CN110163252B - Data classification method and device, electronic equipment and storage medium - Google Patents
Data classification method and device, electronic equipment and storage medium Download PDFInfo
- Publication number
- CN110163252B CN110163252B CN201910309546.8A CN201910309546A CN110163252B CN 110163252 B CN110163252 B CN 110163252B CN 201910309546 A CN201910309546 A CN 201910309546A CN 110163252 B CN110163252 B CN 110163252B
- Authority
- CN
- China
- Prior art keywords
- meta
- data
- prediction model
- data classification
- feature
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 44
- 238000003860 storage Methods 0.000 title claims description 10
- 238000012549 training Methods 0.000 claims abstract description 190
- 239000011159 matrix material Substances 0.000 claims description 35
- 230000006870 function Effects 0.000 claims description 33
- 238000012545 processing Methods 0.000 claims description 14
- 238000004364 calculation method Methods 0.000 claims description 11
- 230000000694 effects Effects 0.000 claims description 11
- 238000012417 linear regression Methods 0.000 claims description 7
- 238000010276 construction Methods 0.000 claims description 6
- 238000004590 computer program Methods 0.000 claims description 4
- 238000013145 classification model Methods 0.000 claims 2
- 238000010801 machine learning Methods 0.000 abstract description 8
- 238000010586 diagram Methods 0.000 description 7
- 238000012360 testing method Methods 0.000 description 3
- 238000004422 calculation algorithm Methods 0.000 description 2
- 230000001419 dependent effect Effects 0.000 description 2
- 229920002430 Fibre-reinforced plastic Polymers 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000002131 composite material Substances 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 239000011151 fibre-reinforced plastic Substances 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000010076 replication Effects 0.000 description 1
- 230000001052 transient effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The application discloses a data classification method and device, and relates to the technical field of machine learning. The method comprises the following steps: generating at least two base learners for predicting data class labels based on the provided training set, and combining the base learners to form a data class prediction model; training the data classification prediction model according to the training set to obtain prediction parameters associated with the data classification prediction model; and predicting the class label of the data sample through the data classification prediction model according to the obtained prediction parameters, and obtaining the class label of the data sample. The method provided by the application can accurately predict the class label of the data sample.
Description
Technical Field
The present application relates to the field of machine learning technologies, and in particular, to a data classification method and apparatus, an electronic device, and a computer readable storage medium.
Background
Machine learning provides important technical support for data classification in its course. In performing supervised machine learning, the goal is to learn a stable and better performing model in every respect, but in practice only a plurality of better performing models in some respects are available. In order to solve the technical problem, the prior art proposes an integrated learning method of combining a plurality of machine learning models into one prediction model, and even if one machine learning model obtains a wrong prediction, other machine learning models can correct the error back.
However, since the prediction data may be constantly changing and learning effects of different machine learning models are different, both factors affect the prediction accuracy of the prediction model on the prediction data, so that the prediction data cannot be accurately classified.
Disclosure of Invention
Based on the technical problems, the application provides a data classification method and device, electronic equipment and a computer readable storage medium.
The technical scheme disclosed by the application comprises the following steps:
a method of data classification, the method comprising: generating at least two base learners for predicting data category labels based on the provided training set, and combining the at least two base learners to form a data category prediction model, wherein each base learner is a combination obtained by training for different historical data samples; training the data classification prediction model according to the training set to obtain prediction parameters associated with the data classification prediction model, wherein the prediction parameters comprise meta-feature learning weights and meta-feature learning parameters of the data prediction model relative to the training set; and predicting the class label of the data sample through the data classification prediction model according to the obtained prediction parameters, and obtaining the class label of the data sample.
Further, the training the data classification prediction model according to the training set to obtain a prediction parameter associated with the data classification prediction model includes: predicting the class labels of all training samples in the training set through the data classification prediction model to obtain a meta-feature matrix associated with the data classification prediction model; according to the obtained meta-feature matrix, computing the meta-feature learning weight of the data classification prediction model relative to the training set, and obtaining the meta-feature learning parameters of the data classification prediction model relative to the training set.
Further, the predicting, by the data classification prediction model, each training sample class label in the training set, to obtain a meta-feature matrix associated with the data classification prediction model, includes: constructing a residual space of the training set according to the prediction bias of the data classification prediction model on each training sample class label; and carrying out soft clustering processing on the constructed residual space according to the prediction effect of each training sample class label to obtain a meta-feature matrix associated with the data classification prediction model.
Further, the calculating the meta-feature learning weight of the data classification prediction model relative to the training set according to the obtained meta-feature matrix includes: acquiring a prediction function of the data classification prediction model, wherein the meta-feature of a training sample contained in the prediction function is derived from the meta-feature matrix; and obtaining the meta-feature learning weight of the data classification prediction model relative to the training set by least square of the prediction function and the objective function of each training sample.
Further, the obtaining, according to the obtained meta-feature matrix, the meta-feature learning parameter of the data classification prediction model relative to the training set includes: copying the training set according to the appointed group number to obtain a training set of the appointed group number; and in each group of training sets, replacing class labels obtained by predicting each training sample with corresponding meta-features in the meta-feature matrix, and taking the obtained data set with the designated group number as the meta-feature learning parameter.
Further, the predicting the class label of the data sample according to the obtained prediction parameter by using the data classification prediction model to obtain the class label of the data sample includes: performing linear regression calculation on the original characteristic data of the data sample and the meta-characteristic learning parameters to obtain meta-characteristics of the data sample relative to the data classification prediction model; calculating the meta-feature weight of the data classification prediction model relative to the data sample through the meta-feature of the data sample and the meta-feature learning weight; and predicting the class labels of the data samples through the data classification prediction model according to the obtained meta-feature weights.
Further, calculating a formula of the data classification prediction model relative to the meta feature weights of the data samples by the meta features of the data samples and the meta feature learning weights is expressed as The data classification prediction model predicts the data sample class label according to the meta-feature weight, and the calculation formula is expressed as +.>Wherein "x * "represents the data sample," T "represents the group number of the meta-feature learning parameters," m j (x * ) "represents the meta-characteristics of the data sample," v ij "represents the meta-feature weight of the data sample," Q "represents the number of category labels preset in the data classification prediction model, +.>Representing the predicted outcome of each of the base learners for the data samples, "S" represents the number of base learners.
A data classification apparatus, the apparatus comprising: the prediction model construction module is used for generating at least two base learners for predicting data category labels based on the provided training set, and combining the at least two base learners to form a data classification prediction model, wherein each base learner is a combination obtained by training different historical data samples; the prediction model training module is used for training the data classification prediction model according to the training set to obtain prediction parameters related to the data classification prediction model, wherein the prediction parameters comprise meta-feature learning weights and meta-feature learning parameters of the data prediction model relative to the training set; and the data sample prediction module is used for predicting the class label of the data sample through the data classification prediction model according to the obtained prediction parameters to obtain the class label of the data sample.
An electronic device, the electronic device comprising:
a processor;
a memory having stored thereon computer readable instructions which, when executed by the processor, implement a data classification method as described above.
A computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements a data classification method as described hereinbefore.
The technical scheme provided by the embodiment of the application can comprise the following beneficial effects:
in the above technical solution, the result of performing the class label prediction on the data sample is related to the prediction parameter of the data classification prediction model, and the prediction parameter of the data classification prediction model is dependent on the training process of the data classification prediction model through a plurality of test samples in the test set. In the application, the prediction parameters of the data classification prediction model comprise the meta-feature learning weight and the meta-feature learning parameter of the data classification prediction model relative to the training set, and the prediction parameters can well reflect the prediction weights of the prediction precision of the prediction capacities of different base learners in the data classification prediction model, so that the data classification prediction model can accurately predict the class label prediction of the data sample by combining the prediction weights of the different base learners.
It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the application as claimed.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the application and together with the description, serve to explain the principles of the application.
FIG. 1 is a schematic diagram illustrating an environment in which the present application may be implemented, according to an exemplary embodiment;
FIG. 2 is a hardware block diagram of a server shown according to an example embodiment;
FIG. 3 is a flow chart illustrating a method of data classification according to an exemplary embodiment;
FIG. 4 is a flow chart depicting step 330, shown in accordance with the corresponding embodiment of FIG. 3;
figures 5 to 8 are schematic diagrams of the data processing procedure involved in the embodiment shown in figure 4;
FIG. 9 is a flow chart depicting step 350, in accordance with the corresponding embodiment of FIG. 3;
fig. 10 is a block diagram illustrating a data sorting apparatus according to an exemplary embodiment.
There has been shown in the drawings, and will hereinafter be described, specific embodiments of the application with the understanding that the present disclosure is to be considered in all respects as illustrative, and not restrictive, the scope of the inventive concepts being indicated by the appended claims.
Detailed Description
Reference will now be made in detail to exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, the same numbers in different drawings refer to the same or similar elements, unless otherwise indicated. The implementations described in the following exemplary examples do not represent all implementations consistent with the application. Rather, they are merely examples of apparatus and methods consistent with aspects of the application as detailed in the accompanying claims.
FIG. 1 is a schematic diagram illustrating an implementation environment according to an example embodiment. As shown in fig. 1, the implementation environment of the present application includes: a terminal 100 and a server 200.
In the present application, the terminal 100 is configured to run a data classification client, which may also be referred to as a front end, for obtaining a data sample and displaying a prediction result of a data sample class label. The terminal 100 may be, but is not limited to, a smart phone, a tablet computer, a notebook computer, a computer, or any other electronic device capable of running a data sorting client.
The server 200 is configured to store the mass data, so as to respond to a service request initiated by the terminal, and perform data processing according to the service request. The server 200 may be a server or a server cluster composed of several servers, which is not limited herein.
It should be noted that, a wired or wireless network connection is pre-established between the terminal 100 and the server 200, so that the terminal 100 can perform data interaction with the server 200.
Fig. 2 is a block diagram illustrating a hardware architecture of the server 200 shown in fig. 1, according to an exemplary embodiment. It should be noted that this server is only an example adapted to the present application, and should not be construed as providing any limitation on the scope of use of the present application.
The hardware structure of the server may vary widely depending on the configuration or performance, as shown in fig. 2, the server 200 includes: a power supply 210, an interface 230, at least one memory 250, and at least one central processing unit 270.
Wherein, the power supply 210 is used for providing an operating voltage for each hardware device on the server 200.
The interface 230 includes at least one wired or wireless network interface 231, at least one serial-to-parallel interface 233, at least one input-output interface 235, and at least one USB interface 237, etc., for communicating with external devices.
The memory 250 may be a read-only memory, a random access memory, a magnetic disk, an optical disk, or the like as a carrier for storing resources, where the resources stored include an operating system 251, an application 253, data 255, or the like, and the storage manner may be transient storage or permanent storage. The operating system 251 is used for managing and controlling various hardware devices and application programs 253 on the server 200, so as to implement calculation and processing of the mass data 255 by the central processor 270. The application 253 is a computer program that performs at least one specific task based on the operating system 251, and may include at least one module (not shown in fig. 2), each of which may respectively include a series of computer readable instructions for the server 200. The data 255 may be key information stored on a disk, etc.
The central processor 270 may include one or more of the above processors and is configured to communicate with the memory 250 via a bus for computing and processing the mass data 255 in the memory 250.
As described in detail above, a server embodying the present application will perform the data sorting method by the cpu 270 reading a series of computer readable instructions stored in memory.
Fig. 3 is a flowchart illustrating a data classification method according to an exemplary embodiment, which is applicable to the server 200 shown in fig. 1. As shown in fig. 3, the method may include the steps of:
at step 310, at least two base learners for predicting data class labels are generated based on the provided training set, and the at least two base learners are combined to form a data class prediction model.
Wherein a training set is a set of several training samples, which should be understood as data samples for training a data classification predictive model. In an exemplary embodiment, the training samples may be text data in particular.
The base learner is a predictive model for predicting class labels of data samples, and can be understood as a class label prediction rule function. The different base learners can be specially trained for different historical data samples, so that the prediction effects of the different base learners are different. The historical data sample is understood to be a data sample for training a base learner, which is a common technical means in integrated learning, and is not described herein.
Based on the training set provided, it may be assumed that S base learners are generated<g 1 (x)，g 2 (x)，…，g S (x)>And the base learners are combined to form a data classification prediction model. That is, the data classification prediction model is formed by stacking at least two base learners having different prediction effects, and the data classification prediction model should be understood as an integrated learning model.
And step 330, training the data classification prediction model according to the training set to obtain prediction parameters associated with the data classification prediction model.
The process of training the data classification prediction model according to the training set should be understood that each base learner in the data classification prediction model needs to be used for predicting the class label of each training sample in the training set to obtain a corresponding prediction result, and the obtained prediction results are integrated according to a certain rule, so that the class label of each training sample is finally obtained.
For example, for training sample x, the class label prediction is performed by using the S base learners generated as described above, so as to obtainObtaining the result obtained by each base learner through predicting the class label of the training sample x<y 1 ，y 2 ，…，y S >It is then necessary to finally obtain the class label of the training sample x by integrating this prediction.
The prediction parameters associated with the data classification prediction model are relevant parameters obtained by training the data classification prediction model for the training set. In an exemplary embodiment, the prediction parameters may include a meta-feature learning weight and a meta-feature learning parameter, wherein the meta-feature learning weight represents: each base learner in the data classification prediction model respectively predicts the result of the training sample, and relative to the importance weight of the predicted result finally obtained by the training sample, the meta-feature learning parameter represents: learning parameters for extracting meta-features of data samples, which are input features for describing the data samples in a data classification predictive model.
Alternatively, in other exemplary embodiments, the prediction parameters may also include other parameters, and are not meant to be limiting of the application to the particular type of prediction parameters.
It should be noted that, the prediction parameters related to the data classification prediction model are obtained by performing corresponding processing on the result obtained by training after the data classification prediction model is trained on the training samples provided by the training set.
The prediction parameters include meta-feature learning weights and meta-feature learning parameters of the data prediction model relative to the training set, as shown in fig. 4, in an exemplary embodiment, obtaining the prediction parameters associated with the data classification prediction model may include the steps of:
and 331, predicting the class labels of each training sample in the training set through the data classification prediction model to obtain a meta-feature matrix associated with the data classification prediction model.
And predicting class labels of all training samples through each base learner in the data classification prediction model to obtain corresponding prediction results, and then combining the prediction results to form a meta-feature matrix associated with the data classification prediction model. The meta-feature matrix is used to describe the input features of each training sample in the training set in the data classification predictive model.
Specifically, according to the prediction deviation of the base learner in the data classification prediction model to each training sample class label, constructing the residual space of the training set, and then according to the prediction effect of the data classification prediction model to each training sample class label, performing soft clustering processing on the constructed residual space to obtain a meta-feature matrix associated with the data classification prediction model.
The prediction bias of each training sample on the different base learners should be understood as the difference between the actual class label of the training sample and the class label predicted by the different base learners, respectively. If the obtained difference value is larger, the prediction effect of the corresponding base learner on the training sample is poorer; if the obtained difference is smaller, the corresponding base learner has better prediction effect on the training sample.
In an exemplary embodiment, as shown in FIG. 5, assume that the residuals (i.e., prediction bias) of training samples x on different base learners are<r 1 ，r 2 ，…，r S >Then, for the training set D containing N training samples x, the generated residual space is shown in fig. 6, where the residual space represents how good the data classification prediction model performs the prediction effect of the class label on each training sample in the training set.
After constructing the residual space of the training set, soft clustering can be performed on the residual space by a fuzzy C-means method so as to gather different training samples in the training set together according to the prediction effect of some specific base learners on different training samples to form a plurality of clusters. That is, training sets are divided into several classes of clusters based on similarity between different training samples.
If the number of the classified clusters is assumed to be T, the membership degree between each training sample and each cluster can be obtained through the fuzzy C-means method, and the membership degree is the meta-feature associated with each training sample and the data classification prediction model, and can be specifically represented through a meta-feature function, and each training is mapped through the meta-feature functionTraining samples to original features of the training samples. For example, the membership between training sample x and different classes of clusters can be expressed as a meta-feature function: f (f) 1 (x)，f 2 (x)，f 3 (x)，…，f T (x) A. The application relates to a method for producing a fibre-reinforced plastic composite The original features of the training samples should be understood as the data content contained in the training samples, for example, when the training set is a text set, the original features of each training sample are corresponding to specific text content.
For the whole training set, as shown in fig. 7, the meta-feature matrix of the training set associated with the data classification prediction model is obtained by combining meta-features of each training sample with the data classification prediction model. Thus, each training sample contains T meta-features f associated with the data classification predictive model j (x)。
Step 333, calculating the meta-feature learning weight of the data classification prediction model relative to the training set according to the obtained meta-feature matrix, and obtaining the meta-feature learning parameter of the data classification prediction model relative to the training set.
As previously described, the data classification predictive model learns weight representation relative to the meta-features of the training set: and each base learner in the data classification prediction model respectively calculates the importance weight of the prediction result of each training sample in the training set and the prediction result finally obtained by the training sample. It can be derived from the meta-feature learning weights that, of the at least two base learners integrated by the data classification prediction model, which base learners have higher accuracy in the prediction of the training samples, then the weights of these base learners are also higher.
The meta-feature learning parameter indicates a learning parameter for extracting meta-features of the data sample, so that the meta-feature learning parameter of the data classification prediction model needs to be obtained in advance, and the meta-feature of the data sample can be obtained according to the meta-feature learning parameter when the data classification prediction model performs class label prediction on the data sample.
In an exemplary embodiment, calculating the meta-feature learning weights of the data classification prediction model may include the steps of:
acquiring a prediction function of the data classification prediction model;
and obtaining the meta-feature learning weight of the data classification prediction model relative to the training set by least square of the prediction function and the objective function of each training sample.
Wherein, after obtaining the meta-feature matrix of the data classification prediction model, the FWLS (Feature Weighted Linear Stacking, feature weighted linear superposition) algorithm may be used to calculate the meta-feature learning weights. Firstly, a prediction function of a data classification prediction model is required to be obtained, wherein the prediction function is a combination function searched based on a standard linear regression stacking process, and the prediction function is specifically as follows:
wherein b (x) represents a prediction function of the data classification prediction model, v ij Meta-feature learning weights, f, representing relative training sets of data classification predictive models j (x) Representing meta-features of training sample x, g i (x) A base learner that predicts class labels for training samples x is shown.
As described above, the meta-feature matrix associated with the data classification prediction model is obtained by combining the meta-features associated with the data classification prediction model with each training sample, so that the meta-features of the training sample x are derived from the meta-feature matrix and correspond to some elements in the meta-feature matrix.
Then, by adopting an optimization method of the FLWS algorithm, the meta-feature learning weight v is performed by performing least square calculation on the prediction function of the data classification prediction model and the objective function of each training sample ij Is calculated, and the calculated meta-feature learning weight v ij Should be the minimum calculated according to the following expression. The expression is as follows:
wherein y (x) represents the target function of training sample xA number. Meta-feature learning weight v ij Information representing the prediction information, confidence level, etc. of the ith base learner on the training sample data, and for a trained data classification prediction model, the meta-feature learning weight v ij Is a constant and is therefore defined by a meta-characteristic function f j (x) The weight of the ith basis learner on the training sample is determined.
It should be appreciated that the above-described computation of the meta-feature learning weights v ij The essence of the minimum process is the process of optimizing each base learner parameter in the data classification prediction model, when the meta-feature learning weight v is obtained ij At the minimum value, the data classification prediction model is trained, and the meta-feature of the data classification prediction model learns the weight v ij Then it is a constant.
In an exemplary embodiment, acquiring the meta-feature learning parameters of the data classification prediction model may include the steps of:
copying the training set according to the appointed group number to obtain the training set of the appointed group number;
and in each group of training sets, replacing the category labels predicted by each training sample with corresponding meta-features in the meta-feature matrix, and taking the obtained data set with the designated group number as a meta-feature learning parameter.
The number of the designated groups is preset in the data classification prediction model and is consistent with the number of the classified clusters (namely T groups).
By copying the training sets into T groups, the class labels predicted by each training sample in each training set are respectively replaced with the corresponding meta-feature functions in the meta-feature matrix, as shown in fig. 8, so that T data sets can be obtained, and the obtained T data sets are meta-feature learning parameters of the data classification prediction model relative to the training sets.
And step 350, according to the obtained prediction parameters, performing class label prediction of the data sample through the data classification prediction model to obtain class labels of the data sample.
When the trained data classification prediction model is used for predicting the class label of the data sample, the prediction parameters obtained by training the data classification prediction model are needed to be used for carrying out.
As shown in fig. 9, in an exemplary embodiment, the classification tag prediction of the data sample by the data classification prediction model may include the steps of:
and 351, linearly regressing the original features and the meta-feature learning parameters of the data sample to obtain the meta-features of the data sample relative to the data classification prediction model.
As previously described, the original characteristics of a data sample represent the data content, such as text data, contained by the data sample. Let the data sample be x * The data sample is x * Performing linear regression (Linear Regression, LR) on the T data sets obtained by training to obtain a test sample x * Meta-features of (2), which can be expressed as m j (x * )。
The meta features of the data samples are obtained on the basis of the parameters obtained by learning in the category label prediction process of the training set by the plurality of base learners, and the input features of the data samples in the data classification prediction model can be more accurately represented, so that the basis is provided for accurate prediction of the category labels of the data samples by the data classification prediction model.
In step 353, the meta-feature weights of the data classification prediction model relative to the data samples are calculated by the meta-features and the meta-feature learning weights of the data samples.
The method comprises the steps of combining meta-feature learning weights acquired by a data classification prediction model in a training stage, calculating importance weights corresponding to all base learners in the process of predicting data sample class labels through the data classification prediction model, and obtaining a result, namely the meta-feature weights of the data classification prediction model relative to the data samples, wherein the calculation formula is as follows:
and 355, predicting the class label of the data sample through the data classification prediction model according to the obtained meta-feature weight.
And after the meta-feature weight of the data classification prediction model relative to the data sample is obtained, the meta-feature weight and the meta-feature of the data sample are brought into a prediction function of the data classification prediction model to be calculated, so that the class label prediction of the data sample is performed through the data classification prediction model.
Specifically, in an exemplary embodiment, for data sample x * The data classification prediction model needs to be preset from the preset class label set { l } 1 ，l 2 ，l 3 ，…，l Q A class label is predicted in the }, and the calculation formula is as follows:
wherein, the classification label needs to be preset as a Q-dimensional vector Wherein (1)>Representation base learner g i In category label l Q And an output from the first and second switches.
It can be seen that, in the data classification prediction model, the meta-feature weight of the data classification prediction model is related to the meta-feature of the data sample, so that it can be understood that in the data classification prediction method provided by the application, the importance weight of each base learner in the data classification prediction model is dependent on the meta-feature of the data sample, and the meta-feature of the data sample further comprises that each base learner learns in the process of performing prediction accuracy training, so that the data classification prediction model provided by the application can accurately predict the class label of the data sample.
Fig. 10 is a diagram illustrating a data sorting apparatus according to an exemplary embodiment. As shown in fig. 10, the apparatus includes a predictive model construction module 410, a predictive model training module 430, and a data sample prediction module 450.
The prediction model construction module 410 is configured to generate at least two base learners for predicting data class labels based on the provided training set, and combine at least two base learners to form a data class prediction model, where each base learner is respectively trained for different historical data samples.
The prediction model training module 430 is configured to train the data classification prediction model according to the training set, and obtain prediction parameters associated with the data classification prediction model, where the prediction parameters include meta-feature learning weights and meta-feature learning parameters of the data prediction model relative to the training set.
The data sample prediction module 450 is configured to predict a class label of a data sample according to the obtained prediction parameter through the data classification prediction model, so as to obtain the class label of the data sample.
In another embodiment, the prediction model training module 430 specifically includes a training sample prediction unit and a prediction parameter acquisition unit.
And the training sample prediction unit is used for predicting each training sample class label in the training set through the data classification prediction model to obtain a meta-feature matrix associated with the data classification prediction model.
The prediction parameter acquisition unit is used for calculating the meta-feature learning weight of the data classification prediction model relative to the training set according to the obtained meta-feature matrix, and acquiring the meta-feature learning parameter of the data classification prediction model relative to the training set.
In another embodiment, the training sample prediction unit specifically comprises a residual space construction subunit and a residual space processing subunit.
And the residual space construction subunit is used for constructing the residual space of the training set according to the prediction deviation of the data classification prediction model on each training sample class label.
And the residual space processing subunit is used for carrying out soft clustering processing on the constructed residual space according to the prediction effect of each training sample class label to obtain a meta-feature matrix associated with the data classification prediction model.
In another embodiment, the prediction parameter acquisition unit specifically includes a function acquisition subunit and a function processing subunit.
The function obtaining subunit is used for obtaining a prediction function of the data classification prediction model, and the meta-feature of the training sample contained in the prediction function is derived from the meta-feature matrix.
The function processing subunit is used for obtaining the meta-feature learning weight of the data classification prediction model relative to the training set by least square of the prediction function and the objective function of each training sample.
In another embodiment, the prediction parameter acquisition unit specifically includes a data replication subunit and a data replacement subunit.
The data copying subunit is used for copying the training set according to the appointed group number to obtain the training set of the appointed group number;
and the data replacing subunit is used for replacing the category labels predicted by the training samples with corresponding meta-features in the meta-feature matrix in each group of training sets, and taking the obtained data set with the designated group number as the meta-feature learning parameter.
In another embodiment, the data sample prediction module 450 specifically includes a meta-feature acquisition unit, a meta-feature weight calculation unit, and a category label prediction unit.
And the meta-feature acquisition unit is used for carrying out linear regression calculation on the original feature data of the data sample and the meta-feature learning parameter to obtain the meta-feature of the data sample relative to the data classification prediction model.
And the meta-feature weight calculation unit is used for calculating the meta-feature weight of the data classification prediction model relative to the data sample through the meta-feature of the data sample and the meta-feature learning weight.
And the class label prediction unit is used for predicting the class label of the data sample through the data classification prediction model according to the obtained meta-feature weight.
It should be noted that, the apparatus provided in the foregoing embodiments and the method provided in the foregoing embodiments belong to the same concept, and a specific manner in which each module performs an operation has been described in detail in the method embodiment, which is not described herein again.
In an exemplary embodiment, the present application also provides an electronic device including:
a processor;
a memory having stored thereon computer readable instructions which, when executed by a processor, implement a data classification method as previously described.
In an exemplary embodiment, the application also provides a computer-readable storage medium, on which a computer program is stored which, when executed by a processor, implements a data classification method as previously indicated.
It is to be understood that the application is not limited to the precise arrangements and instrumentalities shown in the drawings, which have been described above, and that various modifications and changes may be effected without departing from the scope thereof. The scope of the application is limited only by the appended claims.
Claims (8)
1. A method of classifying data, the method comprising:
generating at least two base learners for predicting data category labels based on the provided training set, and combining the at least two base learners to form a data category prediction model, wherein each base learner is respectively trained for different historical data samples;
training the data classification prediction model according to the training set to obtain prediction parameters related to the data classification prediction model, wherein the prediction parameters comprise meta-feature learning weights and meta-feature learning parameters of the data prediction model relative to the training set, the meta-feature learning weights represent importance weights of prediction results of training samples in the data classification model by each base learner relative to prediction results finally obtained by the training samples, and the meta-feature learning parameters represent learning parameters for extracting meta-features of the data samples;
according to the obtained prediction parameters, performing class label prediction of the data sample through the data classification prediction model to obtain class labels of the data sample;
training the data classification prediction model according to the training set to obtain prediction parameters associated with the data classification prediction model, wherein the training set comprises the following steps:
predicting the class labels of all training samples in the training set through the data classification prediction model to obtain a meta-feature matrix associated with the data classification prediction model;
according to the obtained meta-feature matrix, computing meta-feature learning weight of the data classification prediction model relative to the training set, and obtaining meta-feature learning parameters of the data classification prediction model relative to the training set;
and predicting the class label of the data sample through the data classification prediction model according to the obtained prediction parameters to obtain the class label of the data sample, wherein the method comprises the following steps:
performing linear regression calculation on the original characteristic data of the data sample and the meta-characteristic learning parameters to obtain meta-characteristics of the data sample relative to the data classification prediction model;
calculating the meta-feature weight of the data classification prediction model relative to the data sample through the meta-feature of the data sample and the meta-feature learning weight;
and predicting the class labels of the data samples through the data classification prediction model according to the obtained meta-feature weights.
2. The method of claim 1, wherein said predicting, by said data classification prediction model, each training sample class label in said training set to obtain a meta-feature matrix associated with said data classification prediction model, comprises:
constructing a residual space of the training set according to the prediction bias of the data classification prediction model on each training sample class label;
and carrying out soft clustering processing on the constructed residual space according to the prediction effect of each training sample class label to obtain a meta-feature matrix associated with the data classification prediction model.
3. The method of claim 1, wherein calculating the meta-feature learning weights of the data classification prediction model relative to the training set based on the obtained meta-feature matrix comprises:
acquiring a prediction function of the data classification prediction model, wherein the meta-feature of a training sample contained in the prediction function is derived from the meta-feature matrix;
and obtaining the meta-feature learning weight of the data classification prediction model relative to the training set by least square of the prediction function and the objective function of each training sample.
4. The method according to claim 1, wherein the obtaining, according to the obtained meta-feature matrix, meta-feature learning parameters of the data classification prediction model relative to the training set includes:
copying the training set according to the appointed group number to obtain a training set of the appointed group number;
and in each group of training sets, replacing class labels obtained by predicting each training sample with corresponding meta-features in the meta-feature matrix, and taking the obtained data set with the designated group number as the meta-feature learning parameter.
5. The method of claim 1, wherein calculating a formula for the data classification prediction model relative to the meta-feature weights of the data samples by the meta-features of the data samples and the meta-feature learning weights is expressed as
The data classification prediction model predicts the data sample class label according to the meta-feature weight as follows
Wherein "x * "means the data sample," Q * "representing class labels of the data samples," T "representing the number of groups of the meta-feature learning parameters," i "representing each base learner," j "representing each meta-feature," m j (x * ) "represents the meta-characteristics of the data sample," v ij "represents the meta-feature weight of the data sample," Q "represents the number of category labels preset in the data classification prediction model,representing the predicted outcome of each of the base learners for the data samples, "S" represents the number of base learners.
6. A data sorting apparatus, the apparatus comprising:
the prediction model construction module is used for generating at least two base learners for predicting data category labels based on the provided training set, and combining the at least two base learners to form a data classification prediction model, wherein each base learner is trained for different historical data samples;
the prediction model training module is used for training the data classification prediction model according to the training set to obtain prediction parameters related to the data classification prediction model, wherein the prediction parameters comprise meta-feature learning weights and meta-feature learning parameters of the data prediction model relative to the training set, the meta-feature learning weights represent importance weights of prediction results of each base learner in the data classification model on training samples relative to prediction results finally obtained by the training samples, and the meta-feature learning parameters represent learning parameters for extracting meta-features of the data samples;
the data sample prediction module is used for predicting the class label of the data sample through the data classification prediction model according to the obtained prediction parameters to obtain the class label of the data sample;
wherein, the predictive model training module is further configured to:
predicting the class labels of all training samples in the training set through the data classification prediction model to obtain a meta-feature matrix associated with the data classification prediction model;
according to the obtained meta-feature matrix, computing meta-feature learning weight of the data classification prediction model relative to the training set, and obtaining meta-feature learning parameters of the data classification prediction model relative to the training set;
the data sample prediction module is further configured to:
performing linear regression calculation on the original characteristic data of the data sample and the meta-characteristic learning parameters to obtain meta-characteristics of the data sample relative to the data classification prediction model;
calculating the meta-feature weight of the data classification prediction model relative to the data sample through the meta-feature of the data sample and the meta-feature learning weight;
and predicting the class labels of the data samples through the data classification prediction model according to the obtained meta-feature weights.
7. An electronic device, the device comprising:
a processor;
a memory having stored thereon computer readable instructions which, when executed by the processor, implement the data classification method of any of claims 1 to 5.
8. A computer-readable storage medium, on which a computer program is stored which, when executed by a processor, implements the data classification method according to any one of claims 1 to 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910309546.8A CN110163252B (en) | 2019-04-17 | 2019-04-17 | Data classification method and device, electronic equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910309546.8A CN110163252B (en) | 2019-04-17 | 2019-04-17 | Data classification method and device, electronic equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110163252A CN110163252A (en) | 2019-08-23 |
| CN110163252B true CN110163252B (en) | 2023-11-24 |
Family
ID=67638590
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910309546.8A Active CN110163252B (en) | 2019-04-17 | 2019-04-17 | Data classification method and device, electronic equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110163252B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11481623B2 (en) * | 2019-09-25 | 2022-10-25 | International Business Machines Corporation | Systems and methods for training a model using a few-shot classification process |
| CN112825576B (en) * | 2019-11-20 | 2023-05-05 | 中国电信股份有限公司 | Cell capacity expansion determining method, device and storage medium |
| CN110929785B (en) * | 2019-11-21 | 2023-12-05 | 中国科学院深圳先进技术研究院 | Data classification method, device, terminal equipment and readable storage medium |
| CN111126448A (en) * | 2019-11-29 | 2020-05-08 | 无线生活(北京)信息技术有限公司 | Method and device for intelligently identifying fraud users |
| WO2022141094A1 (en) * | 2020-12-29 | 2022-07-07 | 深圳市大疆创新科技有限公司 | Model generation method and apparatus, image processing method and apparatus, and readable storage medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6453307B1 (en) * | 1998-03-03 | 2002-09-17 | At&T Corp. | Method and apparatus for multi-class, multi-label information categorization |
| CN104881689A (en) * | 2015-06-17 | 2015-09-02 | 苏州大学张家港工业技术研究院 | Method and system for multi-label active learning classification |
| CN105354595A (en) * | 2015-10-30 | 2016-02-24 | 苏州大学 | Robust visual image classification method and system |
| CN106446951A (en) * | 2016-09-28 | 2017-02-22 | 中科院成都信息技术股份有限公司 | Singular value selection-based integrated learning device |
| CN108710907A (en) * | 2018-05-15 | 2018-10-26 | 苏州大学 | Handwritten form data classification method, model training method, device, equipment and medium |
| CN109471938A (en) * | 2018-10-11 | 2019-03-15 | 平安科技（深圳）有限公司 | A kind of file classification method and terminal |
-
2019
- 2019-04-17 CN CN201910309546.8A patent/CN110163252B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6453307B1 (en) * | 1998-03-03 | 2002-09-17 | At&T Corp. | Method and apparatus for multi-class, multi-label information categorization |
| CN104881689A (en) * | 2015-06-17 | 2015-09-02 | 苏州大学张家港工业技术研究院 | Method and system for multi-label active learning classification |
| CN105354595A (en) * | 2015-10-30 | 2016-02-24 | 苏州大学 | Robust visual image classification method and system |
| CN106446951A (en) * | 2016-09-28 | 2017-02-22 | 中科院成都信息技术股份有限公司 | Singular value selection-based integrated learning device |
| CN108710907A (en) * | 2018-05-15 | 2018-10-26 | 苏州大学 | Handwritten form data classification method, model training method, device, equipment and medium |
| CN109471938A (en) * | 2018-10-11 | 2019-03-15 | 平安科技（深圳）有限公司 | A kind of file classification method and terminal |
Non-Patent Citations (1)
| Title |
|---|
| SUCE:基于聚类集成的半监督二分类方法;闵帆 等;智能系统学报;第13卷(第06期);第974-980页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110163252A (en) | 2019-08-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110163252B (en) | Data classification method and device, electronic equipment and storage medium | |
| US20220382564A1 (en) | Aggregate features for machine learning | |
| US11694109B2 (en) | Data processing apparatus for accessing shared memory in processing structured data for modifying a parameter vector data structure | |
| CN110196908A (en) | Data classification method, device, computer installation and storage medium | |
| CN109885768A (en) | Worksheet method, apparatus and system | |
| CN112465138A (en) | Model distillation method, device, storage medium and equipment | |
| US20190286978A1 (en) | Using natural language processing and deep learning for mapping any schema data to a hierarchical standard data model (xdm) | |
| CN112883257B (en) | Behavior sequence data processing method and device, electronic equipment and storage medium | |
| US20210216904A1 (en) | Knowledge Aided Feature Engineering | |
| CN114298122A (en) | Data classification method, device, equipment, storage medium and computer program product | |
| CN111191797B (en) | Information processing method, information processing device, electronic equipment and storage medium | |
| CN112200296A (en) | Network model quantification method and device, storage medium and electronic equipment | |
| CN113821527A (en) | Hash code generation method and device, computer equipment and storage medium | |
| Han et al. | SlimML: Removing non-critical input data in large-scale iterative machine learning | |
| CN114912030A (en) | Equity model training method, equity model recommendation method, electronic terminal and computer medium | |
| CN115222443A (en) | Client group division method, device, equipment and storage medium | |
| CN113761375A (en) | Message recommendation method, device, equipment and storage medium based on neural network | |
| CN112269875B (en) | Text classification method, device, electronic equipment and storage medium | |
| CN114020892A (en) | Answer selection method and device based on artificial intelligence, electronic equipment and medium | |
| CN113641835A (en) | Multimedia resource recommendation method and device, electronic equipment and medium | |
| CN114281950B (en) | Data retrieval method and system based on multi-graph weighted fusion | |
| CN112433952B (en) | Method, system, device and medium for testing fairness of deep neural network model | |
| CN111897943A (en) | Session record searching method and device, electronic equipment and storage medium | |
| CN111723254A (en) | Information processing apparatus, recording medium, and information processing method | |
| US11893772B1 (en) | Artificial intelligence system with iterative two-phase active learning |
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 |