CN110058922A - A kind of method, apparatus of the metadata of extraction machine learning tasks - Google Patents
A kind of method, apparatus of the metadata of extraction machine learning tasks Download PDFInfo
- Publication number
- CN110058922A CN110058922A CN201910208590.XA CN201910208590A CN110058922A CN 110058922 A CN110058922 A CN 110058922A CN 201910208590 A CN201910208590 A CN 201910208590A CN 110058922 A CN110058922 A CN 110058922A
- Authority
- CN
- China
- Prior art keywords
- metadata
- machine learning
- training
- container
- program code
- 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
Links
- 238000000034 method Methods 0.000 title claims abstract description 99
- 238000010801 machine learning Methods 0.000 title claims abstract description 89
- 238000000605 extraction Methods 0.000 title claims abstract description 85
- 238000012549 training Methods 0.000 claims abstract description 141
- 230000008569 process Effects 0.000 claims abstract description 59
- 239000000284 extract Substances 0.000 claims abstract description 17
- 238000012545 processing Methods 0.000 claims description 23
- 238000004590 computer program Methods 0.000 claims description 2
- 230000001052 transient effect Effects 0.000 claims description 2
- 230000006870 function Effects 0.000 description 11
- 230000007613 environmental effect Effects 0.000 description 10
- 238000013135 deep learning Methods 0.000 description 6
- 238000013527 convolutional neural network Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 5
- 238000013473 artificial intelligence Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 238000004891 communication Methods 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 230000006399 behavior Effects 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 230000008878 coupling Effects 0.000 description 2
- 230000010354 integration Effects 0.000 description 2
- 230000002093 peripheral effect Effects 0.000 description 2
- 239000007787 solid Substances 0.000 description 2
- 230000007480 spreading Effects 0.000 description 2
- 239000011800 void material Substances 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 230000010485 coping Effects 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000013075 data extraction Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000000802 evaporation-induced self-assembly Methods 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 238000012856 packing Methods 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 230000002085 persistent effect Effects 0.000 description 1
- 238000004321 preservation Methods 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 238000000638 solvent extraction Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/44—Arrangements for executing specific programs
- G06F9/455—Emulation; Interpretation; Software simulation, e.g. virtualisation or emulation of application or operating system execution engines
- G06F9/45533—Hypervisors; Virtual machine monitors
- G06F9/45558—Hypervisor-specific management and integration aspects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/48—Program initiating; Program switching, e.g. by interrupt
- G06F9/4806—Task transfer initiation or dispatching
- G06F9/4843—Task transfer initiation or dispatching by program, e.g. task dispatcher, supervisor, operating system
- G06F9/4881—Scheduling strategies for dispatcher, e.g. round robin, multi-level priority queues
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5005—Allocation of resources, e.g. of the central processing unit [CPU] to service a request
- G06F9/5011—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resources being hardware resources other than CPUs, Servers and Terminals
- G06F9/5016—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resources being hardware resources other than CPUs, Servers and Terminals the resource being the memory
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5061—Partitioning or combining of resources
- G06F9/5077—Logical partitioning of resources; Management or configuration of virtualized resources
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/44—Arrangements for executing specific programs
- G06F9/455—Emulation; Interpretation; Software simulation, e.g. virtualisation or emulation of application or operating system execution engines
- G06F9/45533—Hypervisors; Virtual machine monitors
- G06F9/45558—Hypervisor-specific management and integration aspects
- G06F2009/45562—Creating, deleting, cloning virtual machine instances
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/44—Arrangements for executing specific programs
- G06F9/455—Emulation; Interpretation; Software simulation, e.g. virtualisation or emulation of application or operating system execution engines
- G06F9/45533—Hypervisors; Virtual machine monitors
- G06F9/45558—Hypervisor-specific management and integration aspects
- G06F2009/45583—Memory management, e.g. access or allocation
Landscapes
- Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Theoretical Computer Science (AREA)
- General Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Data Mining & Analysis (AREA)
- Computing Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Mathematical Physics (AREA)
- Artificial Intelligence (AREA)
- Stored Programmes (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
This application provides a kind of method of the metadata in extraction machine learning tasks, the method is applied to virtualized environment, this method comprises: running machine learning task in virtualized environment according to the machine learning program code that user inputs;Metadata is extracted from the machine learning program code, the metadata is for reappearing the running environment of the machine learning task;The metadata is stored in the first memory space.Technical solution provided by the present application is in the training process of target machine learning tasks, automatically extract required relevant metadata when one specific training environment of reproduction, when other developers want one specific training environment of reproduction, specific training environment is reappeared according to the metadata of storage, accelerates the propagation of model.
Description
Technical field
This application involves field of cloud calculation, and more particularly, to a kind of metadata of extraction machine learning tasks
Method, apparatus and computer readable storage medium.
Background technique
Machine learning (machine learning, ML) is a multi-field cross discipline, specializes in computer mould
Quasi- or the realization mankind learning behaviors reorganize the existing structure of knowledge and are allowed to continuous to obtain new knowledge or technical ability
Improve the performance of itself.Machine learning is the core of artificial intelligence intelligence, is the fundamental way for making computer have intelligence, answers
With the every field for spreading artificial intelligence.
The workflow of machine learning task may include environmental structure, model training process and model reasoning process.
After source developer trains a model by the above process, its model trained can be provided to other developers.Its
The desired training process of reappearing of his developer just needs complete reproduction source exploitation environment.But other developers open in reproduction source
It is required a great deal of time during hair ring border to build and debug the training ring compatible with target machine learning tasks
Border brings great inconvenience to the propagation of model.
Summary of the invention
The application provides a kind of method, apparatus of the metadata in extraction machine learning tasks, and developer can be in target
In the training process of machine learning task, required some relevant members when one specific training environment of reproduction are automatically extracted
Data can be according to the relevant metadata of storage to spy when other developers want one specific training environment of reproduction
Fixed training environment is reappeared, and the propagation of model is accelerated.
In a first aspect, providing a kind of method of the metadata in extraction machine learning tasks, the method is applied to void
Quasi-ization environment, which comprises machine is run in the virtualized environment according to the machine learning program code that user inputs
Device learning tasks;Metadata is extracted from the machine learning program code, the metadata is used to appoint the machine learning
The running environment of business is reappeared;The metadata is stored in the first memory space.
In one possible implementation, by way of keyword search, according to the type of the metadata from institute
It states and extracts the metadata in machine learning program code.
In alternatively possible implementation, the virtualized environment runs the machine by least one training container
Device learning tasks, the metadata include first kind metadata.It can be according to the type of the first kind metadata from input
Extracts the first kind metadata in training container starting script, the trained container starting script for start it is described at least
One trained container.
In alternatively possible implementation, the type of the first kind metadata includes any one of following or more
It is a: frame that the machine learning task uses, the model that the machine learning task uses, the machine learning task instruction
Practice data set used in process.
In alternatively possible implementation, the virtualized environment runs the machine by least one training container
Device learning tasks, the metadata include the second class metadata.It can be according to the type of the second class metadata from input
The metadata is extracted in training program code, the training program code is stored at least one described training container carry
The second memory space in, the training program code be used for it is described at least one training container in run the machine learning
The model training process of task.
In alternatively possible implementation, the type of the second class metadata includes any one of following or more
It is a: the training of the processing mode of data set used in the training process of the machine learning task, the machine learning task
Training parameter used in the structure of model used in process, the training process of the machine learning task.
Second aspect, provides a kind of device of the metadata in extraction machine learning tasks, and described device runs on void
Quasi-ization environment, described device include:
Module is run, the machine learning program code for inputting according to user runs machine in the virtualized environment
Learning tasks;
Metadata extraction module, for extracting metadata from the machine learning program code, the metadata is used for
The running environment of the machine learning task is reappeared;
The metadata extraction module is also used to the metadata being stored in the first memory space.
In one possible implementation, the metadata extraction module is specifically used for: passing through the side of keyword search
Formula extracts the metadata from the machine learning program code according to the type of the metadata.
In alternatively possible implementation, the virtualized environment runs the machine by least one training container
Device learning tasks, the metadata include first kind metadata;
The metadata extraction module is specifically used for: according to the type of the first kind metadata from the training container of input
The first kind metadata is extracted in starting script, the trained container starting script is for starting at least one described training
Container.
In alternatively possible implementation, the type of the first kind metadata includes any one of following or more
It is a: frame that the machine learning task uses, the model that the machine learning task uses, the machine learning task instruction
Practice data set used in process.
In alternatively possible implementation, the virtualized environment runs the machine by least one training container
Device learning tasks, the metadata include the second class metadata;
The metadata extraction module is specifically used for: according to the type of the second class metadata from the training program of input
Extract the metadata in code, what the training program code was stored at least one training container carry second deposits
It stores up in space, the training program code is used to run the mould of the machine learning task at least one described training container
Type training process.
In alternatively possible implementation, the type of the second class metadata includes any one of following or more
It is a: the training of the processing mode of data set used in the training process of the machine learning task, the machine learning task
Training parameter used in the structure of model used in process, the training process of the machine learning task.
The third aspect provides a kind of system of the metadata in extraction machine learning tasks, and the system comprises at least
One server, each server include memory and at least one processor, and memory is used for program instruction, and described at least one
When a server is run, at least one processor executes the program instruction in the memory to execute first aspect or first party
Method in face in any possible implementation, or for realizing any possible in second aspect or second aspect
Operation module, metadata extraction module in implementation.
In one possible implementation, operation module may operate on the multiple server, meta-data extraction
Module may operate on each of multiple servers.
In alternatively possible implementation, metadata extraction module may operate in a part in multiple servers
On server.
In alternatively possible implementation, metadata extraction module be may operate in addition to above-mentioned multiple servers
Any one other server on.
Optionally, which can be general processor, can be realized by hardware or by software come real
It is existing.When passing through hardware realization, which can be logic circuit, integrated circuit etc.;When by software to realize, at this
Reason device can be a general processor, be realized by reading the software code stored in memory, which can collect
At in the processor, it can be located at except the processor, be individually present.
Fourth aspect provides a kind of non-transient readable storage medium storing program for executing, including program instruction, when described program instructs quilt
When computer is run, the computer is executed such as the side in first aspect and first aspect in any possible implementation
Method.
5th aspect, provides a kind of computer program product, including program instruction, when described program is instructed by computer
When operation, the computer is executed such as the method in first aspect and first aspect in any possible implementation.
The application can also be further combined on the basis of the implementation that above-mentioned various aspects provide to provide more
More implementations.
Detailed description of the invention
Fig. 1 is a kind of schematic block diagram of device 100 for running machine learning task provided by the embodiments of the present application.
Fig. 2 is a kind of schematic flow chart for executing machine learning task provided by the embodiments of the present application.
Fig. 3 is a kind of schematic block diagram of container environment 300 provided by the embodiments of the present application.
Fig. 4 is the schematic flow for the method that a kind of metadata extraction module provided by the embodiments of the present application extracts metadata
Figure.
Fig. 5 is the system 500 of the metadata during a kind of extraction machine learning training provided by the embodiments of the present application
Schematic diagram.
Specific embodiment
Below in conjunction with attached drawing, the technical solution in the application is described.
Machine learning (machine learning, ML) is a multi-field cross discipline, specializes in computer mould
Quasi- or the realization mankind learning behaviors reorganize the existing structure of knowledge and are allowed to continuous to obtain new knowledge or technical ability
Improve the performance of itself.Machine learning is the core of artificial intelligence intelligence, is the fundamental way for making computer have intelligence, answers
With the every field for spreading artificial intelligence.The workflow of machine learning task may include environmental structure, model training process
And model reasoning process.
Fig. 1 is a kind of schematic block diagram of device 100 for running machine learning task provided by the embodiments of the present application.Device
It may include the second storage for running module 110,120 carry of metadata extraction module 120 and metadata extraction module in 100
Space.Above-mentioned several modules are described in detail separately below.
Run module 110 in may include multiple submodule, such as: environmental structure submodule 111, training submodule 112,
Reasoning submodule 113, environment destroy submodule 114.
It should be understood that operation module 110, metadata extraction module 120 and its submodule may operate in virtualized environment,
For example, can be realized using container, for another example, it can also be realized using virtual machine, the embodiment of the present application is not done this specifically
It limits.
(1) environmental structure submodule 111:
Environmental structure submodule 111 is used to build to what the training environment of machine learning task carried out.Machine learning task
Building for environment is scheduling to computer hardware resource in fact, which can include but is not limited to: computing resource is deposited
Store up resource.
As machine learning task becomes increasingly complex, calculation amount is increasing, the cloud and containerization of machine learning task
As a development trend.It is graduallyd mature by the container technique of representative of docker, creates a virtualization using mirror image
Running environment.Relevant component can be disposed in a reservoir.Computing resource, storage resource are provided by the container technique, thus
Computing resource, the storage resource in physical machine can be called directly, provides hardware resource for machine learning task.For example, with
Kubernetes is the open source container dispatching platform of representative, can be effectively managed to container.
It should be understood that docker be one open source application container engine, the source developer of allowing can be packaged they application with
And packet is relied on into a transplantable container, it is then published on the Linux machine of any prevalence, also may be implemented virtual
Change.For ease of description, below by taking container as an example, technical solution provided by the embodiments of the present application is described in detail.Operation
In the case that virtualized environment where module 110, metadata extraction module 120 is virtual machine, module 110, metadata are run
Extraction module 120 and its submodule can realize by virtual machine
The embodiment of the present application is not specifically limited computing resource.It can be central processing unit (central
Processing unit, CPU), or can also be graphics processor (graphics processing unit, GPU).
Specifically, source developer can by way of being packaged mirror image, by the container mirror image of the associated component of packing, such as
The container mirror image of training assembly pulls in container environment.And the order line or container inputted by source developer starts foot
This, container is trained in creation and starting, and the training process of model is carried out in the training container.
(2) training submodule 112:
The training that training submodule 112 may operate in the above-mentioned container environment built, and be inputted according to source developer
The training process of program code progress model.
Specifically, source developer can be using Network File System (network file system, NFS) shared storage
Or other storage products in cloud platform, for example, distributed file system (distributed file system, DFS)
Training program code is stored in the first memory space 115 by mode, first memory space 115 can with carry starting instruction
Practice in container.Training submodule 112 can be according to getting the training program code of storage to model in the first memory space 115
It is trained.
The model trained can also be stored in the first memory space 115 during training by training submodule 112
In.
(3) reasoning submodule 113:
Accessible first memory space 115 of reasoning submodule 113, and can be stored according in the first memory space 115
Trained model make inferences process.Specifically, reasoning submodule 113 can by according to the training data of input and
Trained model determines the output valve of prediction.And it can be according to the output valve of prediction and the priori knowledge of training data
Whether the error between (prior knowledge), the model for determining that training submodule 112 trains are correct.
It should be understood that priori knowledge is also referred to as true value (ground truth), the training number provided by people is generally comprised
According to corresponding prediction result.
For example, machine learning task application is in field of image recognition.The mode input that above-mentioned trained submodule 112 trains
Training data be image Pixel Information, priori knowledge corresponding to training data is that the label of the image is " dog ".It will figure
The label of picture be " dog " training data be input in trained model, judge the model output predicted value whether be
"dog".If the output of the model is " dog ", it can determine that the model can be predicted accurately.
(4) environment destroys submodule 114:
After above-mentioned training process terminates, environment, which destroys submodule 114, can destroy the container environment of creation.But
First memory space 115 will not be destroyed, trained model is stored in first memory space 115, in order to reasoning
Submodule 113 makes inferences process according to the trained model of storage.
(5) metadata extraction module 120:
Metadata extraction module 120 can be during above-mentioned operation module 110 carries out machine learning task, Cong Yuankai
Metadata is automatically extracted out in the machine learning program code of originator input, which can be used for appointing above-mentioned machine learning
The running environment of business is reappeared.
The metadata extracted can also be generated description file by metadata extraction module 120, and the description of generation is literary
Part is stored in the second memory space 121.In order to which other sources developer wants the running environment to above-mentioned machine learning task
When being reappeared, the description file of storage is obtained from the second memory space 121, and according to some phases for including in description file
Pass metadata directly configures and debugs exploitation environment and accelerates the propagation of model to reappear target training environment out.
In order to reappear the training environment of a specific machine learning task out, source developer would generally be mentioned in the prior art
One of description standard for following three kinds of associated metadatas is a variety of, for example, the frame of the deep learning of source developer selection
The data set (dataset) that model (model) that frame (framework), source developer use, source developer use.It ties below
Above-mentioned several metadata are described in detail in conjunction table 1- table 3.
1 frame of table (framework)
Attribute | Type | Description |
Title (name) | string | The title of the deep learning frame of source developer selection |
Version (version) | string | The version of the deep learning frame of source developer selection |
As shown in table 1, the frame of deep learning can include but is not limited to: tensor stream (tensorflow), convolutional Neural
Network frame (convolutional neural network framework, CNNF), the convolution being embedded in for swift nature
Structure (convolutional architecture for fast feature embedding, CAFFE).
It should be understood that network structure tensorflow common in addition to support, for example, convolutional neural networks
(convolutional neural network, CNN), Recognition with Recurrent Neural Network (recurent neural network, RNN)
Except, it can also support deeply study or even other computation-intensive scientific algorithms (such as Solving Partial Differential Equations).
2 model of table (model)
Attribute | Type | Description |
Title (name) | string | The model name that source developer uses |
Version (version) | string | The model version that source developer uses |
Source (source) | string | The source for the model that source developer uses |
Filename (file) | object | The model file name that source developer uses |
Author (creator) | string | The author for the model that source developer uses |
Time (time) | ISO-8601 | The creation time for the model that source developer uses |
As shown in table 2, the model that source developer uses can include but is not limited to: image recognition model, Text region mould
Type etc..
It should be noted that the model that source developer uses can be disclosed model, it is also possible to privately owned model.Such as
Fruit source developer uses disclosed model, and the disclosure model provides unified resource and positions (uniform resource
Location, URL) link.
It should be noted that.The filename for the model that source developer uses does not store the description file of metadata directly
In, the filename for the model being somebody's turn to do file can be described with metadata with the description form of filename together be packaged.If source is developed
The model that person uses is disclosed model, and it is URL link that metadata, which describes file,.
3 data set of table (dataset)
Attribute | Type | Description |
Title (name) | string | The title for the data set that source developer uses |
Version (version) | string | The version for the data set that source developer uses |
Source (source) | string | The source for the data set that source developer uses |
As shown in table 3, the compressed file of the URL link of the data set that source developer uses or data set itself can be with
Metadata describes file and is packaged together.
Referring to table 1- table 3, the metadata such as said frame, model, data set usually in environmental structure submodule 111, by
Source developer is determined by way of being packaged the container mirror image of training assembly.Another label language is write and started with source developer
Say that (yet another markup language, YAML) file is packaged for the container mirror image of training assembly, source developer exists
It include the frame of source developer selection and the deep learning used when writing and starting YAML file, in the program code of input
(framework), the metadata of the key such as model (model), data set (dataset).
But the metadata for relying solely on the offer in table 1- table 3 is difficult to reappear the training of a machine learning task out
Environment.The embodiment of the present application also provides one of description standards of following three kinds of associated metadatas or a variety of, for example, source is opened
The processing mode (data-process) for the data set that originator uses, the structure (model- for the model that source developer uses
Architecture), the training parameter (training-parameters) that source developer uses in the training process.It ties below
Above-mentioned several metadata are described in detail in conjunction table 4- table 6.
The processing mode (data-process) of 4 data set of table
Referring to table 4, the partitioning scheme for the data set that source developer defines, which can be, handles the data set of input
Process, for example, a part of the data set of input to be used for the training process of model, that is to say, that the partial data collection can be with
As the training data during model training.A part of the data set of input is used for the reasoning process of model, that is,
It says, which can be used as the test data during model reasoning.
The structure (model-architecture) of 5 model of table
6 training parameter of table (training-params)
Referring to table 4- table 6, one of metadata such as above-mentioned data set processing mode, model structure, training parameter or
It is a variety of to be typically hidden in the training program code that source developer is stored in the first memory space 150.
The embodiment of the present application by getting during above-mentioned operation module 110 carries out machine learning task automatically
Metadata shown in table 1- table 6.And exploitation environment is directly configured and debugged according to the metadata, to reappear target machine out
The training environment of habit task.
According to the description standard of metadata shown in above-mentioned table 1- table 6, the metadata of a total of 6 sport needs to extract, i.e.,
The frame (framework) of deep learning of source developer selection, model (model), data set (dataset), data set
Processing mode (data-process), the structure (model-architecture) of model, training ginseng used in training process
Number (training-params).The mode that it is determined due to different metadata is different, extract first number of above-mentioned 6 sport
According to specific implementation it is also not identical.
For extracting the metadata such as frame, model, data set as shown in table 1- table 3, due to the usual feelings of the metadata
It is just to have been had determined by source developer when being packaged the container mirror image of training assembly under condition, which is stored in starting instruction
On the physics host for practicing container.Therefore, metadata extraction module 120 can by physics host send querying command,
To get the metadata as shown in table 1- table 3 being stored on physics host.
For extracting the metadata such as data set processing mode, model structure, training parameter as shown in table 4- table 6, by
It in the metadata is determined after creating and starting training container, source developer is in the memory space to training container carry
It include the metadata in the training program code of middle storage, therefore, metadata extraction module 120 can be by accessing training container
The training program code stored in the memory space of carry, to get the metadata as shown in table 4- table 6.
Metadata extraction module 120 can extract above-mentioned several in the way of keyword search in the embodiment of the present application
The metadata of type.Below with reference to Fig. 2-Fig. 3, the entire flow figure of machine learning task provided by the embodiments of the present application is carried out
Detailed description.
Referring to fig. 2, the entire flow figure of machine learning task may include environmental structure process, training process, reasoning
Journey is separately below described in detail above three process.
(1) environmental structure process:
Step 210: source developer is packaged the mirror image of training assembly mirror image, metadata extraction module.
Source developer can determine whether the members such as frame, model, data set as shown in table 1- table 3 when being packaged training assembly mirror image
Data.
In the case where scheduling of resource platform is kubernetes, training assembly can be jupyter notebook.It should
Jupyter notebook is a kind of interactive web applications, and it is online that source developer can use jupyter notebook
The training program code of input and adjustment model.
Step 215: source developer starts container mirror image.
Source developer can store the mirror image of the training assembly mirror image being packaged in step 210, metadata extraction module 120
In container warehouse.It should be understood that container warehouse can be managed to container mirror image, store and protect container mirror image.For example, should
Container warehouse can be container registration (container registry).
Source developer can start script with input pod or order line carrys out the container mirror of the different editions from container warehouse
As drawing into container environment, start corresponding component in a reservoir.For example, running training assembly in training container, extracting
Metadata extraction module 120 is run in container.
It should be understood that the container mirror image of different editions can correspond to the metadata such as different frames, model, data set.
It should also be understood that may include title, the version of the container mirror image pulled in container starting script or order line, hold
The information such as the time of device image starting.
Specifically, referring to the container environment 300 in Fig. 3, providing in the container group 310 of training function may include training
Container and extraction vessel, training the first memory space of container carry 115, the second memory space of extraction vessel carry 121.
In the case where scheduling of resource platform is kubernetes, container group is properly termed as beanpod (pod).Pod exists
It is the smallest scheduling unit in kubernetes, may include multiple containers in a pod.For pod, it can run
On some physics host, when needing to dispatch, pod can be scheduled as a whole by kubernetes.
The memory space of container carry can be persistently volume (persistent volume, PV) in kubernetes, should
PV is the one section of network storage region distributed by network administrator.PV has the life cycle independently of any single pod, also
It is to say, after the life cycle of pod terminates, the container in pod can be destroyed, but the PV of the container carry in pod will not
It is destroyed.
(2) training process:
Step 220: source developer inputs training program code.
Source developer can be by training the training assembly (such as jupyter notebook) run in container according to table
The description standard of metadata shown in 1- table 6 inputs training program code.It include as shown in table 4- table 6 in the training program code
Data set processing mode, model structure, the metadata such as training parameter.
The training program code of input can store in the first memory space 115 of training container carry.
It should be noted that during model training, if necessary to modify to the training program code of input,
The training program code of model can be inputted and adjusted online by jupyter notebook.
After model training process terminates, trained model can also be stored in the first storage sky of trained container carry
Between in.
Step 225: metadata extraction module 120 extracts metadata, and the second storage for being stored in extraction vessel carry is empty
Between in 121.
The metadata extraction module 120 run in extraction vessel can pass through 6 institute of table 1- table in the way of keyword extraction
The description standard for the metadata shown extracts above-mentioned metadata.
The mode that it is determined due to different metadata is different, extract the specific implementation of the metadata of above-mentioned 6 sport
Mode is not also identical.A kind of possible container starting foot for being achieved in that metadata extraction module 120 and being inputted from source developer
Originally, the metadata such as frame, model, data set as shown in table 1- table 3 are extracted in order line.Alternatively possible implementation
Be metadata extraction module 120 stored from source developer extracted in training program code into the first memory space 115 as
The metadata such as data set processing mode, model structure, training parameter shown in table 4- table 6.Specific reference to the description in Fig. 4,
Details are not described herein again.
It should also be noted that, the pod for providing training function can be destroyed at the end of model training task, but hang
The first memory space 115 and the second memory space 121 carried will not be destroyed.
(3) reasoning process:
Step 230: the container mirror image of starting inference component mirror image, metadata extraction module 120.
Create and start the process and step 215 of the container mirror image of reasoning container mirror image and metadata extraction module 120
It is corresponding, the description in step 215 is specifically referred to, details are not described herein again.
Step 235: reasoning container makes inferences service according to the model trained.
Specifically, referring to the container environment 300 of Fig. 3, providing in the container group 320 of inference function may include that reasoning is held
Device and extraction vessel.First memory space 115 of training container carry can be mounted to reasoning container again, by training function
Container group in extraction vessel carry the second memory space 121 be mounted to again provide inference function container group in mentioning
Extracting container.
Reasoning container can make inferences according to the model trained stored in the first memory space 115 of carry, mention
Issuable metadata in reasoning process can also be got for the extraction vessel in the container group of inference function, and by this yuan
Data are stored in the second memory space 121 of carry.
Below with reference to the example in Fig. 4, the mistake of metadata is extracted to the metadata extraction module 120 run in extraction vessel
Journey is described in detail.
Fig. 4 is the schematic of the method that a kind of metadata extraction module 120 provided by the embodiments of the present application extracts metadata
Flow chart.Method shown in Fig. 4 may include step 410-420, and step 410-420 is described in detail separately below.
It should be understood that according to the difference of the type of the metadata of extraction, it can be by metadata extraction module 120 shown in FIG. 1
It is divided into two parts, respectively the first metadata extraction module and the second metadata extraction module.
First metadata extraction module can be used for being packaged training assembly from extraction source developer on physics host
The metadata such as frame, model, the data set as shown in table 1- table 3 determined when container mirror image.Second metadata extraction module
It can be used for extracting in the training program code stored into the memory space of training container carry such as table from source developer
The metadata such as data set processing mode, model structure, training parameter shown in 4- table 6.
Optionally, in some embodiments, scheduling of resource platform is kubernetes, and the first metadata extraction module can be with
To work extractor (job extractor), job extractor is kubectl order line.Second metadata extraction module is
Code extractor (code extractor).It for ease of description, using resource dispatching platform is below kubernetes as showing
Example is described.
Step 410: the first metadata extraction module sends querying command to physics host's pusher side to extract such as 3 institute of table 1- table
The metadata such as the frame, model, the data set that show.
For metadata such as frame, model, data sets as shown in table 1- table 3 via source developer by being packaged training
The mode of the container mirror image of component determines, and container mirror image is stored in container warehouse.Source developer understands input pod starting
Script or order line pull the container mirror images of different editions from container warehouse, and the container mirror image of different editions corresponds to not
The metadata such as same frame, model, data set.
Since the metadata such as frame, model, data set are stored on physics host, job extractor needs
Service outside access obtains the metadata such as the frame stored on physics host, model, data set.In the embodiment of the present application
The first metadata extraction module (for example, job extractor) can be made by configuring gateway (for example, outlet (egress))
Address Internet protocol (internet protocol, IP) of physics host can be accessed by egress, and pass through hair
Querying command row is sent to obtain the metadata such as frame, model, data set.
In the case where scheduling of resource platform is kubernetes, can sending " kubectl get ", order line is dynamically
From the container of physics host's pusher side starting script, order line in the way of keyword extraction, relevant metadata is extracted
Information, for example, the title of container mirror image, version, the time of container image starting, frame, model, the metadata such as data set.And
With the format of java scripting object symbol (java script object notation, JSON) or alternative document format
Form is stored in the second memory space 121 of its carry.
Step 420: the second metadata extraction module is extracted from the first memory space 115 of training container carry such as table
The metadata such as data set processing mode, model structure, training parameter shown in 4- table 6.
The metadata such as data set processing mode, model structure, training parameter as shown in table 4- table 6 have been opened via source
Originator is stored in the first memory space 115 of trained container carry, therefore, the second metadata extraction module (for example,
It codeextractor) can be by way of keyword search, according to the description standard of metadata shown in table 4- table 6, from instruction
Practice and extracts data set shown in discrepancy table 4- table 6 in the training program code stored in the first memory space 115 of container carry
The metadata such as processing mode, model structure, training parameter.And it is stored in the form of the format of JSON or alternative document format
In second memory space 121 of its carry.
Above-mentioned code extractor and job extractor can will be extracted after extracting corresponding metadata
The metadata integration arrived, and the metadata after integration is deposited in the form second of " metadata describes file+model+data set "
It stores up in space 121.
Source developer can be taken by the workflow of above-mentioned machine learning task carrying out environment in the embodiment of the present application
Build or model training during, automatically obtained by metadata extraction module and storage source developer machine learning appoint
The metadata as shown in table 1- table 6 used in business.After machine learning task terminates, if source developer or other open
Originator needs to reappear source exploitation environment, can realize entire machine learning task Life cycle by the metadata of preservation
Workflow is built, to reappear source exploitation environment out.
Fig. 5 is the system 500 of the metadata during a kind of extraction machine learning training provided by the embodiments of the present application
Schematic diagram may include at least one server in system 500.
For ease of description, it is illustrated by taking server 510, server 520 as an example in Fig. 5.Server 510 and server
520 structure is similar.
Operation module 110 shown in FIG. 1 may operate at least one server, for example, server 510 and server
Operation module 110 has been separately operable on 520.
There are many deployment forms of metadata extraction module 120 shown in FIG. 1, and the embodiment of the present application does not do specific limit to this
It is fixed.As an example, metadata extraction module 120 may operate in each of at least one server, for example, service
Metadata extraction module 120 has been separately operable on device 510 and server 520.As another example, metadata extraction module
120 can also operate in the part of server at least one server, for example, metadata extraction module 120 operates in
On server 510 or operate on server 520.As another example, metadata extraction module 120 can also operate in
On other servers other than at least one above-mentioned server, for example, metadata extraction module 120 operates in server
On 530.
The method that system 500 can execute the metadata in said extracted machine learning training process, specifically, system
It may include at least one processor and memory at least one server in 500.Memory is used to store program instruction,
The processor for including at least one server can execute the program instruction that stores in memory to realize said extracted machine
The method of metadata during learning training, or realize operation module 110 shown in FIG. 1, metadata extraction module in Fig. 1
120.Below with server 510 as an example, realizing the metadata in said extracted machine learning training process to server 510
The detailed process of method be described in detail.
It may include: at least one processor (for example, processor 511, processor 516), memory in server 510
512, communication interface 513, input/output interface 514.
Wherein, at least one processor can be connect with memory 512.The memory 512 can be used for storing program and refer to
It enables.The memory 512 can be the storage unit inside at least one processor, be also possible to and at least one processor independence
External memory unit, can also be including only with the storage unit inside at least one processor and at least one processor
The component of vertical external memory unit.
Memory 512 can be solid state hard disk (solid state drive, SSD), be also possible to hard disk drive
(hard disk drive, HDD) can also be read-only memory (read-only memory, ROM), random access memory
(random access memory, RAM) etc..
Optionally, server 510 can also include bus 515.Wherein, memory 512, input/output interface 514, communication
Interface 513 can be connect by bus 515 at least one processor.Bus 515 can be Peripheral Component Interconnect standard
(peripheral component interconnect, PCI) bus or expanding the industrial standard structure (extended
Industry standard architecture, EISA) bus etc..It is total that the bus 515 can be divided into address bus, data
Line, control bus etc..Only to be indicated with a line in Fig. 5, it is not intended that an only bus or a seed type convenient for indicating
Bus.
Optionally, in some embodiments, system 500 can also include cloud storage 540.Cloud storage 540 can be made
For external memory, it is connect with system 500.Above procedure instruction can store in memory 512, also can store and deposits in cloud
In reservoir 540.
In the embodiment of the present application, at least one processor can use central processing unit (central
Processing unit, CPU), it can also be other general processors, digital signal processor (digital signal
Processor, DSP), it is specific integrated circuit (application specific integrated circuit, ASIC), existing
At programmable gate array (fieldprogrammable gate Array, FPGA) or other programmable logic device, discrete
Door or transistor logic, discrete hardware components etc..General processor can be microprocessor or the processor can also
To be any conventional processor etc..Or using one or more integrated circuits, for executing relative program, to realize this Shen
It please technical solution provided by embodiment.
Referring to Fig. 5, in server 510, by taking processor 511 as an example, operation has operation module 110 in processor 511.Fortune
May include multiple submodule in row module 110, for example, environmental structure submodule 111 shown in FIG. 1, training submodule 112,
Reasoning submodule 113, environment destroy submodule 114.
The training program code of active developer's input, the training are stored in first memory space 115 of memory 512
It include one of metadata such as data set processing mode, model structure, training parameter as described in table 4- table 6 in program code
Or it is a variety of.The metadata that metadata extraction module 120 extracts is stored in second memory space 121.Third memory space
The training container starting script of active developer's input is stored in 5121, includes such as table 1- table in the trained container starting script
The one or more of the metadata such as frame, model shown in 3, data set.
Processor 511 obtains the program instruction of storage from memory 512, to run above-mentioned machine learning task.Specifically
, the environmental structure submodule 111 in operation module 110 obtains container from the third memory space 5121 of memory 512 and opens
In dynamic script, and execute the build process of said vesse environment.The training submodule 112 in module 110 is run from memory 512
The first memory space 115 in obtain training program code, to execute the training process of above-mentioned model, and can be by the instruction of model
Practice result to be stored in first memory space 115.Specific each submodule in relation in operation module 110 executes machine
The specific implementation process of learning tasks, please refers to the description in Fig. 1, details are not described herein again.
In the operational process of above-mentioned machine learning task, metadata extraction module 120 can be from the first of memory 512
Data set processing mode as described in table 4- table 6, model knot are extracted in the training program code stored in memory space 115
One or more of metadata such as structure, training parameter.Metadata extraction module 120 can also be from third memory space 5121
Middle storage container starting script in extract the metadata such as frame, model, data set as shown in table 1- table 3 one kind or
Person is a variety of.
Optionally, in some embodiments, the metadata extracted can also be generated and be described by metadata extraction module 120
File, and the description file of generation is stored in the second memory space 121 of memory 512.Specific related metadata mentions
The process that modulus block 120 extracts metadata please refers to described above, and details are not described herein again.
It should be understood that magnitude of the sequence numbers of the above procedures are not meant to execute suitable in the various embodiments of the application
Sequence it is successive, the execution of each process sequence should be determined by its function and internal logic, the implementation without coping with the embodiment of the present application
Process constitutes any restriction.
Those of ordinary skill in the art may be aware that list described in conjunction with the examples disclosed in the embodiments of the present disclosure
Member and algorithm steps can be realized with the combination of electronic hardware or computer software and electronic hardware.These functions are actually
It is implemented in hardware or software, the specific application and design constraint depending on technical solution.Professional technician
Each specific application can be used different methods to achieve the described function, but this realization is it is not considered that exceed
Scope of the present application.
It is apparent to those skilled in the art that for convenience and simplicity of description, the system of foregoing description,
The specific work process of device and unit, can refer to corresponding processes in the foregoing method embodiment, and details are not described herein.
In several embodiments provided herein, it should be understood that disclosed systems, devices and methods, it can be with
It realizes by another way.For example, the apparatus embodiments described above are merely exemplary, for example, the unit
It divides, only a kind of logical function partition, there may be another division manner in actual implementation, such as multiple units or components
It can be combined or can be integrated into another system, or some features can be ignored or not executed.Another point, it is shown or
The mutual coupling, direct-coupling or communication connection discussed can be through some interfaces, the indirect coupling of device or unit
It closes or communicates to connect, can be electrical property, mechanical or other forms.
The unit as illustrated by the separation member may or may not be physically separated, aobvious as unit
The component shown may or may not be physical unit, it can and it is in one place, or may be distributed over multiple
In network unit.It can select some or all of unit therein according to the actual needs to realize the mesh of this embodiment scheme
's.
It, can also be in addition, each functional unit in each embodiment of the application can integrate in one processing unit
It is that each unit physically exists alone, can also be integrated in one unit with two or more units.
It, can be with if the function is realized in the form of SFU software functional unit and when sold or used as an independent product
It is stored in a computer readable storage medium.Based on this understanding, the technical solution of the application is substantially in other words
The part of the part that contributes to existing technology or the technical solution can be embodied in the form of software products, the meter
Calculation machine software product is stored in a storage medium, including some instructions are used so that a computer equipment (can be a
People's computer, server or network equipment etc.) execute each embodiment the method for the application all or part of the steps.
And storage medium above-mentioned includes: that USB flash disk, mobile hard disk, read-only memory (read-only memory, ROM), arbitrary access are deposited
The various media that can store program code such as reservoir (random access memory, RAM), magnetic or disk.
The above, the only specific embodiment of the application, but the protection scope of the application is not limited thereto, it is any
Those familiar with the art within the technical scope of the present application, can easily think of the change or the replacement, and should all contain
Lid is within the scope of protection of this application.Therefore, the protection scope of the application should be based on the protection scope of the described claims.
Claims (15)
1. a kind of method of the metadata in extraction machine learning tasks, which is characterized in that the method is applied to virtualization ring
Border, which comprises
Machine learning task is run in the virtualized environment according to the machine learning program code that user inputs;
Metadata is extracted from the machine learning program code, the metadata is for the operation to the machine learning task
Environment is reappeared;
The metadata is stored in the first memory space.
2. the method according to claim 1, wherein described extract first number from the machine learning program code
According to, comprising:
By way of keyword search, institute is extracted from the machine learning program code according to the type of the metadata
State metadata.
3. according to the method described in claim 2, it is characterized in that, the virtualized environment is transported by least one training container
The row machine learning task, the metadata includes first kind metadata;
It is described by way of keyword search, extracted from the machine learning program code according to the type of the metadata
The metadata out, comprising:
The first kind member number is extracted from the training container of input starting script according to the type of the first kind metadata
According to the trained container starting script is for starting at least one described training container.
4. according to the method described in claim 3, it is characterized in that, the type of the first kind metadata includes following any one
A or multiple: model that frame that the machine learning task uses, the machine learning task use, the machine learning are appointed
Data set used in the training process of business.
5. the method according to claim 3 or 4, which is characterized in that the virtualized environment is held by least one training
Device runs the machine learning task, and the metadata includes the second class metadata;
It is described by way of keyword search, extracted from the machine learning program code according to the type of the metadata
The metadata out, comprising:
The metadata, the training are extracted from the training program code of input according to the type of the second class metadata
Program code is stored in the second memory space of at least one training container carry, and the training program code is used for
The model training process of the machine learning task is run at least one described training container.
6. according to the method described in claim 5, it is characterized in that, the type of the second class metadata includes following any one
It is a or multiple: the processing mode of data set used in the training process of the machine learning task, the machine learning task
Training process used in the structure of model, training parameter used in the training process of the machine learning task.
7. a kind of device of the metadata in extraction machine learning tasks, which is characterized in that described device runs on virtualization ring
Border, described device include:
Module is run, the machine learning program code for inputting according to user runs machine learning in the virtualized environment
Task;
Metadata extraction module, for extracting metadata from the machine learning program code, the metadata is used for institute
The running environment for stating machine learning task is reappeared;
The metadata extraction module is also used to the metadata being stored in the first memory space.
8. device according to claim 7, which is characterized in that the metadata extraction module is specifically used for:
By way of keyword search, institute is extracted from the machine learning program code according to the type of the metadata
State metadata.
9. device according to claim 8, which is characterized in that the virtualized environment is transported by least one training container
The row machine learning task, the metadata includes first kind metadata;
The metadata extraction module is specifically used for:
The first kind member number is extracted from the training container of input starting script according to the type of the first kind metadata
According to the trained container starting script is for starting at least one described training container.
10. device according to claim 9, which is characterized in that the type of the first kind metadata includes following any
It is one or more: model that frame that the machine learning task uses, the machine learning task use, the machine learning
Data set used in the training process of task.
11. device according to claim 9 or 10, which is characterized in that the virtualized environment is trained by least one
Container runs the machine learning task, and the metadata includes the second class metadata;
The metadata extraction module is specifically used for:
The metadata, the training are extracted from the training program code of input according to the type of the second class metadata
Program code is stored in the second memory space of at least one training container carry, and the training program code is used for
The model training process of the machine learning task is run at least one described training container.
12. device according to claim 11, which is characterized in that the type of the second class metadata includes following any
One or more: the processing mode of data set used in the training process of the machine learning task, the machine learning are appointed
Training parameter used in the structure of model used in the training process of business, the training process of the machine learning task.
13. a kind of system of the metadata in extraction machine learning tasks, the system comprises at least one server, each clothes
Business device includes memory and at least one processor, and the memory is used for program instruction, at least one described processor executes
Program instruction in the memory is in method described in any one of perform claim requirement 1 to 6.
14. a kind of non-transient readable storage medium storing program for executing, which is characterized in that including program instruction, when described program instruction is calculated
When machine is run, the computer executes such as method described in any one of claims 1 to 6.
15. a kind of computer program product, which is characterized in that including program instruction, run when described program is instructed by computer
When, the computer executes such as method described in any one of claims 1 to 6.
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910208590.XA CN110058922B (en) | 2019-03-19 | 2019-03-19 | Method and device for extracting metadata of machine learning task |
PCT/CN2020/070577 WO2020186899A1 (en) | 2019-03-19 | 2020-01-07 | Method and apparatus for extracting metadata in machine learning training process |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910208590.XA CN110058922B (en) | 2019-03-19 | 2019-03-19 | Method and device for extracting metadata of machine learning task |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110058922A true CN110058922A (en) | 2019-07-26 |
CN110058922B CN110058922B (en) | 2021-08-20 |
Family
ID=67317220
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910208590.XA Active CN110058922B (en) | 2019-03-19 | 2019-03-19 | Method and device for extracting metadata of machine learning task |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN110058922B (en) |
WO (1) | WO2020186899A1 (en) |
Cited By (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110532098A (en) * | 2019-08-30 | 2019-12-03 | 广东星舆科技有限公司 | The GPU method and system of service are provided |
CN110795141A (en) * | 2019-10-12 | 2020-02-14 | 广东浪潮大数据研究有限公司 | Training task submitting method, device, equipment and medium |
CN110837896A (en) * | 2019-11-22 | 2020-02-25 | 中国联合网络通信集团有限公司 | Storage and calling method and device of machine learning model |
CN111629061A (en) * | 2020-05-28 | 2020-09-04 | 苏州浪潮智能科技有限公司 | Inference service system based on Kubernetes |
CN111694641A (en) * | 2020-06-16 | 2020-09-22 | 中电科华云信息技术有限公司 | Storage management method and system for container application |
WO2020186899A1 (en) * | 2019-03-19 | 2020-09-24 | 华为技术有限公司 | Method and apparatus for extracting metadata in machine learning training process |
CN112286682A (en) * | 2020-10-27 | 2021-01-29 | 上海淇馥信息技术有限公司 | Machine learning task processing method, device and equipment based on distributed cluster |
CN112311605A (en) * | 2020-11-06 | 2021-02-02 | 北京格灵深瞳信息技术有限公司 | Cloud platform and method for providing machine learning service |
CN112395039A (en) * | 2019-08-16 | 2021-02-23 | 北京神州泰岳软件股份有限公司 | Management method and device for Kubernetes cluster |
CN112819176A (en) * | 2021-01-22 | 2021-05-18 | 烽火通信科技股份有限公司 | Data management method and data management device suitable for machine learning |
WO2021136365A1 (en) * | 2019-12-30 | 2021-07-08 | 第四范式(北京)技术有限公司 | Application development method and apparatus based on machine learning model, and electronic device |
US20210240471A1 (en) * | 2020-01-31 | 2021-08-05 | International Business Machines Corporation | Schema-based machine-learning model task deduction |
TWI772884B (en) * | 2020-09-11 | 2022-08-01 | 英屬維爾京群島商飛思捷投資股份有限公司 | Positioning system and method integrating machine learning positioning model |
WO2023173964A1 (en) * | 2022-03-14 | 2023-09-21 | International Business Machines Corporation | Intelligently optimized machine learning models |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20070283352A1 (en) * | 2005-10-14 | 2007-12-06 | Degenhardt Jon R | Sub-task mechanism for development of task-based user interfaces |
CN104899141A (en) * | 2015-06-05 | 2015-09-09 | 北京航空航天大学 | Test case selecting and expanding method facing network application system |
CN107451663A (en) * | 2017-07-06 | 2017-12-08 | 阿里巴巴集团控股有限公司 | Algorithm assembly, based on algorithm assembly modeling method, device and electronic equipment |
CN108665072A (en) * | 2018-05-23 | 2018-10-16 | 中国电力科学研究院有限公司 | A kind of machine learning algorithm overall process training method and system based on cloud framework |
CN108805282A (en) * | 2018-04-28 | 2018-11-13 | 福建天晴在线互动科技有限公司 | Deep learning data sharing method, storage medium based on block chain mode |
CN109146084A (en) * | 2018-09-06 | 2019-01-04 | 郑州云海信息技术有限公司 | A kind of method and device of the machine learning based on cloud computing |
Family Cites Families (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US10282171B2 (en) * | 2015-03-30 | 2019-05-07 | Hewlett Packard Enterprise Development Lp | Application analyzer for cloud computing |
CN109272116A (en) * | 2018-09-05 | 2019-01-25 | 郑州云海信息技术有限公司 | A kind of method and device of deep learning |
CN110058922B (en) * | 2019-03-19 | 2021-08-20 | 华为技术有限公司 | Method and device for extracting metadata of machine learning task |
-
2019
- 2019-03-19 CN CN201910208590.XA patent/CN110058922B/en active Active
-
2020
- 2020-01-07 WO PCT/CN2020/070577 patent/WO2020186899A1/en active Application Filing
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20070283352A1 (en) * | 2005-10-14 | 2007-12-06 | Degenhardt Jon R | Sub-task mechanism for development of task-based user interfaces |
CN104899141A (en) * | 2015-06-05 | 2015-09-09 | 北京航空航天大学 | Test case selecting and expanding method facing network application system |
CN107451663A (en) * | 2017-07-06 | 2017-12-08 | 阿里巴巴集团控股有限公司 | Algorithm assembly, based on algorithm assembly modeling method, device and electronic equipment |
CN108805282A (en) * | 2018-04-28 | 2018-11-13 | 福建天晴在线互动科技有限公司 | Deep learning data sharing method, storage medium based on block chain mode |
CN108665072A (en) * | 2018-05-23 | 2018-10-16 | 中国电力科学研究院有限公司 | A kind of machine learning algorithm overall process training method and system based on cloud framework |
CN109146084A (en) * | 2018-09-06 | 2019-01-04 | 郑州云海信息技术有限公司 | A kind of method and device of the machine learning based on cloud computing |
Non-Patent Citations (2)
Title |
---|
张蕾 等: "机器学习在网络空间安全研究中的应用", 《计算机学报》 * |
肖建军: "Demo3-保存训练后模型", 《HTTPS://DEVELOPER.ALIYUN.COM/ARTICLE/674117》 * |
Cited By (20)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2020186899A1 (en) * | 2019-03-19 | 2020-09-24 | 华为技术有限公司 | Method and apparatus for extracting metadata in machine learning training process |
CN112395039A (en) * | 2019-08-16 | 2021-02-23 | 北京神州泰岳软件股份有限公司 | Management method and device for Kubernetes cluster |
CN112395039B (en) * | 2019-08-16 | 2024-01-19 | 北京神州泰岳软件股份有限公司 | Method and device for managing Kubernetes cluster |
CN110532098B (en) * | 2019-08-30 | 2022-03-08 | 广东星舆科技有限公司 | Method and system for providing GPU (graphics processing Unit) service |
CN110532098A (en) * | 2019-08-30 | 2019-12-03 | 广东星舆科技有限公司 | The GPU method and system of service are provided |
CN110795141A (en) * | 2019-10-12 | 2020-02-14 | 广东浪潮大数据研究有限公司 | Training task submitting method, device, equipment and medium |
CN110795141B (en) * | 2019-10-12 | 2023-10-10 | 广东浪潮大数据研究有限公司 | Training task submitting method, device, equipment and medium |
CN110837896A (en) * | 2019-11-22 | 2020-02-25 | 中国联合网络通信集团有限公司 | Storage and calling method and device of machine learning model |
WO2021136365A1 (en) * | 2019-12-30 | 2021-07-08 | 第四范式(北京)技术有限公司 | Application development method and apparatus based on machine learning model, and electronic device |
US20210240471A1 (en) * | 2020-01-31 | 2021-08-05 | International Business Machines Corporation | Schema-based machine-learning model task deduction |
WO2021238251A1 (en) * | 2020-05-28 | 2021-12-02 | 苏州浪潮智能科技有限公司 | Inference service system based on kubernetes |
CN111629061B (en) * | 2020-05-28 | 2023-01-24 | 苏州浪潮智能科技有限公司 | Inference service system based on Kubernetes |
CN111629061A (en) * | 2020-05-28 | 2020-09-04 | 苏州浪潮智能科技有限公司 | Inference service system based on Kubernetes |
CN111694641A (en) * | 2020-06-16 | 2020-09-22 | 中电科华云信息技术有限公司 | Storage management method and system for container application |
TWI772884B (en) * | 2020-09-11 | 2022-08-01 | 英屬維爾京群島商飛思捷投資股份有限公司 | Positioning system and method integrating machine learning positioning model |
CN112286682A (en) * | 2020-10-27 | 2021-01-29 | 上海淇馥信息技术有限公司 | Machine learning task processing method, device and equipment based on distributed cluster |
CN112311605A (en) * | 2020-11-06 | 2021-02-02 | 北京格灵深瞳信息技术有限公司 | Cloud platform and method for providing machine learning service |
CN112311605B (en) * | 2020-11-06 | 2023-12-22 | 北京格灵深瞳信息技术股份有限公司 | Cloud platform and method for providing machine learning service |
CN112819176A (en) * | 2021-01-22 | 2021-05-18 | 烽火通信科技股份有限公司 | Data management method and data management device suitable for machine learning |
WO2023173964A1 (en) * | 2022-03-14 | 2023-09-21 | International Business Machines Corporation | Intelligently optimized machine learning models |
Also Published As
Publication number | Publication date |
---|---|
WO2020186899A1 (en) | 2020-09-24 |
CN110058922B (en) | 2021-08-20 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110058922A (en) | A kind of method, apparatus of the metadata of extraction machine learning tasks | |
Abadi et al. | Tensorflow: Large-scale machine learning on heterogeneous distributed systems | |
Butterfield et al. | A dictionary of computer science | |
US20190332230A1 (en) | User interface view generation | |
US9280340B2 (en) | Dynamically building an unstructured information management architecture (UIMA) pipeline | |
CN110580197A (en) | Distributed computing architecture for large model deep learning | |
Bengfort et al. | Data analytics with Hadoop: an introduction for data scientists | |
US11520829B2 (en) | Training a question-answer dialog sytem to avoid adversarial attacks | |
WO2020119268A1 (en) | Model-based prediction method and device | |
CN109154908A (en) | The part of dynamic size with accurate garbage collection report | |
EP3021266A1 (en) | Lean product modeling systems and methods | |
US10191921B1 (en) | System for expanding image search using attributes and associations | |
US10282411B2 (en) | System, method, and recording medium for natural language learning | |
CN113449299A (en) | Projected vector modification as suppression of machine learning model string fill | |
WO2020038376A1 (en) | Method and system for uniformly performing feature extraction | |
KR102002732B1 (en) | Deep neural network based data processing method and apparatus using ensemble model | |
Bauer et al. | A Python extension for the massively parallel multiphysics simulation framework waLBerla | |
KR101494795B1 (en) | Method for representing document as matrix | |
Gesing et al. | Workflows in a dashboard: a new generation of usability | |
US11449797B1 (en) | Secure machine learning workflow automation using isolated resources | |
Achterberg et al. | Fastr: A workflow engine for advanced data flows in medical image analysis | |
Estrada | Fast data processing systems with SMACK stack | |
Karau et al. | Scaling Python with Ray | |
Munerman et al. | Realization of Distributed Data Processing on the Basis of Container Technology | |
US20220067502A1 (en) | Creating deep learning models from kubernetes api objects |
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 | ||
TR01 | Transfer of patent right |
Effective date of registration: 20211221 Address after: 450046 Floor 9, building 1, Zhengshang Boya Plaza, Longzihu wisdom Island, Zhengdong New Area, Zhengzhou City, Henan Province Patentee after: xFusion Digital Technologies Co., Ltd. Address before: 518129 Bantian HUAWEI headquarters office building, Longgang District, Guangdong, Shenzhen Patentee before: HUAWEI TECHNOLOGIES Co.,Ltd. |
|
TR01 | Transfer of patent right |