Method and device for learning rate calculation, and method and device for classification model calculation

A calculation method and technology of classification model, applied in the direction of calculation model, calculation, instrument, etc., can solve the problems of poor data classification effect, unable to obtain optimal model, easy to fall into local optimal solution and so on

Pending Publication Date: 2018-11-27
CHENGDU SEFON SOFTWARE CO LTD
View PDF0 Cites 2 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0004] One of the main defects of the gradient progressive regression tree algorithm is that it is easy to fall into a local optimal solution when solving the learning rate of the cumulative model, and the optimal model cannot be obtained, which leads to poor classification of some data

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Method and device for learning rate calculation, and method and device for classification model calculation
  • Method and device for learning rate calculation, and method and device for classification model calculation
  • Method and device for learning rate calculation, and method and device for classification model calculation

Examples

Experimental program
Comparison scheme
Effect test

no. 1 example

[0076] The classification model calculation method provided by the first embodiment of the present invention is used to train a gradient progressive regression tree model F(x), which is used to classify data x, and F(x) can also be regarded as x obtained through the prediction of the model The classification result, and the actual classification result of x is recorded as y, and y is also called the label of x. The degree of error between y and F(x) is defined by the loss function L(y, F(x)). Usually, the smaller the loss function, the better the performance of the model. Common loss functions include 0-1 loss function, absolute loss function, and absolute loss function. Value loss function, logarithmic loss function, Hinge loss function, etc.

[0077] The gradient asymptotic regression tree model F(x) is an accumulation model, which can be obtained by iterative formula F m (x)=F m-1 (x)+γ m h m (x) to define. where m is the number of iterations, F m-1 (x) is the result ...

no. 2 example

[0130] Figure 4 A functional block diagram of the learning rate calculation apparatus 200 provided by the second embodiment of the present invention is shown. refer to Figure 4 , the device includes an empirical risk acquisition module 210 and a learning rate acquisition module 220 .

[0131] Wherein, the empirical risk obtaining module 210 is used for obtaining the empirical risk of the classification model used for classifying the data, and the parameters of the empirical risk include the learning rate used for iteratively calculating the classification model;

[0132] The learning rate obtaining module 220 is configured to iteratively calculate the learning rate based on random walk, and obtain the value of the learning rate when the empirical risk takes the minimum value.

[0133] The image feature extraction device 200 provided by the second embodiment of the present invention has the same implementation principle and technical effects as the parts about the learning ...

no. 3 example

[0135] Figure 5 A functional block diagram of the classification model computing apparatus 300 provided by the third embodiment of the present invention is shown. refer to Figure 5 , the device includes an initialization module 310 , a fitting module 320 , a learning rate calculation module 330 , an iteration module 340 and a result determination module 350 .

[0136] The initialization module 310 is used to determine the number of iterations M, the initialization model of F(x), and the empirical risk J(γ) of F(x), where γ is the learning rate for iterative calculation of F(x), and M is the positive integer;

[0137] The fitting module 320 is used to take the iteration number m as 1 to M, and at the mth iteration, fit the decision regression tree and denote the decision regression tree as h m (x);

[0138] The learning rate calculation module 330 is configured to use the learning rate calculation method provided by the first aspect or any possible implementation manner o...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

PUM

No PUM Login to view more

Abstract

The invention relates to the technical field of data classification, and provides a method and a device for learning rate calculation, and a method and a device for classification model calculation. The learning rate calculation method comprises: obtaining empirical risk of a classification model for classifying data, parameters of the empirical risk including learning rate for iteratively calculating the classification model; performing iterative computation on the learning rate based on random walk, and obtaining a learning rate value when the empirical risk is minimum. Because random walksare introduced into the iterative computation of learning rate, the learning rate obtained from the computation can converge the empirical risk which is used as an optimization object to a global optimal solution rather than a local optimal solution. Furthermore, the classification model can obtain a higher precision model when the model performs iterative calculation based on the learning rate, so as to improve accuracy of data classification results. The method for classification model calculation is used to calculate a gradient progressive regression tree model, and the method for learningrate calculation is used when the learning rate of the model is calculated.

Description

technical field [0001] The present invention relates to the technical field of data classification, and in particular, to a learning rate calculation method and device, and a classification model calculation method and device. Background technique [0002] Machine Learning (ML), as a multi-domain interdisciplinary subject, is widely used in data mining, big data and other fields. Common machine learning algorithms include: classification algorithm, clustering algorithm, neural network, reinforcement learning, etc. Among them, the classification algorithm is a supervised learning method, which is a very important task in data mining. The purpose is to learn a classifier based on the training samples, so as to map the data to a certain category (value) in a given category . [0003] Common classification algorithms include decision trees, logistic regression, and combinatorial classifiers. Gradient asymptotic regression tree is a combination classifier, proposed by Freidman...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

Application Information

Patent Timeline
no application Login to view more
Patent Type & Authority Applications(China)
IPC IPC(8): G06N99/00
Inventor 覃进学王纯斌詹雪薇
Owner CHENGDU SEFON SOFTWARE CO LTD
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap