CN106067028A - The modeling method of automatic machinery based on GPU study - Google Patents
The modeling method of automatic machinery based on GPU study Download PDFInfo
- Publication number
- CN106067028A CN106067028A CN201510184147.5A CN201510184147A CN106067028A CN 106067028 A CN106067028 A CN 106067028A CN 201510184147 A CN201510184147 A CN 201510184147A CN 106067028 A CN106067028 A CN 106067028A
- Authority
- CN
- China
- Prior art keywords
- model
- gpu
- data
- study
- modeling method
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Abstract
The invention discloses the modeling method of a kind of automatic machinery based on GPU study, comprise the following steps: set up model library, prepare data；The characteristic of the data of input is normalized, selects multiple models of general mode to enter GPU parallel computation；According to input data, automatically preparing the data of cross validation, default is 10 foldings, 9 folding training, a folding validation test；Being calculated the cross validation of multiple model by GPU respectively, respectively obtain the error of each model, the less model of Select Error is as the final model selected；Using all data as training data, the final model selected of training, obtain final model, remember into Ypre=f (X).Automatically calculate for domestic consumer and obtain the model optimized, alleviate the complexity of domestic consumer's machine learning application and the undue dependence to expert.
Description
Technical field
The present invention relates to the modeling method of a kind of machine learning, be specifically related to the modeling method of a kind of automatic machinery based on GPU study.
Background technology
Machine learning is to study the subject how using machine to carry out simulating human learning activity.The strictest lifting manipulation is: machine learning is that a research machine obtains new knowledge and new technical ability, and identifies existing acquainted knowledge.
Machine learning is the science of an artificial intelligence, and the main study subject in this field is artificial intelligence, in particular how improves the performance of specific algorithm in empirical learning.Machine learning generally relates to two big processes, and one is training Procedure Acquisition parameter or model, and another is that application model is predicted or classification etc..
The training process of machine learning is complicated, it usually needs professional designs a model and repeatedly attempts obtaining a model；For domestic consumer, this process is complicated and expensive, it is not easy to automatically obtain；Comparatively speaking, domestic consumer more concerned be the direct application training prediction or the classification results that obtain that model obtains, obtain model without concern for how to train.
Even if the expert to machine learning, training obtain one accurately model the most also have a difficulty much needing to overcome, the choice of such as model, the optimization etc. of model parameter,
The acquisition of machine learning model at present is usually and relies on some models of machine learning expert design, recycling computer aid training obtains the parameter optimized, and obtain final model, this method except relying in addition to expert can not be widely used by domestic consumer, the individual level being also limited by expert and the calculating resource that can arrive.
In order to improve the accuracy of model, needing to select to expand Model Selection scope and the scope of parameter, it means that huge amount of calculation, the most common platform that calculates is not caned rapidly, is effectively provided, and needs to be completed by new method.
Summary of the invention
For the defect overcoming prior art to exist, the present invention provides the modeling method that a kind of automatic machinery based on GPU learns, the present invention provides general mode to need not understand the model of optimization that general users of machine learning details obtain automatically to those, alleviates the complexity of domestic consumer's machine learning application and the undue dependence to expert.
For reaching above-mentioned purpose, the technical scheme is that
The modeling method of a kind of automatic machinery based on GPU study, comprises the following steps:
(1) set up model library, prepare data；
(2) characteristic of the data of input is normalized, selects multiple models of general mode to enter GPU parallel computation；
(3) according to input data, automatically preparing the data of cross validation, default is 10 foldings, 9 folding training, a folding validation test；
(4) calculating the cross validation of multiple model respectively by GPU, respectively obtain the error of each model, the less model of Select Error is as the final model selected；
(5) using all data as training data, the final model selected of training, obtain final model, remember into Ypre=f (X).
Preferably, described model library also includes expert mode, manually arranges the basic model in framework and framework, starts cross validation, obtains new frame model.
Preferably, described model library also includes professional mode, it is provided that the interface of new model algorithm, for adding new model.
Preferably, described general mode at least includes perceptron model, linear regression model (LRM), logistic regression models, supporting vector machine model, Back propagation neural networks model and feedback Hopfield network model.
Preferably, the framework of described expert mode includes Bag, Adaboost, Randon Forest, Deep Learning；Basic model includes decision tree, unsupervised learning K-means algorithm.
Preferably, described GPU parallel computation includes different model parallel computations sample set parallel computation different with during cross validation.
Preferably, described data form is that (X, y), wherein X is characterized, and y is classification or predictive value.
The invention has the beneficial effects as follows:
The present invention provides domestic consumer automatically to calculate and obtains the model optimized, and alleviates the complexity of domestic consumer's machine learning application and the undue dependence to expert.
The method of the automatic modeling of the present invention is also the platform tools of machine learning expert's training pattern, and meanwhile, its autgmentability is that senior expert user provides new model customization and extension.
Big data age, the sample size of machine learning is huge, so-called big data, brings huge computing cost, and big data disclosed by the invention calculate platform and use GPU parallel computation, and the complexity of automatic modeling is controlled.
Accompanying drawing explanation
Fig. 1 is the integral platform of the modeling method of present invention automatic machinery based on GPU study；
Fig. 2 is the applicable domestic consumer automation modeling platform of the modeling method of present invention automatic machinery based on GPU study；
Fig. 3 is the computation complexity block diagram based on parallel computation of the modeling method of present invention automatic machinery based on GPU study.
Detailed description of the invention
For making the object, technical solutions and advantages of the present invention of greater clarity, below in conjunction with detailed description of the invention and referring to the drawings, the present invention is described in more detail.It should be understood that these describe the most exemplary, and it is not intended to limit the scope of the present invention.Additionally, in the following description, the description to known features and technology is eliminated, to avoid unnecessarily obscuring idea of the invention.
As it is shown in figure 1, set up the platform of energy automatic machinery learning model building.
1 sets up model library, including general mode, expert mode, professional mode
2 totally obtain model and parameter by cross validation
3 cross-validation data prepare automatically, generate training sample set test sample collection
4 models configurations (default be universal model, manually can enter expert mode and professional mode, and candidate family and parameter can be revised)
5 cross validation preference patterns
6 training obtain the model parameter optimized
Fig. 2 is the example of a platform, and conventional model has RBF RBF, support vector machines, Polynomial SVM, AdaBoost etc..
Fig. 3 gives the solution block diagram of computation complexity.Available GPU can be on a main frame, it is also possible on multiple host.
Automatically working flow process is as follows:
(1) user prepares data, and data prepare according to certain call format, and the data of disaggregated model are data characteristics and mark (classification) (X thereof, y), it was predicted that the data of model be data characteristics and actual measured value thereof (X, y), X is characterized, and y is classification or predictive value.
(2) user enters data into the device of this invention；Below as a example by disaggregated model.
(3) characteristic (except the mark outer portion) X of the data of input is done normalization etc. and processes by invention device.
(4) general mode is started, without loss of generality, it is assumed that only perceptron and two kinds of models of support vector machines；Perceptron model and SVM model enter GPU parallel computation.
(5) according to input data, automatically preparing the data of cross validation, default is 10 foldings, 9 folding training, a folding validation test.
(6) identical data enter two groups of GPU and calculate perceptron model and the cross validation of SVM model respectively, respectively obtain the error of the two model, the less model of Select Error is as the final model selected, without loss of generality, it is assumed that select here is SVM model.
(7) using all data as training data, train SVM model, obtain final model, remember into Ypre=f (X).
(8) for new feature Xnew, its classification is unknown, but just can directly predict now with model f (X), Ynew=f (Xnew).
(9) if reality is applied, find the precision needing to improve model, start expert mode, the basic model in framework and framework, startup cross validation are manually set, obtain new frame model.
(10) the SVM model of Comparison framework model and general mode, selects wherein error reckling.
(11) if finding that certain model is that this device is not to be covered in actual applications, and excellent performance, it being worth subsequent reuse, the interface that the present invention provides allows user add model.This is professional mode.
It should be appreciated that the above-mentioned detailed description of the invention of the present invention is used only for exemplary illustration or explains the principle of the present invention, and it is not construed as limiting the invention.Therefore, any modification, equivalent substitution and improvement etc. done in the case of without departing from the spirit and scope of the present invention, should be included within the scope of the present invention.Additionally, claims of the present invention be intended to fall in the equivalents on scope and border or this scope and border whole change and modifications example.
Claims (7)
1. the modeling method of automatic machinery based on a GPU study, it is characterised in that comprise the following steps:
(1) set up model library, prepare data；
(2) characteristic of the data of input is normalized, selects multiple models of general mode to enter GPU parallel computation；
(3) according to input data, automatically preparing the data of cross validation, default is 10 foldings, 9 folding training, a folding validation test；
(4) calculating the cross validation of multiple model respectively by GPU, respectively obtain the error of each model, the less model of Select Error is as the final model selected；
(5) using all data as training data, the final model selected of training, obtain final model, remember into Ypre=f (X).
The modeling method of automatic machinery based on GPU the most according to claim 1 study, it is characterised in that described model library also includes expert mode, manually arranges the basic model in framework and framework, starts cross validation, obtains new frame model.
The modeling method of automatic machinery based on GPU the most according to claim 1 study, it is characterised in that described model library also includes professional mode, it is provided that the interface of new model algorithm, for adding new model.
The modeling method of automatic machinery based on GPU the most according to claim 1 study, it is characterized in that, described general mode at least includes perceptron model, linear regression model (LRM), logistic regression models, supporting vector machine model, Back propagation neural networks model and feedback Hopfield network model.
The modeling method of automatic machinery based on GPU the most according to claim 2 study, it is characterised in that the framework of described expert mode includes Bag, Adaboost, Randon Forest, Deep Learning；Basic model includes decision tree, unsupervised learning K-means algorithm.
The modeling method of automatic machinery based on GPU the most according to claim 1 study, it is characterised in that described GPU parallel computation includes different model parallel computations sample set parallel computation different with during cross validation.
The modeling method of automatic machinery based on GPU the most according to claim 1 study, it is characterised in that described data form is that (X, y), wherein X is characterized, and y is classification or predictive value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510184147.5A CN106067028A (en) | 2015-04-19 | 2015-04-19 | The modeling method of automatic machinery based on GPU study |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510184147.5A CN106067028A (en) | 2015-04-19 | 2015-04-19 | The modeling method of automatic machinery based on GPU study |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106067028A true CN106067028A (en) | 2016-11-02 |
Family
ID=57419584
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510184147.5A Pending CN106067028A (en) | 2015-04-19 | 2015-04-19 | The modeling method of automatic machinery based on GPU study |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106067028A (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107908536A (en) * | 2017-11-17 | 2018-04-13 | 华中科技大学 | To the performance estimating method and system of GPU applications in CPU GPU isomerous environments |
| CN108062509A (en) * | 2017-10-30 | 2018-05-22 | 广东工业大学 | A kind of intelligent video camera head emergent dialing method based on machine learning |
| CN108710949A (en) * | 2018-04-26 | 2018-10-26 | 第四范式（北京）技术有限公司 | The method and system of template are modeled for creating machine learning |
| CN108830850A (en) * | 2018-06-28 | 2018-11-16 | 信利（惠州）智能显示有限公司 | Automatic optics inspection picture analyzing method and apparatus |
| CN109063576A (en) * | 2018-07-05 | 2018-12-21 | 北京泛化智能科技有限公司 | Management method and device for flight movement node |
| CN109101547A (en) * | 2018-07-05 | 2018-12-28 | 北京泛化智能科技有限公司 | Management method and device for wild animal |
| CN109145942A (en) * | 2018-07-05 | 2019-01-04 | 北京泛化智能科技有限公司 | Image processing method and device for intelligent recognition |
| WO2019114413A1 (en) * | 2017-12-11 | 2019-06-20 | 北京三快在线科技有限公司 | Model training |
| CN110235137A (en) * | 2017-02-24 | 2019-09-13 | 欧姆龙株式会社 | Learning data obtains device and method, program and storage medium |
| WO2019233231A1 (en) * | 2018-06-08 | 2019-12-12 | 上海寒武纪信息科技有限公司 | General machine learning model, and model file generation and parsing method |
| WO2020147450A1 (en) * | 2019-01-15 | 2020-07-23 | 探智立方(北京)科技有限公司 | Ai model automatic generation method based on computational graph evolution |
| CN111475775A (en) * | 2020-04-14 | 2020-07-31 | 腾讯科技（深圳）有限公司 | Data processing method, text processing method, device and equipment of graphic processor |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101782976A (en) * | 2010-01-15 | 2010-07-21 | 南京邮电大学 | Automatic selection method for machine learning in cloud computing environment |
| CN103426007A (en) * | 2013-08-29 | 2013-12-04 | 人民搜索网络股份公司 | Machine learning classification method and device |
| CN103502899A (en) * | 2011-01-26 | 2014-01-08 | 谷歌公司 | Dynamic predictive modeling platform |
| CN104063744A (en) * | 2014-04-15 | 2014-09-24 | 浙江大学 | Time series prediction method for medicine consumption |
-
2015
- 2015-04-19 CN CN201510184147.5A patent/CN106067028A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101782976A (en) * | 2010-01-15 | 2010-07-21 | 南京邮电大学 | Automatic selection method for machine learning in cloud computing environment |
| CN103502899A (en) * | 2011-01-26 | 2014-01-08 | 谷歌公司 | Dynamic predictive modeling platform |
| CN103426007A (en) * | 2013-08-29 | 2013-12-04 | 人民搜索网络股份公司 | Machine learning classification method and device |
| CN104063744A (en) * | 2014-04-15 | 2014-09-24 | 浙江大学 | Time series prediction method for medicine consumption |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110235137A (en) * | 2017-02-24 | 2019-09-13 | 欧姆龙株式会社 | Learning data obtains device and method, program and storage medium |
| CN108062509A (en) * | 2017-10-30 | 2018-05-22 | 广东工业大学 | A kind of intelligent video camera head emergent dialing method based on machine learning |
| CN107908536B (en) * | 2017-11-17 | 2020-05-19 | 华中科技大学 | Performance evaluation method and system for GPU application in CPU-GPU heterogeneous environment |
| CN107908536A (en) * | 2017-11-17 | 2018-04-13 | 华中科技大学 | To the performance estimating method and system of GPU applications in CPU GPU isomerous environments |
| WO2019114413A1 (en) * | 2017-12-11 | 2019-06-20 | 北京三快在线科技有限公司 | Model training |
| CN108710949A (en) * | 2018-04-26 | 2018-10-26 | 第四范式（北京）技术有限公司 | The method and system of template are modeled for creating machine learning |
| US11726754B2 (en) | 2018-06-08 | 2023-08-15 | Shanghai Cambricon Information Technology Co., Ltd. | General machine learning model, and model file generation and parsing method |
| WO2019233231A1 (en) * | 2018-06-08 | 2019-12-12 | 上海寒武纪信息科技有限公司 | General machine learning model, and model file generation and parsing method |
| CN108830850A (en) * | 2018-06-28 | 2018-11-16 | 信利（惠州）智能显示有限公司 | Automatic optics inspection picture analyzing method and apparatus |
| CN108830850B (en) * | 2018-06-28 | 2020-10-23 | 信利（惠州）智能显示有限公司 | Automatic optical detection picture analysis method and equipment |
| CN109063576A (en) * | 2018-07-05 | 2018-12-21 | 北京泛化智能科技有限公司 | Management method and device for flight movement node |
| CN109145942A (en) * | 2018-07-05 | 2019-01-04 | 北京泛化智能科技有限公司 | Image processing method and device for intelligent recognition |
| CN109101547B (en) * | 2018-07-05 | 2021-11-12 | 北京泛化智能科技有限公司 | Management method and device for wild animals |
| CN109063576B (en) * | 2018-07-05 | 2021-12-17 | 北京泛化智能科技有限公司 | Management method and device for flight action nodes |
| CN109145942B (en) * | 2018-07-05 | 2022-02-01 | 北京泛化智能科技有限公司 | Image processing method and device for intelligent recognition |
| CN109101547A (en) * | 2018-07-05 | 2018-12-28 | 北京泛化智能科技有限公司 | Management method and device for wild animal |
| WO2020147450A1 (en) * | 2019-01-15 | 2020-07-23 | 探智立方(北京)科技有限公司 | Ai model automatic generation method based on computational graph evolution |
| CN111475775A (en) * | 2020-04-14 | 2020-07-31 | 腾讯科技（深圳）有限公司 | Data processing method, text processing method, device and equipment of graphic processor |
| CN111475775B (en) * | 2020-04-14 | 2023-09-15 | 腾讯科技（深圳）有限公司 | Data processing method, text processing method, device and equipment of graphic processor |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106067028A (en) | The modeling method of automatic machinery based on GPU study | |
| Zhang et al. | Deep learning-enabled intelligent process planning for digital twin manufacturing cell | |
| Gaier et al. | Data-efficient design exploration through surrogate-assisted illumination | |
| US20230082597A1 (en) | Neural Network Construction Method and System | |
| KR101729694B1 (en) | Method and Apparatus for Predicting Simulation Results | |
| CN113361680B (en) | Neural network architecture searching method, device, equipment and medium | |
| TW201903624A (en) | Training task optimization system, training task optimization method and non-transitory computer readable medium thereof | |
| CN111611085B (en) | Yun Bian collaboration-based man-machine hybrid enhanced intelligent system, method and device | |
| CN113128671B (en) | Service demand dynamic prediction method and system based on multi-mode machine learning | |
| KR20170078256A (en) | Method and apparatus for time series data prediction | |
| CN113505883A (en) | Neural network training method and device | |
| CN113326852A (en) | Model training method, device, equipment, storage medium and program product | |
| WO2022076797A1 (en) | Computer architecture for generating footwear digital asset | |
| CN112052027A (en) | Method and device for processing AI task | |
| CN114127803A (en) | Multi-method system for optimal prediction model selection | |
| CN110222734B (en) | Bayesian network learning method, intelligent device and storage device | |
| CN113378414A (en) | Cornea shaping lens fitting method, device, equipment and readable storage medium | |
| Zhang et al. | Interactive AI with Retrieval-Augmented Generation for Next Generation Networking | |
| Laskar et al. | Artificial Neural Networks and Gene Expression Programing based age estimation using facial features | |
| Andrés-Pérez et al. | Aerodynamic shape design by evolutionary optimization and support vector machines | |
| CN108898648A (en) | A kind of K line chart building method, system and relevant device | |
| Li | Prediction of body temperature from smart pillow by machine learning | |
| CN106934480A (en) | Insure grade analysis method, server and terminal | |
| Ghosh et al. | Fast approximate Bayesian computation for estimating parameters in differential equations | |
| Qiuju | PERSONAL CREDIT SCORING MODEL RESEARCHBASED ON THE RF-GA-SVM MODEL |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20161102 |
|
| RJ01 | Rejection of invention patent application after publication |