analysis accuracy rate parameter difference C in nearest 24 hours of corresponding convolution time network model11Are as follows: C11=Yn1-Y11, face Identify application environment the corresponding convolution time network model of the 1st model on-line time section per minute be averaged call number parameter difference Value C12Are as follows: C12=Yn2-Y12, the corresponding model running parameter differences value of the 1st model on-line time section of face recognition application environment f(C1)=a1C11+a2C12=a1(Yn1-Y11)+a2(Yn2-Y12), a1It is accurate for model analysis in predetermined nearest 24 hours The corresponding weighted value of rate parameter difference, a2For the predetermined corresponding weighted value of call number parameter difference that is averaged per minute.
S24, the server 1 are according to the corresponding model running parameter differences of calculated each model on-line time section Value, determine with the currently upper most matched model on-line time section of line model, the model on-line time section determined is right The upper line model answered is determined as and the currently upper most matched line model in history of line model；
Optionally, the server 1 is according to the corresponding model running parameter difference of calculated each model on-line time section Different value is determined and the currently upper most matched model on-line time section of line model, comprising:
If the corresponding model running parameter differences value of all model on-line time sections is all larger than or is equal to preset threshold, Determining does not terminate with the current above most matched model on-line time section of line model, the process；Or
If the corresponding model running parameter differences value of an only model on-line time section is less than preset threshold, it is determined that should The most matched model on-line time section of line model is currently gone up described in model on-line time Duan Weiyu；Or
It, will be multiple if there is the corresponding model running parameter differences value of multiple model on-line time sections to be less than preset threshold In the corresponding model running parameter differences value of model on-line time section on the corresponding model of the smallest model running parameter differences value The line period currently goes up the most matched model on-line time section of line model as with described.
For example, setting preset threshold as 10, if the corresponding mould of all model on-line time sections of face recognition application environment Type operating parameter difference value is all larger than or is equal to 10, then does not go up the most matched model on-line time section of line model with current； If only the corresponding model running parameter differences value of the 2nd model on-line time section of face recognition application environment is less than 10, it is determined that 2nd model on-line time Duan Weiyu of face recognition application environment currently goes up the most matched model on-line time section of line model, will The corresponding supporting vector machine model of the 2nd model on-line time section of face recognition application environment is determined as with currently upper line model most Matched line model in history；If the corresponding model running parameter difference of the 1st model on-line time section of face recognition application environment Different value is 6, and the corresponding model running parameter differences value of the 2nd model on-line time section of face recognition application environment is 3, and face is known Other corresponding model running parameter differences values of model on-line time section of other application environment are all larger than 10, then select recognition of face The corresponding 2nd model on-line time Duan Weiyu of model running parameter differences value the smallest 3 of application environment currently goes up line model most Matched model on-line time section；If the corresponding model running parameter of the 1st model on-line time section of face recognition application environment Difference value is 6, and the corresponding model running parameter differences value of the 2nd model on-line time section of face recognition application environment is 3, face The corresponding model running parameter differences value of the 5th model on-line time section for identifying application environment is also 3, face recognition application environment Other corresponding model running parameter differences values of model on-line time section be all larger than 10, then in the mould of face recognition application environment One is randomly choosed in the corresponding 2nd model on-line time section of type operating parameter difference value the smallest 3 and the 5th model on-line time section It is a as with the current upper most matched model on-line time section of line model.
S25, the server 1 analyse whether to be changed with determining line model in history according to predetermined parser Generation current online model, it is to current online model described in iteration, then the current online model is offline and online The line model in history of the determination.
Optionally, the server 1 is analysed whether according to predetermined parser with the determining online mould of history The current online model of type iteration, comprising:
It is trained using in history line model of the predetermined model training data to the determination, generation trains Line model in history；
Using predetermined model verify data, the trained line model in history and described current is separately verified The accuracy rate of upper line model；
If the accuracy rate of the trained line model in history is higher, it is determined that with the line model in history of the determination Current online model described in iteration.
It should be noted last that the above examples are only used to illustrate the technical scheme of the present invention and are not limiting, although ginseng It is described the invention in detail according to embodiment, those skilled in the art should understand that, it can be to skill of the invention Art scheme is modified or equivalent replacement, without departing from the spirit and scope of the technical solution of the present invention.

Claims (10)

1. a kind of server of implementation model iteration control, which is characterized in that the server includes memory and processor, described It is stored with model iteration control program on memory, is realized when the model iteration control program is executed by the processor as follows Step:
The iterative operation of S11, start by set date model, alternatively, after receiving the iterative instruction for model, the iteration of Boot Model Operation；
S12, after the iterative operation of Boot Model, the preset kind model running parameter of current upper line model is obtained, according to pre- The mapping relations number of the preset kind model running parameter of the model on-line time section, online model, upper line model that first store According to, the preset kind model running parameter that will acquire preset kind model running corresponding with each model on-line time section respectively Parameter carries out parameter matching, obtains the corresponding match parameter group of each model on-line time section, calculates separately out each matching ginseng The parameter difference or parameter difference absolute value of array；
It is S13, respectively that the parameter difference of the corresponding each match parameter group of each model on-line time section or parameter difference is exhausted Value is substituted into predetermined calculation formula, the corresponding model running parameter difference of each model on-line time section is calculated separately out Different value；
S14, according to the corresponding model running parameter differences value of calculated each model on-line time section, determine to work as with described The corresponding upper line model of model on-line time section determined is determined as by the most matched model on-line time section of preceding upper line model With the currently upper most matched line model in history of line model；
S15, it is analysed whether according to predetermined parser with the current online mould of the determining iteration of line model in history Type, it is to current online model described in iteration, then the current online model is offline, and the history of the online determination Upper line model.
2. server as described in claim 1, which is characterized in that the predetermined calculation formula are as follows: f (Ci)=a1Ci1+ a2Ci2+……+amCim, in formula, f (Ci) represent the corresponding model running parameter differences value of the i-th on-line time section, Ci1For the first kind The parameter difference or parameter difference absolute value of match parameter group, a1For the parameter difference of predetermined first kind match parameter group Value or the corresponding weighted value of parameter difference absolute value, Ci2It is exhausted for the parameter difference or parameter difference of the second class match parameter group To value, a2For the parameter difference or the corresponding weighted value of parameter difference absolute value of predetermined second class match parameter group, CimFor the parameter difference or parameter difference absolute value of m class match parameter group, amFor predetermined m class match parameter group Parameter difference or the corresponding weighted value of parameter difference absolute value, i, m be positive integer.
3. server as claimed in claim 1 or 2, which is characterized in that described according to calculated each model on-line time The corresponding model running parameter differences value of section is determined and the currently upper most matched model on-line time section of line model, packet It includes:
If the corresponding model running parameter differences value of all model on-line time sections is all larger than or is equal to preset threshold, it is determined that Do not terminate with the current above most matched model on-line time section of line model, the process；Or
If the corresponding model running parameter differences value of an only model on-line time section is less than preset threshold, it is determined that the model The most matched model on-line time section of line model is currently gone up described in on-line time Duan Weiyu；Or
If thering is the corresponding model running parameter differences value of multiple model on-line time sections to be less than preset threshold, by multiple model When the corresponding model of the smallest model running parameter differences value is online in the corresponding model running parameter differences value of on-line time section Between section, as with the currently upper most matched model on-line time section of line model.
4. server as claimed in claim 1 or 2, which is characterized in that described to be according to the analysis of predetermined parser The no current online model of the determining iteration of line model in history, comprising:
It is trained using in history line model of the predetermined model training data to the determination, generates trained go through Line model in history；
Using predetermined model verify data, the trained line model in history and described current online is separately verified The accuracy rate of model；
If the accuracy rate of the trained line model in history is higher, it is determined that with the line model iteration in history of the determination The current online model.
5. a kind of server of implementation model iteration control, which is characterized in that the server includes memory and processor, described It is stored with model iteration control program on memory, is realized when the model iteration control program is executed by the processor as follows Step:
The iterative operation of S21, start by set date model, alternatively, after receiving the iterative instruction for model, the iteration of Boot Model Operation；
S22, after the iterative operation of Boot Model, select application environment one by one, after selecting an application environment, obtain the fortune The preset kind model running parameter of the current upper line model of row environment, according to the corresponding pre-stored model of the application environment The mapping relations data of on-line time section, the preset kind model running parameter of online model, upper line model, what be will acquire is pre- If preset kind model running parameter corresponding with each model on-line time section carries out parameter to Type model operating parameter respectively Matching, obtains the corresponding match parameter group of each model on-line time section, calculates separately out the parameter difference of each match parameter group Value or parameter difference absolute value；
It is S23, respectively that the parameter difference of the corresponding each match parameter group of each model on-line time section or parameter difference is exhausted Value is substituted into predetermined calculation formula, the corresponding model running parameter difference of each model on-line time section is calculated separately out Different value；
S24, according to the corresponding model running parameter differences value of calculated each model on-line time section, determine to work as with described The corresponding upper line model of model on-line time section determined is determined as by the most matched model on-line time section of preceding upper line model With the currently upper most matched line model in history of line model；
S25, it is analysed whether according to predetermined parser with the current online mould of the determining iteration of line model in history Type, it is to current online model described in iteration, then the current online model is offline, and the history of the online determination Upper line model.
6. server as claimed in claim 5, which is characterized in that described right according to calculated each model on-line time section The model running parameter differences value answered is determined and the currently upper most matched model on-line time section of line model, comprising:
If the corresponding model running parameter differences value of all model on-line time sections is all larger than or is equal to preset threshold, it is determined that Do not terminate with the current above most matched model on-line time section of line model, the process；Or
If the corresponding model running parameter differences value of an only model on-line time section is less than preset threshold, it is determined that the model The most matched model on-line time section of line model is currently gone up described in on-line time Duan Weiyu；Or
If thering is the corresponding model running parameter differences value of multiple model on-line time sections to be less than preset threshold, by multiple model When the corresponding model of the smallest model running parameter differences value is online in the corresponding model running parameter differences value of on-line time section Between section, as with the currently upper most matched model on-line time section of line model.
7. such as server described in claim 5 or 6, which is characterized in that described to be according to the analysis of predetermined parser The no current online model of the determining iteration of line model in history, comprising:
It is trained using in history line model of the predetermined model training data to the determination, generates trained go through Line model in history；
Using predetermined model verify data, the trained line model in history and described current online is separately verified The accuracy rate of model；
If the accuracy rate of the trained line model in history is higher, it is determined that with the line model iteration in history of the determination The current online model.
8. a kind of method of implementation model iteration control, which is characterized in that this method comprises:
The iterative operation of S11, server start by set date model, alternatively, server opens after receiving the iterative instruction for model The iterative operation of movable model；
S12, after the iterative operation of Boot Model, the server obtains the preset kind model running ginseng of current upper line model Number, model on-line time section according to the pre-stored data, the preset kind model running parameter of online model, upper line model are reflected Relation data is penetrated, the preset kind model running parameter that will acquire preset kind corresponding with each model on-line time section respectively Model running parameter carries out parameter matching, obtains the corresponding match parameter group of each model on-line time section, calculates separately out each The parameter difference or parameter difference absolute value of a match parameter group；
S13, the server respectively by the parameter difference of the corresponding each match parameter group of each model on-line time section or Parameter difference absolute value substitutes into predetermined calculation formula, calculates separately out the corresponding model of each model on-line time section Operating parameter difference value；
S14, the server are determined according to the corresponding model running parameter differences value of calculated each model on-line time section It is out with the currently upper most matched model on-line time section of line model, the model on-line time section determined is corresponding online Model is determined as and the currently upper most matched line model in history of line model；
S15, the server analyse whether to be worked as with the determining iteration of line model in history according to predetermined parser Preceding online model, it is to current online model described in iteration, then the current online model is offline and online described Determining line model in history.
9. a kind of method of implementation model iteration control, which is characterized in that this method comprises:
The iterative operation of S21, server start by set date model, alternatively, server opens after receiving the iterative instruction for model The iterative operation of movable model；
S22, after the iterative operation of Boot Model, the server selects application environment one by one, select an application environment Afterwards, the preset kind model running parameter of the current upper line model of the running environment is obtained, it is corresponding pre- according to the application environment The mapping relations number of the preset kind model running parameter of the model on-line time section, online model, upper line model that first store According to, the preset kind model running parameter that will acquire preset kind model running corresponding with each model on-line time section respectively Parameter carries out parameter matching, obtains the corresponding match parameter group of each model on-line time section, calculates separately out each matching ginseng The parameter difference or parameter difference absolute value of array；
S23, the server respectively by the parameter difference of the corresponding each match parameter group of each model on-line time section or Parameter difference absolute value substitutes into predetermined calculation formula, calculates separately out the corresponding model of each model on-line time section Operating parameter difference value；
S24, the server are determined according to the corresponding model running parameter differences value of calculated each model on-line time section It is out with the currently upper most matched model on-line time section of line model, the model on-line time section determined is corresponding online Model is determined as and the currently upper most matched line model in history of line model；
S25, the server analyse whether to be worked as with the determining iteration of line model in history according to predetermined parser Preceding online model, it is to current online model described in iteration, then the current online model is offline and online described Determining line model in history.
10. a kind of computer readable storage medium, which is characterized in that be stored with model on the computer readable storage medium and change Generation control program, the model iteration control program can execute by one or more processor, with realize as claim 8 to The step of method of any one of 9 implementation model iteration controls.
CN201910422133.0A 2019-05-21 2019-05-21 Method, server and computer readable storage medium for realizing model iteration control Active CN110286587B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910422133.0A CN110286587B (en) 2019-05-21 2019-05-21 Method, server and computer readable storage medium for realizing model iteration control

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910422133.0A CN110286587B (en) 2019-05-21 2019-05-21 Method, server and computer readable storage medium for realizing model iteration control

Publications (2)

Publication Number Publication Date
CN110286587A true CN110286587A (en) 2019-09-27
CN110286587B CN110286587B (en) 2022-02-22

Family

ID=68002006

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910422133.0A Active CN110286587B (en) 2019-05-21 2019-05-21 Method, server and computer readable storage medium for realizing model iteration control

Country Status (1)

Country Link
CN (1) CN110286587B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112561332A (en) * 2020-12-16 2021-03-26 北京百度网讯科技有限公司 Model management method, model management apparatus, electronic device, storage medium, and program product
WO2021174934A1 (en) * 2020-03-06 2021-09-10 深圳前海微众银行股份有限公司 Federated model parameter determination method, apparatus and device, and computer storage medium

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP1298512A2 (en) * 2001-09-26 2003-04-02 Coltec Industries Inc. Adaptive aero-thermodynamic engine model
CN101533261A (en) * 2007-09-28 2009-09-16 费舍-柔斯芒特系统股份有限公司 Method and apparatus for intelligent control and monitoring in a process control system
CN103080855A (en) * 2010-07-02 2013-05-01 西门子公司 Method and device for determining model parameters for controlling a steam power plant block, control unit for a steam generator and computer program product
US20150213358A1 (en) * 2009-11-17 2015-07-30 Hawk Network Defense Inc. Methods and apparatus for analyzing system events
CN107577660A (en) * 2017-07-21 2018-01-12 阿里巴巴集团控股有限公司 Category information recognition methods, device and server
CN108804161A (en) * 2018-06-21 2018-11-13 北京字节跳动网络技术有限公司 Initial method, device, terminal and the storage medium of application
CN109558952A (en) * 2018-11-27 2019-04-02 北京旷视科技有限公司 Data processing method, system, equipment and storage medium

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP1298512A2 (en) * 2001-09-26 2003-04-02 Coltec Industries Inc. Adaptive aero-thermodynamic engine model
CN101533261A (en) * 2007-09-28 2009-09-16 费舍-柔斯芒特系统股份有限公司 Method and apparatus for intelligent control and monitoring in a process control system
US20150213358A1 (en) * 2009-11-17 2015-07-30 Hawk Network Defense Inc. Methods and apparatus for analyzing system events
CN103080855A (en) * 2010-07-02 2013-05-01 西门子公司 Method and device for determining model parameters for controlling a steam power plant block, control unit for a steam generator and computer program product
CN107577660A (en) * 2017-07-21 2018-01-12 阿里巴巴集团控股有限公司 Category information recognition methods, device and server
CN108804161A (en) * 2018-06-21 2018-11-13 北京字节跳动网络技术有限公司 Initial method, device, terminal and the storage medium of application
CN109558952A (en) * 2018-11-27 2019-04-02 北京旷视科技有限公司 Data processing method, system, equipment and storage medium

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2021174934A1 (en) * 2020-03-06 2021-09-10 深圳前海微众银行股份有限公司 Federated model parameter determination method, apparatus and device, and computer storage medium
CN112561332A (en) * 2020-12-16 2021-03-26 北京百度网讯科技有限公司 Model management method, model management apparatus, electronic device, storage medium, and program product
CN112561332B (en) * 2020-12-16 2023-07-25 北京百度网讯科技有限公司 Model management method, device, electronic equipment, storage medium and program product

Also Published As

Publication number Publication date
CN110286587B (en) 2022-02-22

Similar Documents

Publication Publication Date Title
US20230381624A1 (en) Method and System for Interactive, Interpretable, and Improved Match and Player Performance Predictions in Team Sports
CN113574327B (en) Method and system for controlling an environment by selecting a control setting
CN107632995B (en) The method and model training control system of Random Forest model training
CN107818491A (en) Electronic installation, Products Show method and storage medium based on user's Internet data
WO2005010700A2 (en) System and method for representing and incorporating available information into uncertainty-based forecasts
CN112765462B (en) Data processing method and cloud server for big data service and artificial intelligence
CN110310114A (en) Object classification method, device, server and storage medium
CN111177568B (en) Object pushing method based on multi-source data, electronic device and storage medium
CN110286587A (en) Method, server and the computer readable storage medium of implementation model iteration control
CN109670835A (en) Air control method, apparatus, equipment and readable storage medium storing program for executing based on service node
CN113449753B (en) Service risk prediction method, device and system
CN110458184A (en) Optical character identification householder method, device, computer equipment and storage medium
CN110222243A (en) Determine the method, apparatus and storage medium of abnormal behaviour
US20160028769A1 (en) Policy evaluation trees
CN108376301A (en) Shares changing tendency prediction meanss, method and readable storage medium storing program for executing
US20160004583A1 (en) System for project management from non-function evaluation, method for project management from non-function evaluation, and program for project management from non-function evaluation
CN110210880A (en) Data processing method, device and computer readable storage medium
CN112579721A (en) Method and system for constructing crowd distribution map, terminal device and storage medium
CN110134361A (en) Volume adjusting method, computer equipment and the computer readable storage medium of equipment
CN116483983B (en) Method and related equipment for generating emotion change quantity of virtual character
CN114942881B (en) Method and device for A/B test experiment, electronic equipment and medium
US20240066355A1 (en) Live Tournament Predictions in Tennis
CN115170313A (en) Transaction risk determination method and device and server
CN104834979B (en) Leader row group method and system
CN118520168A (en) Game virtual article pushing control method and device and electronic equipment

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
GR01 Patent grant
GR01 Patent grant