High-dimensional feature extraction method, device, computer equipment and storage medium

Abstract

The embodiments of this application belong to the field of artificial intelligence and relate to a high-dimensional feature extraction method, comprising obtaining feature data of raw high-dimensional data, constructing an observation data sample, wherein the observation data sample includes discrete variables and continuous variables; obtaining category labels, grouping the continuous variables according to the category labels to obtain grouped variables, and calculating the sorted sum of the grouped variables; calculating the sorted sum based on a preset detection algorithm to obtain evaluation parameters, transposing all evaluation parameters to obtain a correlation vector, and performing feature screening on the raw high-dimensional data based on the correlation vector to obtain marker features; obtaining a target feature screening model and continuous parameters, inputting the marker features and continuous parameters into the target feature screening model, and calculating target dimensionality reduction features. This application also provides a high-dimensional feature extraction device, a computer device, and a storage medium. In addition, the target dimensionality reduction features can be stored in a blockchain. This application achieves accurate feature extraction from high-dimensional data.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a high-dimensional feature extraction method, apparatus, computer equipment, and storage medium. Background Art

[0002] With the rapid development of biological detection, detecting biological features and extracting effective features from them is an urgent problem that needs to be solved. However, biological-related omics data are usually typical ultra-high-dimensional data, such as proteomes, transcriptomes (RNA), genomes (DNA), etc. In this era of increasingly in-depth integration of data science, artificial intelligence, and medicine, feature screening models based on ultra-high-dimensional data will become increasingly important. Currently, feature extraction from high-dimensional data through models mainly uses regularization methods. However, when the number of input features is exponentially larger than the sample size (for example, when the sample size is 100 and the number of features is 10,000), it is difficult to accurately extract features from high-dimensional data through regularization methods. Summary of the Invention

[0003] The purpose of the embodiments of the present application is to propose a high-dimensional feature extraction method, apparatus, computer equipment and storage medium to solve the technical problem of low accuracy in feature extraction of high-dimensional data.

[0004] In order to solve the above technical problems, the present invention provides a high-dimensional feature extraction method, which adopts the following technical solutions:

[0005] Acquire feature data of the original high-dimensional data, and construct observation data samples of each feature data, wherein the observation data samples include discrete variables and continuous variables;

[0006] Obtaining a category label of the discrete variable, grouping the continuous variable according to the category label to obtain a plurality of grouping variables, and calculating a ranking sum of each of the grouping variables;

[0007] Calculating all the sorting sums of each feature data based on a preset detection algorithm to obtain evaluation parameters for each feature data, transposing all the evaluation parameters to obtain a correlation vector of the original high-dimensional data, and performing feature screening on the original high-dimensional data according to the correlation vector to obtain a marker feature;

[0008] A target feature screening model and continuous parameters corresponding to each feature data are obtained, the marker features and the continuous parameters are input into the target feature screening model, and the target dimensionality reduction features of the original high-dimensional data are calculated.

[0009] Furthermore, the step of calculating all the sorting and the calculation of each feature data based on a preset detection algorithm to obtain an evaluation parameter of each feature data includes:

[0010] Obtaining a detection function of the preset detection algorithm, and calculating all the sorting sums of each feature data according to the detection function to obtain a detection statistic of the feature data;

[0011] A target distribution table is obtained, and evaluation parameters of the feature data are obtained by searching the target distribution table based on the detection statistic.

[0012] Furthermore, the step of obtaining the evaluation parameter of the feature data from the target distribution table based on the detection statistic includes:

[0013] Calculating the degree of freedom of the feature data according to the category label;

[0014] An evaluation parameter of the feature data is obtained by searching the target distribution table based on the degrees of freedom and the detection statistic.

[0015] Furthermore, the step of calculating the sorted sum of each of the grouping variables includes:

[0016] Sorting all the continuous variables of the observation data sample in ascending order according to their numerical values ​​to obtain a sorting number for each continuous variable;

[0017] The sample size in the grouping variable is obtained, and the ranking sum of the grouping variable is calculated according to the sample size and the ranking number.

[0018] Furthermore, the step of performing feature screening on the original high-dimensional data according to the correlation vector to obtain a marker feature includes:

[0019] Sorting all elements of the correlation vector in ascending order to obtain a total feature set;

[0020] A preset screening threshold is obtained, elements in the total feature set are screened according to the preset screening threshold to obtain a screening vector, feature data corresponding to the screening vector is obtained, and the marking feature is obtained.

[0021] Furthermore, the step of inputting the marker features and the continuous parameters into the target feature screening model to calculate the target dimensionality reduction features of the original high-dimensional data includes:

[0022] Obtaining a penalty function for the target feature screening model;

[0023] The penalty function is solved according to the marker feature and the continuous variable to obtain a sparse solution, and the sparse solution is determined to be the target dimensionality reduction feature.

[0024] Furthermore, the step of inputting the marker features and the continuous parameters into the target feature screening model to calculate the target dimensionality reduction features of the original high-dimensional data further includes:

[0025] The correlation index between the marker feature and the continuous parameter is calculated based on the target feature screening model, and the marker feature is screened according to the correlation index to obtain the target dimensionality reduction feature.

[0026] In order to solve the above technical problems, the present application also provides a high-dimensional feature extraction device, which adopts the following technical solution:

[0027] A construction module is used to obtain feature data of the original high-dimensional data and construct an observation data sample of each feature data, wherein the observation data sample includes discrete variables and continuous variables;

[0028] a grouping module, configured to obtain a category label of the discrete variable, group the continuous variable according to the category label to obtain a plurality of grouping variables, and calculate a ranking sum of each of the grouping variables;

[0029] An evaluation module is configured to calculate all the sorting sums of each feature data based on a preset detection algorithm to obtain evaluation parameters for each feature data, transpose all the evaluation parameters to obtain a correlation vector of the original high-dimensional data, and perform feature screening on the original high-dimensional data according to the correlation vector to obtain a marker feature;

[0030] A calculation module is used to obtain a target feature screening model and continuous parameters corresponding to each feature data, input the marker features and the continuous parameters into the target feature screening model, and calculate the target dimensionality reduction features of the original high-dimensional data.

[0031] In order to solve the above technical problems, the embodiment of the present application further provides a computer device, which adopts the following technical solution:

[0032] Acquire feature data of the original high-dimensional data, and construct observation data samples of each feature data, wherein the observation data samples include discrete variables and continuous variables;

[0033] Obtaining a category label of the discrete variable, grouping the continuous variable according to the category label to obtain a plurality of grouping variables, and calculating a ranking sum of each of the grouping variables;

[0034] Calculating all the sorting sums of each feature data based on a preset detection algorithm to obtain evaluation parameters for each feature data, transposing all the evaluation parameters to obtain a correlation vector of the original high-dimensional data, and performing feature screening on the original high-dimensional data according to the correlation vector to obtain a marker feature;

[0035] A target feature screening model and continuous parameters corresponding to each feature data are obtained, the marker features and the continuous parameters are input into the target feature screening model, and the target dimensionality reduction features of the original high-dimensional data are calculated.

[0036] In order to solve the above technical problems, the embodiment of the present application further provides a computer-readable storage medium, which adopts the following technical solution:

[0037] Acquire feature data of the original high-dimensional data, and construct observation data samples of each feature data, wherein the observation data samples include discrete variables and continuous variables;

[0038] Obtaining a category label of the discrete variable, grouping the continuous variable according to the category label to obtain a plurality of grouping variables, and calculating a ranking sum of each of the grouping variables;

[0039] Calculating all the sorting sums of each feature data based on a preset detection algorithm to obtain evaluation parameters for each feature data, transposing all the evaluation parameters to obtain a correlation vector of the original high-dimensional data, and performing feature screening on the original high-dimensional data according to the correlation vector to obtain a marker feature;

[0040] A target feature screening model and continuous parameters corresponding to each feature data are obtained, the marker features and the continuous parameters are input into the target feature screening model, and the target dimensionality reduction features of the original high-dimensional data are calculated.

[0041] The high-dimensional feature extraction method proposed in this application obtains the feature data of the original high-dimensional data and constructs an observation data sample for each feature data, wherein the observation data sample includes discrete variables and continuous variables; then, the category labels of the discrete variables are obtained, the continuous variables are grouped according to the category labels to obtain multiple grouping variables, and the ranking sum of each grouping variable is calculated. The ranking sum can be used to accurately calculate the evaluation parameters of different feature data, thereby improving the accuracy of feature screening; then, all the ranking sums of each feature data are calculated based on a preset detection algorithm to obtain the evaluation parameters of each feature data, and the importance of different feature data can be evaluated by the evaluation parameters. , all evaluation parameters are transposed to obtain the correlation vector of the original high-dimensional data, and the original high-dimensional data is feature screened according to the correlation vector to obtain the marker feature, thereby realizing the initial feature screening of the original high-dimensional data; then, the target feature screening model and the continuous parameters corresponding to each feature data are obtained, the marker feature and the continuous parameter are input into the target feature screening model, and the target dimensionality reduction feature of the original high-dimensional data is calculated, and finally multiple feature screening of the high-dimensional data is realized. In addition, the model is robust to data outliers and long-tail distributions, and can process data of continuous dependent variables and discrete features, thereby improving the accuracy and efficiency of high-dimensional data feature screening. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the solutions in this application, a brief introduction will be given below to the drawings required for use in the description of the embodiments of this application. Obviously, the drawings described below are some embodiments of this application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0043] Figure 1 is an exemplary system architecture diagram to which the present application may be applied;

[0044] Figure 2 A flowchart of an embodiment of a high-dimensional feature extraction method according to the present application;

[0045] Figure 3 is a structural diagram of an embodiment of a high-dimensional feature extraction device according to the present application;

[0046] Figure 4 It is a structural diagram of an embodiment of a computer device according to the present application.

[0047] Reference numerals: high-dimensional feature extraction device 300 , construction module 301 , grouping module 302 , evaluation module 303 , and calculation module 304 . DETAILED DESCRIPTION

[0048] Unless otherwise defined, all technical and scientific terms used herein have the same meanings as commonly understood by those skilled in the art to which this application belongs. The terms used in the specification of the application are for the purpose of describing specific embodiments only and are not intended to limit this application. The terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of this application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0049] References herein to "embodiments" mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0050] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings.

[0051] like Figure 1 As shown, system architecture 100 may include terminal devices 101, 102, 103, a network 104, and a server 105. Network 104 is a medium for providing communication links between terminal devices 101, 102, 103 and server 105. Network 104 may include various connection types, such as wired or wireless communication links or fiber optic cables.

[0052] Users can use terminal devices 101, 102, and 103 to interact with server 105 via network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, and 103, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0053] Terminal devices 101, 102, and 103 can be various electronic devices with display screens and support web browsing, including but not limited to smartphones, tablet computers, e-book readers, MP3 players (Moving Picture Experts Group Audio Layer III), MP4 (Moving Picture Experts Group Audio Layer IV), laptop computers, desktop computers, etc.

[0054] The server 105 may be a server that provides various services, such as a background server that provides support for web pages displayed on the terminal devices 101 , 102 , and 103 .

[0055] It should be noted that the high-dimensional feature extraction method provided in the embodiment of the present application is generally executed by a server / terminal device, and accordingly, the high-dimensional feature extraction device is generally set in the server / terminal device.

[0056] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is merely illustrative. Any number of terminal devices, networks and servers may be provided as required.

[0057] Continue to refer Figure 2 , shows a flow chart of an embodiment of a method for extracting high-dimensional features according to the present application. The high-dimensional feature extraction method comprises the following steps:

[0058] Step S201 : Acquire feature data of original high-dimensional data, and construct observation data samples for each feature data, wherein the observation data samples include discrete variables and continuous variables.

[0059] In this embodiment, the original high-dimensional data is the initial high-dimensional data. In the medical field, biological omics data is usually high-dimensional data, such as proteome, transcriptome (RNA) and genome (DNA). The feature data of the original high-dimensional data is obtained. The feature data is a discrete feature obtained by collecting the original high-dimensional data, and an observation data sample of each feature data is constructed. Specifically, one feature data can construct multiple observation data samples, and the number of observation data samples is determined by the category label of the original high-dimensional data. When obtaining the feature data, the category label of the original high-dimensional data and the feature value of the feature data under each category label are obtained. The feature value is a discrete variable; then, the continuous variable corresponding to the discrete variable under each category label is obtained, and the observation data sample of the feature data is constructed according to the discrete variable and the continuous variable. The observation data sample can be obtained by (X i ,Y i ), where Y is a continuous variable, X is a discrete variable, and X i ∈{1,…,K}, where K is the observed class label. Discrete variables are feature variables whose values ​​can be listed one by one in a certain order, usually taking integer values, such as the feature data under different class labels of the proteome or transcriptome. Continuous variables are variables that can take any value within a certain interval and whose values ​​are continuous, such as the probability values ​​corresponding to feature data under different class labels. The observed data samples corresponding to each feature data are obtained by combining the discrete and continuous variables.

[0060] Step S202 : obtaining the category label of the discrete variable, grouping the continuous variable according to the category label to obtain a plurality of grouping variables, and calculating the ranking sum of each of the grouping variables.

[0061] In this embodiment, the category label is the category to which the feature data in the original high-dimensional data belongs. The original high-dimensional data usually corresponds to multiple different category labels. The same feature data may have different values ​​under different category labels. Therefore, the category label of the original high-dimensional data is obtained, and the continuous variables in the observation data sample are grouped according to the category label to obtain multiple grouping variables. The number of groups of the grouping variables is the same as the number of category labels. For example, if a discrete variable corresponds to K category labels, then the continuous variable Y is grouped according to the category label, and Y can be divided into K grouping variables, which include one or more continuous variables. When the grouping variable is obtained, the sorting sum of the grouping variable is calculated. Specifically, the continuous variables of all observation data samples are sorted in ascending order to obtain the permutation number of each continuous variable; when the grouping variable is obtained, the permutation number of all continuous variables in each grouping variable is obtained, and the permutation number is summed to obtain the sorting sum of each grouping variable.

[0062] Step S203: Based on a preset detection algorithm, all the sorting and calculation of each feature data are performed to obtain evaluation parameters of each feature data, all the evaluation parameters are transposed to obtain a correlation vector of the original high-dimensional data, and feature screening of the original high-dimensional data is performed according to the correlation vector to obtain a marker feature.

[0063] In this embodiment, the preset detection algorithm is a correlation detection algorithm, such as the KW (Kruskal-Wallis) detection algorithm. By calculating all the sorting sums of each feature data through the preset detection algorithm, the evaluation parameters of each feature data can be calculated. Among them, the number of sorting sums calculated for each feature data is the same as the number of grouping variables of the feature data. When the evaluation parameters of all feature data of the original high-dimensional data are obtained, an evaluation matrix is ​​formed according to the evaluation parameters, and the evaluation matrix is ​​transposed to obtain the correlation vector of the original high-dimensional data. The correlation vector ω can be expressed as: ω=(ω1,…,ω p ) T , where ω p is the evaluation parameter. When the correlation vector is obtained, the feature data is sorted according to the correlation vector, and a preset number of feature data are selected as marker features. The marker features are the feature data obtained by preliminary screening of the original high-dimensional data.

[0064] Step S204: obtaining a target feature screening model and continuous parameters corresponding to each feature data, inputting the marker features and the continuous parameters into the target feature screening model, and calculating the target dimensionality reduction features of the original high-dimensional data.

[0065] In this embodiment, the target feature screening model is a pre-trained feature screening model, such as the SCAD (Smoothly Clipped Absolute Deviation) model and the LASSO model. The continuous parameter is a continuous parameter corresponding to each feature data in the original high-dimensional data, and the continuous parameter is also a continuous variable. When the marker feature is obtained, the marker feature and the continuous parameter are input into the target feature screening model, and the target dimensionality reduction feature is obtained based on the target feature screening calculation.

[0066] It should be emphasized that in order to further ensure the privacy and security of the above-mentioned target dimensionality reduction features, the above-mentioned target dimensionality reduction features can also be stored in a node of a blockchain.

[0067] The blockchain referred to in this application is a new application model for computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and encryption algorithms. Blockchain is essentially a decentralized database, a series of data blocks generated using cryptographic methods. Each data block contains information about a batch of network transactions, which is used to verify the validity of this information (to prevent counterfeiting) and generate the next block. Blockchain can include the underlying blockchain platform, the platform product service layer, and the application service layer.

[0068] This application realizes the feature screening of high-dimensional data, and the model is robust to data outliers and long-tail distributions. It can process data with continuous dependent variables and discrete features, thereby improving the accuracy and efficiency of high-dimensional data feature screening.

[0069] In some optional implementations of this embodiment, the step of performing all the sorting and calculations on each feature data based on a preset detection algorithm to obtain an evaluation parameter for each feature data includes:

[0070] Obtaining a detection function of the preset detection algorithm, and calculating all the sorting sums of each feature data according to the detection function to obtain a detection statistic of the feature data;

[0071] A target distribution table is obtained, and evaluation parameters of the feature data are obtained by searching the target distribution table based on the detection statistic.

[0072] In this embodiment, when a preset detection algorithm is obtained, the sorting and calculation are performed according to the detection function in the preset detection algorithm to obtain the detection statistic of the feature data. The calculation formula of the detection function is as follows:

[0073]

[0074] Where H is the test statistic, n i is the number of samples in the i-th grouping variable, R i is the sort sum, and K is the number of category labels.

[0075] This embodiment calculates the evaluation parameters of each feature data of the preset detection algorithm, so that the features of the original high-dimensional data can be further preliminarily screened through the evaluation parameters, further improving the accuracy of high-dimensional feature screening.

[0076] In some optional implementations of this embodiment, the step of obtaining the evaluation parameter of the feature data from the target distribution table based on the detection statistic includes:

[0077] Calculating the degree of freedom of the feature data according to the category label;

[0078] An evaluation parameter of the feature data is obtained by searching the target distribution table based on the degrees of freedom and the detection statistic.

[0079] In this embodiment, the test statistic approximately follows a chi-square distribution with (K-1) degrees of freedom under the null hypothesis. Therefore, the degrees of freedom of the feature data in the original high-dimensional data can be calculated based on the number of class labels. A preset target distribution table, such as a chi-square test table, is then obtained. Based on the test statistic and the degrees of freedom, the evaluation parameters corresponding to the current feature data can be found in the target distribution table.

[0080] This embodiment searches the target distribution table through the degrees of freedom and the detection statistic to obtain the evaluation parameters of each feature data, so that the discrete features of the original high-dimensional data can be directly preliminarily screened through the evaluation parameters, thereby reducing the dimensionality of the original high-dimensional data and improving the accuracy of the feature screening of the original high-dimensional data.

[0081] In some optional implementations of this embodiment, the step of calculating the sorted sum of each of the grouping variables includes:

[0082] Sorting all the continuous variables of the observation data sample in ascending order according to their numerical values ​​to obtain a sorting number for each continuous variable;

[0083] The sample size in the grouping variable is obtained, and the ranking sum of the grouping variable is calculated according to the sample size and the ranking number.

[0084] In this embodiment, the grouping variables include one or more continuous variables. To obtain the grouping variables, the continuous variables for all observed data samples in the feature data are sorted from smallest to largest numerically, obtaining a ranking number for the continuous variables. The ranking numbers are 1, ..., n, where n is the total number of continuous variables in the observed data samples of the feature data. The number of samples in each grouping variable is obtained, and the ranking sum of each grouping variable is calculated based on the sample number and the ranking number.

[0085] This embodiment calculates the ranking sum of each grouping variable, so that the correlation between continuous variables and discrete variables can be further evaluated and calculated through the ranking sum, thereby improving the accuracy of high-dimensional data feature screening.

[0086] In some optional implementations of this embodiment, the step of performing feature screening on the original high-dimensional data according to the correlation vector to obtain a marker feature includes:

[0087] Sorting all elements of the correlation vector in ascending order to obtain a total feature set;

[0088] A preset screening threshold is obtained, elements in the total feature set are screened according to the preset screening threshold to obtain a screening vector, feature data corresponding to the screening vector is obtained, and the marking feature is obtained.

[0089] In this embodiment, the correlation vector represents the correlation between discrete variables and continuous variables. When the correlation vector is obtained, all elements in the correlation vector are sorted in ascending order to obtain the total feature set. Then, a preset screening threshold is obtained, which is d n is a pre-set screening threshold, such as d n = [n / log(n)], filter the elements in the total feature set according to the preset screening threshold, and determine the elements in the total feature set before the preset screening threshold as the screening vector. Obtain the feature vector corresponding to the screening vector, that is, obtain the marker feature of the original high-dimensional data. The screening vector can be expressed as: Where p is the total amount of feature data in the original high-dimensional data.

[0090] This embodiment performs feature selection on original high-dimensional data based on correlation vectors, thereby achieving initial feature extraction of discrete variables and improving the accuracy of high-dimensional feature extraction.

[0091] In some optional implementations of this embodiment, the step of inputting the marker feature and the continuous parameter into the target feature screening model to calculate the target dimensionality reduction feature of the original high-dimensional data includes:

[0092] Obtaining a penalty function for the target feature screening model;

[0093] The penalty function is solved according to the marker feature and the continuous parameter to obtain a sparse solution, and the sparse solution is determined to be the target dimensionality reduction feature.

[0094] In the present embodiment, when the marker features and continuous parameters are obtained, the marker features and continuous parameters are input into the target feature screening model, wherein the target feature screening model is a preset linear regression mathematical model, such as the lasso model and the SCAD (Smoothly Clipped Absolute Deviation) model. Based on the target feature screening model, feature selection can be performed on the marker features according to the input continuous variables. Specifically, selecting features by the target feature screening model actually refers to the process of inputting continuous parameters and marker features into the target feature screening model, solving the features where the coefficients are 0 and estimating the features where other coefficients are not 0, and finally obtaining a sparse solution. Therefore, the penalty function of the target feature screening model is obtained, the marker features and continuous parameters are input into the penalty function, and a sparse solution is obtained by solving the penalty function. The sparse solution is the final target dimensionality reduction feature of the original high-dimensional data obtained by screening.

[0095] This embodiment obtains the target dimensionality reduction feature by solving the penalty function of the target feature screening model, thereby achieving re-screening of the feature, reducing the prediction variance of the model, and further improving the accuracy and speed of feature screening.

[0096] In some optional implementations of this embodiment, the step of inputting the marker feature and the continuous parameter into the target feature screening model to calculate the target dimensionality reduction feature of the original high-dimensional data includes:

[0097] The correlation index between the marker feature and the continuous parameter is calculated based on the target feature screening model, and the marker feature is screened according to the correlation index to obtain the target dimensionality reduction feature.

[0098] In this embodiment, the target feature screening model can also use a random forest model, such as xgboost, lightgbm, etc. The correlation between the marker feature and the continuous parameter is calculated based on the target feature screening model to obtain a correlation index. Then, a random forest is generated based on the random forest recursive algorithm and the correlation index. The number of times each marker feature appears in each tree in the random forest is calculated. Based on this number of occurrences, the variable importance of each marker feature in the entire random forest is calculated. Ultimately, a preset number of marker features whose variable importance is greater than or equal to an importance threshold are determined as target dimensionality reduction features for the original high-dimensional data.

[0099] This embodiment screens marker features through a target feature screening model, thereby further improving the efficiency of feature screening.

[0100] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing related hardware via computer-readable instructions. The computer-readable instructions can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes in the above-described method embodiments. The aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0101] It should be understood that although the steps in the flowcharts of the accompanying drawings are shown in sequence as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some of the steps in the flowcharts of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a portion of the sub-steps or stages of other steps.

[0102] Further references Figure 3 , as a response to the above Figure 2 The present application provides an embodiment of a high-dimensional feature extraction device. Figure 2 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0103] like Figure 3 As shown, the high-dimensional feature extraction device 300 of this embodiment includes: a construction module 301, a grouping module 302, an evaluation module 303 and a calculation module 304. Among them:

[0104] A construction module 301 is used to obtain feature data of original high-dimensional data and construct observation data samples for each feature data, wherein the observation data samples include discrete variables and continuous variables;

[0105] In this embodiment, the original high-dimensional data is the initial high-dimensional data. In the medical field, biological omics data is usually high-dimensional data, such as proteome, transcriptome (RNA) and genome (DNA). The feature data of the original high-dimensional data is obtained. The feature data is a discrete feature obtained by collecting the original high-dimensional data, and an observation data sample of each feature data is constructed. Specifically, one feature data can construct multiple observation data samples, and the number of observation data samples is determined by the category label of the original high-dimensional data. When obtaining the feature data, the category label of the original high-dimensional data and the feature value of the feature data under each category label are obtained. The feature value is a discrete variable; then, the continuous variable corresponding to the discrete variable under each category label is obtained, and the observation data sample of the feature data is constructed according to the discrete variable and the continuous variable. The observation data sample can be obtained by (X i ,Y i ), where Y is a continuous variable, X is a discrete variable, and X i ∈{1,…,K}, where K is the observed class label. Discrete variables are feature variables whose values ​​can be listed one by one in a certain order, usually taking integer values, such as the feature data under different class labels of the proteome or transcriptome. Continuous variables are variables that can take any value within a certain interval and whose values ​​are continuous, such as the probability values ​​corresponding to feature data under different class labels. The observed data samples corresponding to each feature data are obtained by combining the discrete and continuous variables.

[0106] A grouping module 302 is configured to obtain a category label of the discrete variable, group the continuous variable according to the category label to obtain a plurality of grouping variables, and calculate a ranking sum of each of the grouping variables;

[0107] In some optional implementations of this embodiment, the grouping module 302 includes:

[0108] A first sorting unit is configured to sort all the continuous variables of the observation data sample in ascending order according to their numerical values ​​to obtain a sorting number for each continuous variable;

[0109] A calculation unit is used to obtain the sample size in the grouping variable, and calculate the ranking sum of the grouping variable according to the sample size and the ranking number.

[0110] In this embodiment, the category label is the category to which the feature data in the original high-dimensional data belongs. The original high-dimensional data usually corresponds to multiple different category labels. The same feature data may have different values ​​under different category labels. Therefore, the category label of the original high-dimensional data is obtained, and the continuous variables in the observation data sample are grouped according to the category label to obtain multiple grouping variables. The number of groups of the grouping variables is the same as the number of category labels. For example, if a discrete variable corresponds to K category labels, then the continuous variable Y is grouped according to the category label, and Y can be divided into K grouping variables, which include one or more continuous variables. When the grouping variable is obtained, the sorting sum of the grouping variable is calculated. Specifically, the continuous variables of all observation data samples are sorted in ascending order to obtain the permutation number of each continuous variable; when the grouping variable is obtained, the permutation number of all continuous variables in each grouping variable is obtained, and the permutation number is summed to obtain the sorting sum of each grouping variable.

[0111] An evaluation module 303 is configured to calculate all the sorting sums of each feature data based on a preset detection algorithm to obtain an evaluation parameter for each feature data, transpose all the evaluation parameters to obtain a correlation vector of the original high-dimensional data, and perform feature screening on the original high-dimensional data based on the correlation vector to obtain a marker feature;

[0112] In some optional implementations of this embodiment, the evaluation module 303 includes:

[0113] a first acquiring unit, configured to acquire a detection function of the preset detection algorithm, and calculate all the sorting sums of each feature data according to the detection function to obtain a detection statistic of the feature data;

[0114] A search unit is configured to obtain a target distribution table, and to search the target distribution table to obtain an evaluation parameter of the feature data based on the detection statistic.

[0115] In some optional implementations of this embodiment, the searching unit includes:

[0116] a calculation subunit, configured to calculate the degree of freedom of the feature data according to the category label;

[0117] A search subunit is configured to search for an evaluation parameter of the feature data from the target distribution table based on the degrees of freedom and the detection statistic.

[0118] In some optional implementations of this embodiment, the evaluation module 303 further includes:

[0119] a second sorting unit, configured to sort all elements of the correlation vector in ascending order to obtain a total feature set;

[0120] The first screening unit is configured to obtain a preset screening threshold, screen the elements in the total feature set according to the preset screening threshold to obtain a screening vector, obtain feature data corresponding to the screening vector, and obtain the marking feature.

[0121] In this embodiment, the preset detection algorithm is a correlation detection algorithm, such as the KW (Kruskal-Wallis) detection algorithm. By calculating all the sorting sums of each feature data through the preset detection algorithm, the evaluation parameters of each feature data can be calculated. Among them, the number of sorting sums calculated for each feature data is the same as the number of grouping variables of the feature data. When the evaluation parameters of all feature data of the original high-dimensional data are obtained, an evaluation matrix is ​​formed according to the evaluation parameters, and the evaluation matrix is ​​transposed to obtain the correlation vector of the original high-dimensional data. The correlation vector ω can be expressed as: ω=(ω1,…,ω p ) T , where ω p is the evaluation parameter. When the correlation vector is obtained, the feature data is sorted according to the correlation vector, and a preset number of feature data are selected as marker features. The marker features are the feature data obtained by preliminary screening of the original high-dimensional data.

[0122] The calculation module 304 is used to obtain a target feature screening model and continuous parameters corresponding to each feature data, input the marker features and the continuous parameters into the target feature screening model, and calculate the target dimensionality reduction features of the original high-dimensional data.

[0123] In some optional implementations of this embodiment, the calculation module 304 includes:

[0124] A second acquisition unit is used to obtain a penalty function of the target feature screening model;

[0125] A solving unit is used to solve the penalty function according to the label feature and the continuous variable to obtain a sparse solution, and determine the sparse solution as the target dimensionality reduction feature.

[0126] In some optional implementations of this embodiment, the calculation module 304 further includes:

[0127] The second screening unit is used to calculate the correlation index between the marker feature and the continuous parameter based on the target feature screening model, and screen the marker feature according to the correlation index to obtain the target dimensionality reduction feature.

[0128] In this embodiment, the target feature screening model is a pre-trained feature screening model, such as the SCAD (Smoothly Clipped Absolute Deviation) model and the LASSO model. The continuous parameter is a continuous parameter corresponding to each feature data in the original high-dimensional data, and the continuous parameter is also a continuous variable. When the marker feature is obtained, the marker feature and the continuous parameter are input into the target feature screening model, and the target dimensionality reduction feature is obtained based on the target feature screening calculation.

[0129] It should be emphasized that in order to further ensure the privacy and security of the above-mentioned target dimensionality reduction features, the above-mentioned target dimensionality reduction features can also be stored in a node of a blockchain.

[0130] The blockchain referred to in this application is a new application model for computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and encryption algorithms. Blockchain is essentially a decentralized database, a series of data blocks generated using cryptographic methods. Each data block contains information about a batch of network transactions, which is used to verify the validity of this information (to prevent counterfeiting) and generate the next block. Blockchain can include the underlying blockchain platform, the platform product service layer, and the application service layer.

[0131] The high-dimensional feature extraction device proposed in this embodiment realizes feature screening of high-dimensional data, and the model is robust to data outliers and long-tail distributions, and can process data with continuous dependent variables and discrete features, thereby improving the accuracy and efficiency of high-dimensional data feature screening.

[0132] To solve the above technical problems, the present application also provides a computer device. Figure 4 , Figure 4 This is a basic structural block diagram of the computer device in this embodiment.

[0133] The computer device 6 includes a memory 61, a processor 62, and a network interface 63 that are interconnected through a system bus. It should be noted that the figure only shows a computer device 6 with components 61-63, but it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0134] The computer device may be a desktop computer, notebook computer, PDA, cloud server, etc. The computer device may interact with the user via a keyboard, mouse, remote control, touchpad, or voice control device.

[0135] The memory 61 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic storage, magnetic disk, optical disk, etc. In some embodiments, the memory 61 can be an internal storage unit of the computer device 6, such as the hard disk or memory of the computer device 6. In other embodiments, the memory 61 can also be an external storage device of the computer device 6, such as a plug-in hard disk equipped on the computer device 6, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. Of course, the memory 61 can also include both the internal storage unit of the computer device 6 and its external storage device. In this embodiment, the memory 61 is generally used to store the operating system and various application software installed on the computer device 6, such as computer-readable instructions for the high-dimensional feature extraction method. In addition, the memory 61 can also be used to temporarily store various types of data that have been output or are to be output.

[0136] In some embodiments, the processor 62 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip. The processor 62 is generally used to control the overall operation of the computer device 6. In this embodiment, the processor 62 is used to execute computer-readable instructions stored in the memory 61 or process data, such as computer-readable instructions for executing the high-dimensional feature extraction method.

[0137] The network interface 63 may include a wireless network interface or a wired network interface. The network interface 63 is generally used to establish a communication connection between the computer device 6 and other electronic devices.

[0138] The computer device proposed in this embodiment realizes feature screening of high-dimensional data, and the model is robust to data outliers and long-tail distributions. It can process data with continuous dependent variables and discrete features, thereby improving the accuracy and efficiency of feature screening of high-dimensional data.

[0139] The present application also provides another embodiment, namely, providing a computer-readable storage medium, which stores computer-readable instructions, and the computer-readable instructions can be executed by at least one processor to enable the at least one processor to perform the steps of the high-dimensional feature extraction method as described above.

[0140] The computer-readable storage medium proposed in this embodiment realizes feature screening of high-dimensional data, and the model is robust to data outliers and long-tail distributions, and can process data with continuous dependent variables and discrete features, thereby improving the accuracy and efficiency of feature screening of high-dimensional data.

[0141] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for enabling a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0142] Obviously, the embodiments described above are only some of the embodiments of the present application, rather than all of the embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application has been described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions described in the aforementioned specific embodiments, or to make equivalent replacements for some of the technical features therein. Any equivalent structure made using the contents of the present application specification and the accompanying drawings, directly or indirectly used in other related technical fields, is also within the scope of patent protection of the present application.

Claims

1. A high-dimensional feature extraction method, characterized in that: The steps include: Acquire feature data of the original high-dimensional data, and construct observation data samples for each of the feature data, wherein the observation data samples include discrete variables and continuous variables, the feature data are omics data, the discrete variables are feature values ​​of the feature data under each category label of the original high-dimensional data, and the continuous variables are continuous variables corresponding to the discrete variables under each category label; Obtaining a category label of the discrete variable, grouping the continuous variable according to the category label to obtain a plurality of grouping variables, and calculating a ranking sum of each of the grouping variables; Calculating all the sorting sums of each feature data based on a preset detection algorithm to obtain evaluation parameters for each feature data, transposing all the evaluation parameters to obtain a correlation vector of the original high-dimensional data, and performing feature screening on the original high-dimensional data according to the correlation vector to obtain a marker feature, where the marker feature is feature data obtained by preliminary screening of the original high-dimensional data; Obtaining a target feature screening model and a continuous parameter corresponding to each feature data, inputting the marker feature and the continuous parameter into the target feature screening model, and calculating the target dimensionality reduction feature of the original high-dimensional data; The step of calculating the sorted sum of each of the grouping variables specifically includes: Sort the continuous variables of all observation data samples in ascending order to obtain the arrangement sequence number of each continuous variable; When the grouping variables are obtained, the arrangement numbers of all continuous variables in each grouping variable are obtained, and the arrangement numbers are summed to obtain the sorted sum of each grouping variable.

2. The high-dimensional feature extraction method according to claim 1, wherein The step of calculating all the sorting and summing of each feature data based on a preset detection algorithm to obtain an evaluation parameter of each feature data comprises: Obtaining a detection function of the preset detection algorithm, and calculating all the sorting sums of each feature data according to the detection function to obtain a detection statistic of the feature data; A target distribution table is obtained, and evaluation parameters of the feature data are obtained by searching the target distribution table based on the detection statistic.

3. The high-dimensional feature extraction method according to claim 2, characterized in that The step of obtaining the evaluation parameter of the feature data from the target distribution table based on the detection statistic comprises: Calculating the degree of freedom of the feature data according to the category label; An evaluation parameter of the feature data is obtained by searching the target distribution table based on the degrees of freedom and the detection statistic.

4. The high-dimensional feature extraction method according to claim 1, wherein The step of performing feature screening on the original high-dimensional data according to the correlation vector to obtain a marker feature comprises: Sorting all elements of the correlation vector in ascending order to obtain a total feature set; A preset screening threshold is obtained, elements in the total feature set are screened according to the preset screening threshold to obtain a screening vector, feature data corresponding to the screening vector is obtained, and the marking feature is obtained.

5. The high-dimensional feature extraction method according to claim 1, wherein The step of inputting the marker features and the continuous parameters into the target feature screening model to calculate the target dimensionality reduction features of the original high-dimensional data comprises: Obtaining a penalty function for the target feature screening model; The penalty function is solved according to the marker feature and the continuous variable to obtain a sparse solution, and the sparse solution is determined to be the target dimensionality reduction feature.

6. The high-dimensional feature extraction method according to claim 1, characterized in that: The step of inputting the marker features and the continuous parameters into the target feature screening model to calculate the target dimensionality reduction features of the original high-dimensional data further includes: The correlation index between the marker feature and the continuous parameter is calculated based on the target feature screening model, and the marker feature is screened according to the correlation index to obtain the target dimensionality reduction feature.

7. A high-dimensional feature extraction device, characterized in that: include: A construction module is used to obtain feature data of the original high-dimensional data and construct observation data samples for each feature data, wherein the observation data samples include discrete variables and continuous variables, the feature data is omics data, the discrete variables are the feature values ​​of the feature data under each category label of the original high-dimensional data, and the continuous variables are the continuous variables corresponding to the discrete variables under each category label; a grouping module, configured to obtain a category label of the discrete variable, group the continuous variable according to the category label to obtain a plurality of grouping variables, and calculate a ranking sum of each of the grouping variables; Wherein, the grouping module includes a first sorting unit and a calculation unit; The first sorting unit is used to sort the continuous variables of all observation data samples in ascending order to obtain the arrangement sequence number of each continuous variable; The calculation unit is used to obtain the arrangement sequence numbers of all continuous variables in each grouping variable when obtaining the grouping variables, and sum the arrangement sequence numbers to obtain the sorted sum of each grouping variable; an evaluation module, configured to calculate all the sorting sums of each feature data based on a preset detection algorithm to obtain evaluation parameters for each feature data, transpose all the evaluation parameters to obtain a correlation vector of the original high-dimensional data, perform feature screening on the original high-dimensional data according to the correlation vector to obtain a marker feature, wherein the marker feature is feature data obtained by preliminary screening of the original high-dimensional data; A calculation module is used to obtain a target feature screening model and continuous parameters corresponding to each feature data, input the marker features and the continuous parameters into the target feature screening model, and calculate the target dimensionality reduction features of the original high-dimensional data.

8. A computer device comprising a memory and a processor, wherein the memory stores computer-readable instructions, and the processor implements the steps of the high-dimensional feature extraction method according to any one of claims 1 to 6 when executing the computer-readable instructions.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the high-dimensional feature extraction method according to any one of claims 1 to 6.

Images