CN107679584A - Data classification method and system based on integrated resource allocation network RAN - Google Patents
Data classification method and system based on integrated resource allocation network RAN Download PDFInfo
- Publication number
- CN107679584A CN107679584A CN201711022308.6A CN201711022308A CN107679584A CN 107679584 A CN107679584 A CN 107679584A CN 201711022308 A CN201711022308 A CN 201711022308A CN 107679584 A CN107679584 A CN 107679584A
- Authority
- CN
- China
- Prior art keywords
- sample
- training sample
- ran
- training
- entered
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Abstract
The embodiment of the present invention provides a kind of data classification method and system based on integrated resource allocation network RAN, to solve the larger technical problem of existing training sample demand when being trained to RAN in the prior art.Wherein, method includes handling each training sample in training sample set based on the input weights set of default sample, obtain training sample set to be entered, P RAN graders to be trained in RAN are trained based on the training sample set to be entered, P RAN grader after being trained;The data to be sorted are classified based on P RAN grader after the training, and output category result.
Description
Technical field
The present invention relates to field of computer technology, more particularly to a kind of data based on integrated resource allocation network RAN point
Class method and system.
Background technology
In the last few years, artificial intelligence, particularly machine learning, achieve the extensive concern of academia and industrial quarters and answer
With.Machine learning refers to that algorithm dynamically adjusts model structure or parameter using data so that model have Treatment Analysis this
The ability of a kind of data, this is also the intension place of " study ".During model learning, the quality and quantity of training data
Have a great impact to model.Correspondingly, data collection and processing also become particularly important.But in many Practical Projects
In, the collection of data is the work taken time and effort very much, it is difficult to collect enough data volumes or sufficiently accurate
Data value.Such as the fault data of some diseases data and aero-engine, because the incidence of itself is especially low, data volume
Collection become extremely difficult.Further, since the sensor of measurement or the problem of measuring method, cause in the data that collect
More noise signal be present.Under this problem, it is more high-precision how using data limited amount, comprising noise to obtain one
Degree, stable machine learning model, also becomes current study hotspot.
Resource allocation network (Resource Allocating Network, RAN) is that a kind of incremental learning realizes mould
Type, the quantity of its network hidden layer node can enter Mobile state increase according to the complexity of training sample, realize higher calculating
Ability, therefore also received in industrial quarters and be widely applied exploration interest.Meanwhile integrated study can also significantly improve machine learning
Robustness and computational accuracy of the model on small-scale training dataset.But the existing integrated learning approach based on RAM is also
There are problems that.
Generally only each RAN is trained using the training sample of single classification or pattern in the prior art, and made
Single RAN is only trained using the fraction of training sample during with the training sample of single classification or pattern, therefore to total
Training sample demand it is larger, can not be applied to Small Sample Database collection on.
In summary, the larger technical problem of training sample demand when being trained to RAN in the prior art be present.
The content of the invention
The embodiment of the present invention provides a kind of data classification method and system based on integrated resource allocation network RAN, to
Solves the larger technical problem of existing training sample demand when being trained to RAN in the prior art.
In a first aspect, the embodiment of the present invention provides a kind of data classification method based on integrated resource allocation network RAN, bag
Include:
Each training sample in training sample set is handled based on the input weights set of default sample, treated
Input training sample set;Wherein, the default sample input weights set includes P M dimension sample input weight vector, described
Each training sample in training sample set includes M dimension sampling feature vectors, and the M dimensions sampling feature vectors correspondingly indicate phase
The M kind sample properties of training sample are answered, a training sample to be entered in the training sample set to be entered is tieed up by a M
Sample inputs weight vector and a training sample determines, each training sample to be entered includes M sample components, described P, M
For the integer more than or equal to 1;
P RAN graders to be trained in RAN are trained based on the training sample set to be entered, instructed
P RAN grader after white silk;
The data to be sorted are classified based on P RAN grader after the training, and output category result.
It is described that weights set is inputted in training sample set based on default sample in a kind of possible implementation
Each training sample is handled, and obtains training sample set to be entered, including:
Each training sample in training sample set is handled based on the input weights set of default sample, obtained more
Individual training sample to be entered;Wherein, when obtaining a training sample to be entered in the multiple training sample to be entered, perform
Operate below:Based on k-th of sample on the M dimension sample input weight vectors in the default sample input weights set
K-th of the sample inputted on the M dimension sampling feature vectors of a training sample in weights component, and the training sample set
Characteristic component, determine k-th of sample components on M sample components of one training sample to be entered;Wherein, k is successively
Take 1 to M integer;It is determined that the training sample being made up of the M sample components is one training sample to be entered;
It is determined that the collection being made up of the multiple training sample to be entered is combined into the training sample set to be entered.
It is described that P in RAN are waited to instruct based on the training sample set to be entered in a kind of possible implementation
Practice RAN graders to be trained, P RAN grader after being trained, including:
Based at least one training sample to be entered in the training sample set to be entered, P in RAN are waited to instruct
Any RAN graders to be trained practiced in RAN graders are trained, after being trained a RAN graders;
Determine a RAN grader after multiple training for P RAN grader after training.
In a kind of possible implementation, the P RAN grader based on after the training is treated grouped data and entered
Row classification, and output category result, including:
The data to be sorted are divided using each RAN graders in P RAN grader after the training
Class, obtain P output result;Wherein, each output result in the P output result is used to indicate the data to be sorted
The probability of the sample properties stressed when being classified by corresponding RAN graders;
Based on the P output result, the classification results are exported.
It is described to be based on the P output result in a kind of possible implementation, the classification results are exported, including:
The probability of identical output result in the P output result is counted respectively;
The probability highest output result for determining the identical output result is the classification results;
Export the classification results.
Second aspect, the embodiment of the present invention provide a kind of data sorting system, including:
Processing module, for being entered based on the input weights set of default sample to each training sample in training sample set
Row processing, obtains training sample set to be entered;Wherein, the default sample input weights set includes P M dimension sample input
Weight vector, each training sample in the training sample set include M dimension sampling feature vectors, and the M ties up sample characteristics
The corresponding M kind sample properties for indicating corresponding training sample of vector, an instruction to be entered in the training sample set to be entered
Practice sample and determine that each training sample to be entered includes M sample by a M dimension sample input weight vector and a training sample
This component, described P, M are the integer more than or equal to 1;
Training module, for being entered based on the training sample set to be entered to P RAN graders to be trained in RAN
Row training, P RAN grader after being trained;
Sort module, classified for treating grouped data based on P RAN grader after the training, and exported
Classification results.
In a kind of possible implementation, the processing module is used for:
Each training sample in training sample set is handled based on the input weights set of default sample, obtained more
Individual training sample to be entered;Wherein, when obtaining a training sample to be entered in the multiple training sample to be entered, perform
Operate below:Based on k-th of sample on the M dimension sample input weight vectors in the default sample input weights set
K-th of the sample inputted on the M dimension sampling feature vectors of a training sample in weights component, and the training sample set
Characteristic component, determine k-th of sample components on M sample components of one training sample to be entered;Wherein, k is successively
Take 1 to M integer;It is determined that the training sample being made up of the M sample components is one training sample to be entered;
It is determined that the collection being made up of the multiple training sample to be entered is combined into the training sample set to be entered.
In a kind of possible implementation, the training module is used for:
Based at least one training sample to be entered in the training sample set to be entered, P in RAN are waited to instruct
Any RAN graders to be trained practiced in RAN graders are trained, after being trained a RAN graders;
Determine a RAN grader after multiple training for P RAN grader after training.
In a kind of possible implementation, the sort module is used for:
The data to be sorted are divided using each RAN graders in P RAN grader after the training
Class, obtain P output result;Wherein, each output result in the P output result is used to indicate the data to be sorted
The probability of the sample properties stressed when being classified by corresponding RAN graders;
Based on the P output result, the classification results are exported.
In a kind of possible implementation, the sort module is specifically used for:
The probability of identical output result in the P output result is counted respectively;
The probability highest output result for determining the identical output result is the classification results;
Export the classification results.
The third aspect, the embodiment of the present invention provide a kind of computer installation, including:
At least one processor, and
The memory that is connected with least one processor communication, communication interface;
Wherein, have can be by the instruction of at least one computing device, at least one place for the memory storage
Reason device utilizes the communication interface to perform method as described in relation to the first aspect by performing the instruction of the memory storage.
Fourth aspect, the embodiment of the present invention provide a kind of computer-readable recording medium, including:
The computer-readable recording medium storage has computer instruction, when the computer instruction is run on computers
When so that computer performs method as described in relation to the first aspect.
The embodiment of the present invention provides a kind of data classification method based on integrated resource allocation network RAN, by presetting sample
The set of this input weights is handled each training sample in training sample set, obtains training sample set to be entered,
Then parallel training is carried out to P RAN graders to be trained in RAN according to the training sample set to be entered of acquisition, obtained
P RAN grader after training, and then grouped data is treated using P RAN grader after training and classified, and export
Classification results.Due to the training sample to be entered two-by-two in treated training sample set to be entered in the embodiment of the present invention
Between have differences, i.e., only need the training sample of small data set to can be obtained by various instruction to be entered in the embodiment of the present invention
Practice sample, thus it is relatively low to the demand of training sample.
Brief description of the drawings
In order to illustrate the technical solution of the embodiments of the present invention more clearly, it will make below to required in the embodiment of the present invention
Accompanying drawing is briefly described, it should be apparent that, accompanying drawing described below is only some embodiments of the present invention, for
For those of ordinary skill in the art, on the premise of not paying creative work, other can also be obtained according to these accompanying drawings
Accompanying drawing.
Fig. 1 is the system architecture diagram of data sorting system in the embodiment of the present invention;
Fig. 2 is the model schematic of a RAN grader in the embodiment of the present invention;
Fig. 3 is the schematic flow sheet of the data classification method based on integrated resource allocation network RAN in the embodiment of the present invention;
Fig. 4 is the module diagram of data sorting system in the embodiment of the present invention;
Fig. 5 is the structural representation of Computer device of the embodiment of the present invention.
Embodiment
In order that the purpose, technical scheme and advantage of the embodiment of the present invention are clearer, implement below in conjunction with the present invention
Accompanying drawing in example, the technical scheme in the embodiment of the present invention is clearly and completely described.
First, the system architecture applied to the embodiment of the present invention is introduced, so as to skilled artisan understands that.
Fig. 1 is referred to, for the system architecture diagram of the data sorting system of application General layout Plan of the embodiment of the present invention.Under
Each structure in Fig. 1 and mark are simply introduced from left to right in face.
In Fig. 1, x can represent the training sample set in the embodiment of the present invention, can include one or more training
Sample, the quantity of training sample can determine that the embodiment of the present invention is not particularly limited according to practical application.Training sample can be with
For pixel and corresponding pixel value, text data etc..
ExtremelyThe collection of composition is combined into the default sample input weights set in the embodiment of the present invention, wherein, with
Exemplified by,Generally recorded in vector form during individualism, i.e. the sample input weight vector of M dimensions.
RAN can include P subnet, i.e., the RAN1 to RANp shown in Fig. 1, and RAN1 to RANp is claimed in the embodiment of the present invention
In each RAN be RAN graders.
O (1) a to o (i) in o (p) can be corresponded to and be represented output result corresponding to a RAN multiple output results
Set, and y (1) a to y (i) in y (p) can represent to carry out normalizing to each output result in an output result set
Corresponding set after change processing, wherein, p is the integer more than or equal to 1, and i takes 1 to p integer.Voting (VOTE) expression can be right
Y (1) to y (p) final output result is voted, and obtains last classification results, i.e. in Fig. 1
For ease of understanding, the model of a RAN grader in the embodiment of the present invention is simply introduced below, please
Referring to Fig. 2, for the model schematic of a RAN grader in the embodiment of the present invention.
As shown in Figure 2, each RAN graders are considered as the RAN neutral nets with three-decker, it is assumed that input, be defeated
It is respectively M, L to go out a layer dimension, hidden layer node, i.e. c1 to cN, quantity N may in the training process, with the instruction of input
The novelty for practicing sample occurs to be incremented by;b(1),b(2),...,b(L)Each component corresponding to the amount of bias of output layer can be represented.
For example, for the input set X={ x of T training sample1,x2,...,xT, wherein, each training sample
xiSampling feature vectors are tieed up for M, the M kind sample properties for indicating corresponding training sample can be corresponded to, can be mathematically expressed as:
Wherein, [] ' transposition computing can be represented,The determinant that M rows * 1 is arranged can be represented, it is each in determinant
Component can be real number, similarly hereinafter.
So, the center vector of j-th of node of hidden layer node can be expressed as:
The center width vector of j-th of node of hidden layer node can be expressed as:
The amount of bias of output layer can be expressed as:
B=[b(1),b(2),...,b(L)]′
Therefore, corresponding to input quantity xiNetwork output be:
In calculating process, the state of j-th of node in hidden layer node can be expressed as:
Wherein, | | | | modulo operation can be expressed as,J-th of component on training sample is represented,Represent hidden
J-th of component on the center vector of j-th of node containing node layer,Represent the center of j-th of node of hidden layer node
J-th of component on width vector.
The output of l-th of output node layer can be expressed as:
For more classification problems, classified using softmax algorithms,
Final classification exports:
The preferred embodiment of the present invention is described in detail below in conjunction with the accompanying drawings.
Embodiment one
Fig. 3 is referred to, the embodiment of the present invention provides a kind of data classification method based on integrated resource allocation network RAN,
It can apply in the data sorting system shown in Fig. 1, wherein, the implementation process of method can be described as follows:
S101:Each training sample in training sample set is handled based on the input weights set of default sample,
Obtain training sample set to be entered;Wherein, presetting sample input weights set includes P M dimension sample input weight vector, instruction
Practicing each training sample in sample set includes M dimension sampling feature vectors, the corresponding training of the corresponding instruction of M dimension sampling feature vectors
The M kind sample properties of sample, a training sample to be entered in training sample set to be entered tie up sample input power by a M
Value vector determines that each training sample to be entered includes M sample components with a training sample, and P, M are whole more than or equal to 1
Number;
S102:P RAN graders to be trained in RAN are trained based on training sample set to be entered, instructed
P RAN grader after white silk;
S103:Grouped data is treated based on P RAN grader after training to be classified, and output category result.
In the embodiment of the present invention, in order to obtain laying particular emphasis on what the different sample properties for treating grouped data were classified
Different RAN graders, default sample input weights set can be set, for each training sample in training sample set
This is handled, and to obtain training sample set to be entered, and then can use the training sample set to be entered obtained to RAN
RAN graders in network are trained, and obtain multiple different RAN graders.
In S101, presetting the set of sample input weights can be included as shown in Figure 1ExtremelyThat is P M dimension
Sample inputs weight vector, and each M dimensions sample input weight vector includes M weights component, its dimension and each training sample
Sampling feature vectors dimension it is corresponding.
Wherein, it can be relevant parameter of the data sorting system according to training sample to preset sample input weights set, such as
Training sample number, abundant degree of training sample etc., what random initializtion obtained, and subsequently to the instruction of RAN graders
Can be changeless during white silk.
Or default sample input weights set can also be by the self-defined setting of engineer.Default sample input power
The number for the sample input weight vector that value set includes is corresponding with the RAN RAN to be trained included number, i.e. a sample
Input the corresponding RAN to be trained of weight vector.
Training sample set can include one or more training sample, and the quantity of training sample can answer according to actual
With determination.For example under initial situation, training samples number may limited, the training samples number that training sample set includes
May be exactly limited, a small amount of, and as the continuous accumulation of training data, training samples number increase, then training sample gathers
The training samples number that closing includes also constantly increases.
One training sample is considered as a M dimension sampling feature vectors, such as foregoing described training sample xi.Citing comes
Say, it is assumed that training sample includes 10*10, totally 100 pixels, then the training sample is considered as one 100 dimension sample characteristics
Vector;Because pixel value corresponding to each pixel can be different, therefore, 100 dimension sampling feature vectors can correspond to instruction should
100 kinds of sample properties of training sample.
For example pixel value corresponding to 100 pixels and each pixel may be constructed the picture for including alphabetical A;Or
Pixel value corresponding to 100 pixels and each pixel can also form picture including letter b etc., and including alphabetical A's
There may be between 100 kinds of sample characteristics of picture and 100 kinds of sample characteristics of the picture including letter b part it is identical or
May be entirely different.
In a kind of possible implementation, weights set can be inputted according to default sample in training sample set
Each training sample is handled, and then obtains training sample set to be entered.
Below exemplified by obtaining a training sample to be entered in training sample set to be entered:
The M dimension sample input weight vectors that can be inputted according to default sample in weights set, and training sample set
A training sample in conjunction, determines a training sample to be entered.
Due to the corresponding M dimensions sample input weight vector of each RAN graders to be trained, by training sample to be entered
Input before RAN graders, each training sample in training sample set can be handled.
Assuming that i-th of pending training sample in training sample set is xi, including 100 pixels, mathematical table
It is shown as:Wherein,Correspond to respectively and represent the 1st
To the 100th pixel, each pixel is corresponding with pixel value;The determinant of 100 row * 1 row, and ranks can be represented
Each component in formula can be real number.
RAN graders to be trained are the RAN1 in Fig. 1, and corresponding M dimensions sample input weight vector isIts mathematical table
Up to for:
Therefore, sample input weight vector can be tieed up according to MTo training sample xiHandled, obtain one treat it is defeated
Enter training sampleMathematics can be expressed as:
In a kind of possible implementation, a training sample to be entered is determined it can be seen from above-mentioned mathematic(al) representation
This, k-th of the sample that can be tieed up according to a M on sample input weight vector inputs weights component, and the M of a training sample
K-th of sample characteristics component on sampling feature vectors is tieed up, determines the on M sample components of a training sample to be entered
K sample components;Wherein, k takes 1 to M integer successively;Such as on M sample components of a training sample to be entered
K sample components can be expressed as:
Then, the M sample components determined through the above way may be constructed a training sample to be entered.
After training sample set to be entered is obtained by the above method, S102 can be entered, you can with according to be entered
Training sample set is trained to P RAN graders to be trained in RAN, P RAN grader after being trained.
After being handled by the above method training sample set, appointing in the training sample set to be entered of acquisition
One sample to be entered is had differences when RAN graders trains in input P, therefore, in the individual RAN graders to be trained of P
The training of any two RAN graders to be trained independently is carried out, and then allows the RAN graders after each training only
Stand and classified for treating grouped data.
And because any training sample to be entered in the training sample set to be entered of acquisition in input P is waited to train
Had differences during RAN graders, therefore the RAN graders after each training have when treating grouped data and being classified
At least one sample properties of its data to be sorted stressed.
For example, any RAN graders in P RAN grader after training, which are laid particular emphasis on, treats grouped data
A kind of sample characteristics are classified, such as, by taking the RAN1-RANp in Fig. 1 as an example, it is assumed that p 26, then RAN1-RANp may divide
Do not lay particular emphasis on to this 26 alphabetical classification of A-Z;Any RAN graders in P RAN grader after training are laid particular emphasis on pair
Two kinds of data to be sorted or two or more sample characteristics are classified, such as, it is false exemplified by the RAN1-RANp in still Fig. 1
If p is 10, then in RAN1-RAN10, RAN1 may be laid particular emphasis on to classify to A, B and C letter, corresponding A, B and C letter
Stressing weight may be identical possible different;And RAN2 may lay particular emphasis on the classification to B and D letters, B and the D weight that stresses may
Identical possible different, RAN3-RAN10 is similar, and the embodiment of the present invention does not repeat.
In a kind of possible implementation, any RAN graders to be trained in P RAN graders to be trained
When being trained, at least one training sample to be entered in training sample set to be entered can be used, obtains corresponding instruct
RAN graders after white silk.
Due to after being handled by the above method training sample set, in the training sample set to be entered of acquisition
Any training sample to be entered had differences in input P when RAN graders are trained, i.e., in training sample set to be entered
Any two training sample to be entered between there may be difference, P is being waited to train using training sample set to be entered
Any one in RAN graders can be used in training sample set to be entered when training RAN graders to be trained
One or more training sample to be entered, and then had differences between P RAN grader after the training for obtain, carry
High RAN treats nicety of grading when grouped data is classified.
Generally use is based on the RAN integrated studies with long-term memory (Long-Term Memory, LTM) in the prior art
Method, and the input of next RAN networks is calculated using AdaBoost.M1 algorithms according to the output error of current RAN networks
Sample weights.The computational accuracy of this method is better than the algorithm model of single grader.However, because this method trains the latter
RAN graders need to use current RAN output error, and single RAN graders can only be trained at each time point, can not be simultaneously
Row trains all RAN set, it is therefore desirable to consumes the substantial amounts of training time.
Meanwhile also there is document to propose a kind of RAN system integrating methods based on Bagging technologies, and in order to ensure
The RAN networks integrated have certain otherness, and each RAN networks are simply trained with the sample data of N-1 classification,
Wherein, N is the categorical measure of sample label altogether.For some RAN sub-network in integrating, if the sample of input and the net
The distance between nearest node is more than a certain threshold value in network, then shows when the sample label of input is not belonging to the network training
Label space.For each RAN sub-networks, the selection of the threshold value becomes particularly important, and the threshold level is dependent on being made
The training sample of N-1 classification, and the feature Distribution value of the training sample of remaining 1 classification.Therefore, the threshold value
Selection will become very difficult and cumbersome, or even can not obtain, and training effectiveness is relatively low.
And in embodiments of the present invention, each RAN graders can be trained by the way of parallel training, you can with
RAN1-RANp graders in Fig. 1 are trained simultaneously, it is only necessary to seldom time can to complete whole training process,
Improve training effectiveness.
In a kind of possible implementation, grouped data is treated based on P RAN grader after training and classified,
And output category result, can by but be not limited only in the following manner carry out:
Because each RAN graders in P RAN grader after training can be independently used for treating grouped data
Classification, therefore, can treat grouped data using each RAN graders in P RAN grader after training simultaneously and be divided
Class, obtain P output result;Wherein, each output result in P output result is used to indicate data to be sorted by corresponding
The probability for the sample properties that RAN graders stress when being classified.
For example data to be sorted are the picture for including alphabetical A, and P RAN grader after training may be incorporated for identifying
26 letters, wherein, RAN1 identifies that alphabetical A precision is 80%, and to remaining 25 alphabetical accuracy of identification in addition to alphabetical A
Respectively less than 80%;RAN2 identification letter bs are 80%, and 25 alphabetical accuracy of identification of remaining in addition to letter b are respectively less than
80%;RAN3 identifies that alphabetical A is 60%, and is respectively less than 60% etc. to remaining 25 alphabetical accuracy of identification in addition to alphabetical A.
Then, the picture comprising alphabetical A is classified with P RAN grader after training, P output can be obtained
As a result, each output result of P output result here is the final output result of corresponding RAN graders.
Fig. 2 is may refer to, because each RAN graders are considered as the RAN neutral nets of three-decker, wherein, three layers
Structure is respectively input layer, hidden layer and output layer, respectively including multiple input nodes, multiple hidden layer nodes and multiple outputs
Node.
Using training sample to be enteredWhen being trained to RAN graders, for RAN
K-th of node on the input layer of grader, its input value can correspond toWherein,It is expressed as
K-th of component on corresponding sample input weight vector,It is expressed as k-th of component of training sample to be entered.
Therefore, multiple output nodes of each RAN graders may correspond to multiple output results, and each RAN graders
Final output result can by multiple output results vote produce.
For example final output result corresponding to RAN1 is A, final output result corresponding to RAN2 is B, corresponding to RAN3 most
Whole output result is A etc..
And then can be according to this P output result, output category result.
In a kind of possible implementation, based on P output result, output category result, can by but not only limit
Carried out in the following manner:The probability of identical output result in P output result is counted respectively, and then determines identical output result
Probability highest output result is classification results, and exports the classification results.
For example RAN includes the RAN graders after 100 training, when treating category images using RAN and being classified, essence
On be independently to treat category images using each RAN graders in the RAN graders after 100 training and classified, accordingly
Ground, 100 output results can be obtained.
Assuming that indicate that picture to be sorted is totally 20, picture for including alphabetical A in this 100 output results, then output result
For:Picture to be sorted is that picture, the probability for including alphabetical A are:20%;It is that the picture for including letter b is total to indicate picture to be sorted
30, then output result is:Picture to be sorted is that picture, the probability for including letter b are:30%;Indicate picture to be sorted for bag
Totally 50, the picture of letter C is included, then output result is:Picture to be sorted is that picture, the probability for including letter C are:50%.Enter
And final classification results can be obtained and be:Category images is the picture for including letter C.
In summary, one or more technical scheme of the embodiment of the present invention, has the following technical effect that or advantage:
Firstth, weights set is inputted by default sample to handle each training sample in training sample set,
Training sample set to be entered is obtained, then according to the training sample set to be entered of acquisition to P RAN to be trained in RAN
Grader carries out parallel training, P RAN grader after being trained, wherein the RAN graders after each training can be independent
Classified for treating grouped data, and then grouped data is treated using P RAN grader after training and classified, and
Output category result.Due to the sample to be entered two-by-two in treated training sample set to be entered in the embodiment of the present invention
Between have differences, i.e., only need the training sample of small data set to can be obtained by various sample to be entered in the embodiment of the present invention
This, thus it is relatively low to the demand of training sample.
Secondth, due to weights set in the embodiment of the present invention, can be inputted according to default sample in training sample set
Each training sample handled, any two training sample to be entered in this training sample set to be entered allowed for
Between have differences, after being trained using training sample set to be entered to P RAN graders to be trained, obtain
There is also difference between P RAN grader after training, i.e., the RAN graders after each training can be independently used for treating point
Class data are classified, and the RAN graders after a training lay particular emphasis on and treat a kind of sample properties of grouped data and divided
Class, and then improve the nicety of grading of data handling system.
3rd, in the embodiment of the present invention, a small amount of training sample can be used to carry out RAN graders under initial situation
Training, each RAN graders after training by starting working.And with the continuous accumulation of following training sample, it can carry out
The scale and parameter of line study constantly adjustment RAN graders, so that each subnet has good calculating performance, whole RAN
Integrated system possesses higher nicety of grading.
Embodiment two
Fig. 4 is referred to, based on same inventive concept, a kind of data sorting system, including place are provided in the embodiment of the present invention
Manage module 41, training module 42 and sort module 43.
Wherein, processing module 41, for inputting weights set to each instruction in training sample set based on default sample
Practice sample to be handled, obtain training sample set to be entered;Wherein, the default sample input weights set includes P M dimension
Sample inputs weight vector, and each training sample in the training sample set includes M dimension sampling feature vectors, the M dimensions
Sampling feature vectors correspondingly indicate the M kind sample properties of corresponding training sample, one in the training sample set to be entered
Training sample to be entered ties up sample input weight vector by a M and a training sample determines, each training sample to be entered
Including M sample components, described P, M are the integer more than or equal to 1;
Training module 42, for based on the training sample set to be entered to the P RAN graders to be trained in RAN
It is trained, P RAN grader after being trained;
Sort module 43, for being classified based on P RAN grader after the training to the data to be sorted,
And output category result.
In a kind of possible implementation, the processing module 41 is used for:
Each training sample in training sample set is handled based on the input weights set of default sample, obtained more
Individual training sample to be entered;Wherein, when obtaining a training sample to be entered in the multiple training sample to be entered, perform
Operate below:Based on k-th of sample on the M dimension sample input weight vectors in the default sample input weights set
K-th of the sample inputted on the M dimension sampling feature vectors of a training sample in weights component, and the training sample set
Characteristic component, determine k-th of sample components on M sample components of one training sample to be entered;Wherein, k is successively
Take 1 to M integer;It is determined that the training sample being made up of the M sample components is one training sample to be entered;
It is determined that the collection being made up of the multiple training sample to be entered is combined into the training sample set to be entered.
In a kind of possible implementation, the training module 42 is used for:
Based at least one training sample to be entered in the training sample set to be entered, P in RAN are waited to instruct
Any RAN graders to be trained practiced in RAN graders are trained, after being trained a RAN graders;
Determine a RAN grader after multiple training for P RAN grader after training.
In a kind of possible implementation, the sort module 43 is used for:
The data to be sorted are divided using each RAN graders in P RAN grader after the training
Class, obtain P output result;Wherein, each output result in the P output result is used to indicate the data to be sorted
The probability of the sample properties stressed when being classified by corresponding RAN graders;
Based on the P output result, the classification results are exported.
In a kind of possible implementation, the sort module 43 is specifically used for:
The probability of identical output result in the P output result is counted respectively;
The probability highest output result for determining the identical output result is the classification results;
Export the classification results.
Embodiment three
Fig. 5 is referred to, based on same inventive concept, a kind of computer installation is provided in the embodiment of the present invention, including at least
One processor 51, and the memory 52 and communication interface 53 communicated to connect with least one processor 51, in Fig. 5 with
Exemplified by one processor 51 is shown.
Wherein, the memory 52 is stored with the instruction that can be performed by least one processor 51, and described at least one
The instruction that individual processor 51 is stored by performing the memory 52, is performed such as institute in embodiment one using the communication interface 53
The method stated.
Example IV
Based on same inventive concept, the embodiment of the present invention provides a kind of computer-readable recording medium, and the computer can
Read storage medium and be stored with computer instruction, when the computer instruction is run on computers so that computer performs such as
Method described in embodiment one.
In specific implementation process, computer-readable recording medium includes:General serial bus USB
(UMiversal Serial Bus flash drive, USB), mobile hard disk, read-only storage (Read-OMlyPePory,
ROP), random access memory (RaMdoP Access PePory, RAP), magnetic disc or CD etc. are various can be with storage program
The storage medium of code.
Device embodiment described above is only schematical, wherein the units/modules illustrated as separating component
It can be or may not be physically separate, can be as the part that units/modules are shown or may not be
Physical location/module, you can with positioned at a place, or can also be distributed in multiple NE/modules.Can basis
It is actual to need to select some or all of module therein to realize the purpose of this embodiment scheme.Ordinary skill people
Member is not in the case where paying performing creative labour, you can to understand and implement.
Through the above description of the embodiments, those skilled in the art can be understood that each embodiment can
Realized by the mode of software plus required general hardware platform, naturally it is also possible to pass through hardware.Based on such understanding, on
The part that technical scheme substantially in other words contributes to prior art is stated to embody in the form of software product, should
Computer software product can store in a computer-readable storage medium, such as ROP/RAP, magnetic disc, CD, including some fingers
Make to cause a computer equipment (can be personal computer, server, or network equipment etc.) to perform each implementation
Method described in some parts of example or embodiment.
Finally it should be noted that:The above embodiments are merely illustrative of the technical solutions of the present invention, rather than its limitations;Although
The present invention is described in detail with reference to the foregoing embodiments, it will be understood by those within the art that:It still may be used
To be modified to the technical scheme described in foregoing embodiments, or equivalent substitution is carried out to which part technical characteristic;
And these modification or replace, do not make appropriate technical solution essence depart from various embodiments of the present invention technical scheme spirit and
Scope.
Claims (12)
- A kind of 1. data classification method based on integrated resource allocation network RAN, it is characterised in that including:Each training sample in training sample set is handled based on the input weights set of default sample, obtained to be entered Training sample set;Wherein, the default sample input weights set includes P M dimension sample input weight vector, the training Each training sample in sample set includes M dimension sampling feature vectors, and the M dimensions sampling feature vectors are corresponding to indicate corresponding instruction Practice the M kind sample properties of sample, a training sample to be entered in the training sample set to be entered ties up sample by a M Input weight vector and a training sample determines, each training sample to be entered includes M sample components, and described P, M are big In the integer equal to 1;P RAN graders to be trained in RAN are trained based on the training sample set to be entered, after being trained P RAN grader;Grouped data is treated based on P RAN grader after the training to be classified, and output category result.
- 2. the method as described in claim 1, it is characterised in that described that weights set is inputted to training sample based on default sample Each training sample in set is handled, and obtains training sample set to be entered, including:Each training sample in training sample set is handled based on the input weights set of default sample, obtains multiple treat Input training sample;Wherein, when obtaining a training sample to be entered in the multiple training sample to be entered, perform following Operation:Based on k-th of sample input on the M dimension sample input weight vectors in the default sample input weights set K-th of sample characteristics on the M dimension sampling feature vectors of a training sample in weights component, and the training sample set Component, determine k-th of sample components on M sample components of one training sample to be entered;Wherein, k take successively 1 to M integer;It is determined that the training sample being made up of the M sample components is one training sample to be entered;It is determined that the collection being made up of the multiple training sample to be entered is combined into the training sample set to be entered.
- 3. method as claimed in claim 1 or 2, it is characterised in that described to be based on the training sample set to be entered to RAN In P RAN graders to be trained be trained, P RAN grader after being trained, including:Based at least one training sample to be entered in the training sample set to be entered, P in RAN are waited to train Any RAN graders to be trained in RAN graders are trained, after being trained a RAN graders;Determine a RAN grader after multiple training for P RAN grader after training.
- 4. method as claimed in claim 3, it is characterised in that the P RAN grader based on after the training is treated point Class data are classified, and output category result, including:The data to be sorted are classified using each RAN graders in P RAN grader after the training, obtained Obtain P output result;Wherein, each output result in the P output result is used to indicate the data to be sorted by phase The probability of the sample properties stressed when answering RAN graders to be classified;Based on the P output result, the classification results are exported.
- 5. method as claimed in claim 4, it is characterised in that it is described to be based on the P output result, export the classification knot Fruit, including:The probability of identical output result in the P output result is counted respectively;The probability highest output result for determining the identical output result is the classification results;Export the classification results.
- 6. a kind of data sorting system, it is characterised in that the system includes:Processing module, at based on the input weights set of default sample to each training sample in training sample set Reason, obtains training sample set to be entered;Wherein, the default sample input weights set includes P M dimension sample input weights Vector, each training sample in the training sample set include M dimension sampling feature vectors, and the M ties up sampling feature vectors The corresponding M kind sample properties for indicating corresponding training sample, a training sample to be entered in the training sample set to be entered This ties up sample input weight vector by a M and a training sample determines, each training sample to be entered includes M sample point Amount, described P, M are the integer more than or equal to 1;Training module, for being instructed based on the training sample set to be entered to P RAN graders to be trained in RAN Practice, P RAN grader after being trained;Sort module, classified for treating grouped data based on P RAN grader after the training, and output category As a result.
- 7. system as claimed in claim 6, it is characterised in that the processing module is used for:Each training sample in training sample set is handled based on the input weights set of default sample, obtains multiple treat Input training sample;Wherein, when obtaining a training sample to be entered in the multiple training sample to be entered, perform following Operation:Based on k-th of sample input on the M dimension sample input weight vectors in the default sample input weights set K-th of sample characteristics on the M dimension sampling feature vectors of a training sample in weights component, and the training sample set Component, determine k-th of sample components on M sample components of one training sample to be entered;Wherein, k take successively 1 to M integer;It is determined that the training sample being made up of the M sample components is one training sample to be entered;It is determined that the collection being made up of the multiple training sample to be entered is combined into the training sample set to be entered.
- 8. system as claimed in claims 6 or 7, it is characterised in that the training module is used for:Based at least one training sample to be entered in the training sample set to be entered, P in RAN are waited to train Any RAN graders to be trained in RAN graders are trained, after being trained a RAN graders;Determine a RAN grader after multiple training for P RAN grader after training.
- 9. system as claimed in claim 8, it is characterised in that the sort module is used for:The data to be sorted are classified using each RAN graders in P RAN grader after the training, obtained Obtain P output result;Wherein, each output result in the P output result is used to indicate the data to be sorted by phase The probability of the sample properties stressed when answering RAN graders to be classified;Based on the P output result, the classification results are exported.
- 10. system as claimed in claim 9, it is characterised in that the sort module is specifically used for:The probability of identical output result in the P output result is counted respectively;The probability highest output result for determining the identical output result is the classification results;Export the classification results.
- 11. a kind of computer installation, it is characterised in that the computer installation includes:At least one processor, andThe memory that is connected with least one processor communication, communication interface;Wherein, have can be by the instruction of at least one computing device, at least one processor for the memory storage By performing the instruction of the memory storage, performed using the communication interface as any one of claim 1-5 Method.
- A kind of 12. computer-readable recording medium, it is characterised in that:The computer-readable recording medium storage has computer instruction, when the computer instruction is run on computers, So that computer performs the method as any one of claim 1-5.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711022308.6A CN107679584A (en) | 2017-10-27 | 2017-10-27 | Data classification method and system based on integrated resource allocation network RAN |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711022308.6A CN107679584A (en) | 2017-10-27 | 2017-10-27 | Data classification method and system based on integrated resource allocation network RAN |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107679584A true CN107679584A (en) | 2018-02-09 |
Family
ID=61142688
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201711022308.6A Pending CN107679584A (en) | 2017-10-27 | 2017-10-27 | Data classification method and system based on integrated resource allocation network RAN |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107679584A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112633360A (en) * | 2020-12-18 | 2021-04-09 | 中国地质大学(武汉) | Classification method based on cerebral cortex learning mode |
-
2017
- 2017-10-27 CN CN201711022308.6A patent/CN107679584A/en active Pending
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112633360A (en) * | 2020-12-18 | 2021-04-09 | 中国地质大学(武汉) | Classification method based on cerebral cortex learning mode |
CN112633360B (en) * | 2020-12-18 | 2024-04-05 | 中国地质大学(武汉) | Classification method based on cerebral cortex learning mode |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106874688B (en) | Intelligent lead compound based on convolutional neural networks finds method | |
Hu et al. | Deep learning for use in lumber classification tasks | |
CN110298663B (en) | Fraud transaction detection method based on sequence wide and deep learning | |
CN107103332A (en) | A kind of Method Using Relevance Vector Machine sorting technique towards large-scale dataset | |
CN103810101B (en) | A kind of Software Defects Predict Methods and software defect forecasting system | |
CN103853786B (en) | The optimization method and system of database parameter | |
CN108304842A (en) | Meter reading recognition methods, device and electronic equipment | |
CN110210625A (en) | Modeling method, device, computer equipment and storage medium based on transfer learning | |
CN103177265B (en) | High-definition image classification method based on kernel function Yu sparse coding | |
CN105988889A (en) | Data check method and apparatus | |
CN113610069B (en) | Knowledge distillation-based target detection model training method | |
CN106779050A (en) | The optimization method and device of a kind of convolutional neural networks | |
CN106681305A (en) | Online fault diagnosing method for Fast RVM (relevance vector machine) sewage treatment | |
CN107480702A (en) | Towards the feature selecting and Feature fusion of the identification of HCC pathological images | |
CN111882426B (en) | Method, device, equipment and storage medium for training business risk classifier | |
CN109615015A (en) | A kind of data preprocessing method based on block chain intelligence contract and machine learning | |
CN107545038A (en) | A kind of file classification method and equipment | |
CN109635104A (en) | Data classification identification method, device, computer equipment and readable storage medium storing program for executing | |
CN111126627B (en) | Model training system based on separation index | |
CN116366313A (en) | Small sample abnormal flow detection method and system | |
CN109325530A (en) | Compression method based on the depth convolutional neural networks on a small quantity without label data | |
CN114021425B (en) | Power system operation data modeling and feature selection method and device, electronic equipment and storage medium | |
Ramik et al. | Fuzzy mathematical programming: a unified approach based on fuzzy relations | |
CN113724061A (en) | Consumer financial product credit scoring method and device based on customer grouping | |
CN107292320A (en) | System and its index optimization method and device |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180209 |