A data processing method, apparatus and related device

Abstract

The application provides a data processing method for improving the efficiency of solving the eigenvalue and eigenvector of a matrix. The data processing method comprises: obtaining a target matrix; searching a first database according to a plurality of characteristic parameters of the target matrix to determine at least one target reference eigenvector, the first database storing a plurality of sets of first reference data, each set of first reference data corresponding to a first reference matrix, each set of first reference data comprising a plurality of characteristic parameters and an eigenvector of the first reference matrix, the plurality of target reference eigenvectors being the eigenvectors of a target first reference matrix, the plurality of characteristic parameters of the target first reference matrix matching the plurality of characteristic parameters of the target matrix; and determining an initial reference vector according to the at least one target reference eigenvector, the initial reference vector being used to solve the eigenvalue and eigenvector of the target matrix. In addition, the application also provides a corresponding device, a computing device cluster, a computer readable storage medium and a computer program product.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a data processing method, apparatus and related equipment. Background Technology

[0002] Eigenvalue solving refers to finding the eigenvalues ​​and eigenvectors of a matrix, and it is a common application scenario in engineering fields. Eigenvalue solving is required in fields such as mechanics simulation, circuit simulation, signal processing, and image processing.

[0003] In practical applications, the matrices to be solved often have high dimensionality. For eigenvalue problems involving large sparse matrices, due to limitations in computer memory and computational power, iterative methods are currently widely used to solve for the eigenvalues ​​and eigenvectors. In the iterative calculation process, the initial reference vector is iteratively expanded to ultimately determine the eigenvalues ​​and eigenvectors of the matrix. If the initial reference vector has a high similarity to the actual eigenvectors, the required number of iterations may be less, and the computational load during the iteration process may also be reduced.

[0004] It is evident that, in order to improve the solution efficiency, the similarity between the initial reference vector and the actual eigenvectors of the matrix can be increased. Summary of the Invention

[0005] In view of this, this application provides a data processing method to improve the efficiency of feature solving. This application also provides corresponding apparatus, computing device clusters, computer-readable storage media, and computer program products.

[0006] Firstly, this application provides a data processing method. This method, which can be implemented by a data processing device, improves the efficiency of solving for the eigenvalues ​​and eigenvectors of a matrix by increasing the similarity between the initial vector and the actual eigenvectors. Specifically, if it is necessary to solve for the eigenvalues ​​and eigenvectors of a target matrix, the target matrix can first be obtained. Next, the eigenvalues ​​of the target matrix can be determined, and then a first database is searched based on the eigenvalues ​​of the target matrix to determine at least one target reference eigenvector. The first database stores multiple sets of first reference data. Each set of first reference data corresponds to a first reference matrix. The first reference data includes multiple eigenvalues ​​and eigenvectors of the corresponding first reference matrix. The target reference eigenvector is the eigenvector of the target first reference matrix. The eigenvalues ​​of the target first reference matrix match the multiple eigenvalues ​​of the target matrix. After determining at least one target reference eigenvector, an initial reference vector can be determined based on the at least one target reference eigenvector so that the initial reference vector can be used as the initial vector to solve for the eigenvalues ​​and eigenvectors of the target matrix. That is, the initial reference vector used to solve for the eigenvalues ​​and eigenvectors of the target matrix is ​​neither a randomly generated vector nor a vector manually set by a technician, but is obtained based on the eigenvectors of the target first reference matrix. Since the eigenvalues ​​and eigenvectors of the first reference matrix match those of the target matrix, the similarity between their eigenvectors and the target matrix can be considered high. Therefore, by using multiple target reference eigenvectors to obtain initial reference vectors, we can obtain initial reference vectors with high similarity to the eigenvectors of the target matrix. This improves the similarity between the initial reference vectors and the eigenvectors when solving for eigenvalues ​​and eigenvectors, thus increasing the efficiency of solving for eigenvalues ​​and eigenvectors.

[0007] In some possible implementations, the target matrix's eigenvalues ​​include at least one or more of the following: the number of columns, the rank, the number of non-zero elements, and the bandwidth of its diagonal elements. Thus, by searching based on these eigenvalues, a first reference matrix that is similar to the target matrix can be found.

[0008] In some possible implementations, the target feature reference vector can be determined through vector matching. Specifically, a feature parameter vector can be established based on multiple feature parameters of the target matrix, and vector matching can be performed from a first database based on the feature parameter vector to determine the vector corresponding to the target first reference matrix. The vector composed of the feature parameters of the target first reference matrix matches the feature parameter vector. In this way, through vector matching, a target first reference matrix similar to the target matrix can be quickly and accurately determined, thereby determining at least one target reference feature vector.

[0009] In some possible implementations, the eigenvalues ​​of the matrix can be divided into first-class eigenvalues ​​and second-class eigenvalues, and the matching of the eigenvalues ​​of the target matrix with those of the first reference matrix can be determined based on different criteria. Specifically, for the first-class eigenvalues, the first-class eigenvalues ​​of the target first reference matrix must be identical to those of the target matrix. The first-class eigenvalues ​​may include, for example, the number of rows in the matrix. For the second-class eigenvalues, the similarity between the second-class eigenvalues ​​of the target first reference matrix and those of the target matrix must meet a preset condition. In this way, matching using different criteria ensures that the eigenvectors of the target first reference matrix can be used as reference vectors for solving the eigenvalues ​​of the target matrix, and improves the efficiency of eigenvalue solving.

[0010] In some possible implementations, the initial spectral shift can be determined based on the matrix's eigenvalues, and the target matrix can be obtained through the spectral shift. Specifically, the original matrix can be obtained first, and then a second database can be searched based on multiple eigenvalues ​​of the original matrix to determine at least one target reference eigenvalue. Similar to the first database, the second database stores multiple sets of second reference data, each set corresponding to a second reference matrix, including multiple eigenvalues ​​and eigenvalues ​​of the corresponding second reference matrix. The original matrix corresponds to the target second reference matrix among the multiple second reference matrices, and the eigenvalues ​​of the target second reference matrix match the eigenvalues ​​of the original matrix. The eigenvalues ​​of the target second reference matrix are the aforementioned target reference eigenvalues. After determining at least one target reference eigenvalue, the spectral shift parameters can be determined based on at least one target reference eigenvalue, and the original matrix can be spectrally shifted according to the spectral shift parameters to obtain the target matrix. In this way, the similarity between the spectral shift parameters and the eigenvalues ​​of the original matrix is ​​improved, thus improving the efficiency of eigenvalue solving.

[0011] In some possible implementations, the spectral shift parameter can be selected from at least one target reference eigenvalue as required. Specifically, configuration information can be obtained first. This configuration information indicates the requirement for the spectral shift parameter. After determining at least one target reference eigenvalue, a target reference eigenvalue can be selected as the spectral shift parameter according to the requirements indicated by the configuration information. In this way, by selecting a suitable target reference eigenvalue as the spectral shift parameter according to the requirements, a suitable spectral shift parameter can be found.

[0012] In some possible implementations, the configuration information can be user-configurable. Optionally, the user can input the configuration information via a client. Accordingly, the configuration information represents the user's requirements for the spectral shift parameters. For example, the user can specify the algorithm for feature solving. When selecting the spectral shift parameters, a suitable target reference eigenvalue can be chosen as the spectral shift parameter based on the algorithm's requirements.

[0013] In some possible implementations, the configuration information can be obtained based on a recommended algorithm. Specifically, if the user does not specify a feature-solving algorithm, a recommendation algorithm can be used to analyze and recommend suitable target algorithms for multiple feature parameters of the original matrix, thereby determining the appropriate target algorithm for matrix solving. Correspondingly, the configuration information is the configuration information corresponding to the target algorithm. In this way, the most suitable algorithm can be selected for feature solving based on the actual situation of the matrix to be solved.

[0014] Secondly, this application provides a data processing apparatus, the apparatus comprising: an acquisition unit for acquiring a target matrix; a search unit for searching a first database based on multiple feature parameters of the target matrix to determine at least one target reference feature vector, the first database storing multiple sets of first reference data, each set of first reference data corresponding to a first reference matrix, each set of first reference data including multiple feature parameters and feature vectors of the first reference matrix, the at least one target reference feature vector being a feature vector of the target first reference matrix, the multiple feature parameters of the target first reference matrix matching the multiple feature parameters of the target matrix; and a vector determination unit for determining an initial reference vector based on the at least one target reference feature vector, the initial reference vector being used to solve for the eigenvalues ​​and eigenvectors of the target matrix.

[0015] In some possible implementations, the target matrix may include at least one or more of the following characteristic parameters: the number of columns of the target matrix, the number of non-zero elements of the target matrix, and the bandwidth of the diagonal elements of the target matrix.

[0016] In some possible implementations, the search unit is specifically used to determine a feature parameter vector based on multiple feature parameters of the target matrix; perform vector matching from the first database based on the feature parameter vector to determine the at least one target reference feature vector, wherein the vector composed of the feature parameters of the target first reference matrix matches the feature parameter vector.

[0017] In some possible implementations, the plurality of feature parameters include at least one first type of feature parameter and at least one second type of feature parameter; the first type of feature parameter of the target matrix is ​​the same as the first type of feature parameter of the target first reference matrix; or, the similarity between the second type of feature parameter of the target matrix and the second type of feature parameter of the target first reference matrix satisfies a preset condition.

[0018] In some possible implementations, the device further includes a spectral shift unit; the acquisition unit is specifically used to acquire the original matrix; the search unit is further used to search a second database based on multiple feature parameters of the original matrix to determine at least one target reference feature value, the second database storing multiple sets of second reference data, each set of second reference data corresponding to a second reference matrix, including multiple feature parameters and feature values ​​of the second reference matrix, the at least one target reference feature value being a feature value of a target second reference matrix, and at least one feature parameter corresponding to the target first reference matrix matching multiple feature parameters of the original matrix; the spectral shift unit is used to determine spectral shift parameters based on the at least one target reference feature value; and to perform spectral shift on the original matrix based on the spectral shift parameters to obtain the target matrix.

[0019] In some possible implementations, the acquisition unit is further configured to acquire configuration information indicating the requirement for the spectral shift parameter;

[0020] The spectral shift unit is further configured to select a target reference feature value from the at least one target reference feature value as the spectral shift parameter according to the configuration information.

[0021] In some possible implementations, the device further includes an algorithm recommendation unit; the algorithm recommendation unit is used to determine a target solving algorithm based on multiple feature parameters of the original matrix through an algorithm recommendation model, the target solving algorithm is used to solve for the eigenvalues ​​and eigenvectors of the target matrix, and the configuration information is the configuration information corresponding to the target solving algorithm.

[0022] Thirdly, this application provides a computing device, the computing device including at least one processor and at least one memory; the at least one memory is used to store instructions, and the at least one processor executes the instructions stored in the at least one memory to cause the computing device to perform the method in the first aspect or any possible implementation thereof. It should be noted that the memory may be integrated into the processor or may be independent of the processor. The at least one computing device may also include a bus. The processor is connected to the memory via the bus. The memory may include readable storage and random access memory.

[0023] Fourthly, this application provides a computing device cluster, the computing device including at least one computing device, the at least one computing device including at least one processor and at least one memory; the at least one memory is used to store instructions, and the at least one processor executes the instructions stored in the at least one memory to cause the computing device cluster to perform the method in the first aspect or any possible implementation of the first aspect. It should be noted that the memory can be integrated into the processor or can be independent of the processor. The at least one computing device may also include a bus. The processor is connected to the memory via the bus. The memory may include readable storage and random access memory.

[0024] Fifthly, this application provides a computer-readable storage medium storing instructions that, when executed on at least one computing device, cause the at least one computing device to perform the method described in the first aspect or any implementation thereof.

[0025] In a sixth aspect, this application provides a computer program product containing instructions that, when run on at least one computing device, cause the at least one computing device to perform the method described in the first aspect or any implementation thereof.

[0026] Based on the implementation methods provided in the above aspects, this application can be further combined to provide more implementation methods. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings.

[0028] Figure 1a A schematic diagram illustrating an application scenario provided in this application embodiment;

[0029] Figure 1b A schematic diagram illustrating an application scenario provided in this application embodiment;

[0030] Figure 2 A flowchart of a data processing method provided in the embodiments of this application;

[0031] Figure 3 Another method flowchart for the data processing method provided in the embodiments of this application;

[0032] Figure 4 A schematic flowchart illustrating the feature solving process provided in this application embodiment;

[0033] Figure 5 This is a schematic diagram of a data processing apparatus provided in an embodiment of this application.

[0034] Figure 6 A schematic diagram of the structure of a computing device provided in an embodiment of this application;

[0035] Figure 7 This is a schematic diagram of a computing device cluster provided in an embodiment of this application;

[0036] Figure 8 This is a schematic diagram illustrating one implementation of a computing device cluster provided in an embodiment of this application. Detailed Implementation

[0037] The solutions in the embodiments provided in this application will now be described with reference to the accompanying drawings.

[0038] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such terms can be used interchangeably where appropriate; this is merely a way of distinguishing objects with the same attributes in the embodiments of this application.

[0039] First, let me introduce some of the terms used in this application.

[0040] Eigenvalues ​​and eigenvectors: Eigenvalues ​​and eigenvectors are parameters related to a matrix. Finding the eigenvalues ​​and eigenvectors of a matrix is ​​a common step in many data processing procedures.

[0041] Eigenvalue derivation: Eigenvalue derivation refers to finding the eigenvalues ​​and eigenvectors of a matrix or a pair of matrices. In a generalized sense, eigenvalue derivation refers to finding an n-dimensional non-zero column vector for two n*n dimensional matrices A and B. And a number λ (λ can be a real number or a complex number) to satisfy The conditions. In this way, it can be... Let λ be an eigenvector of the matrix pair (A, B), and let λ be an eigenvalue of the matrix pair (A, B). In practical applications, to simplify the solution process, matrix B is often transformed into an identity matrix. This simplifies the above conditions to... Accordingly, λ can be called the eigenvector of matrix A, and λ can be called the eigenvalue of matrix A. Based on the condition... The problem of finding eigenvalues ​​and eigenvectors can be called the standardized eigenvalue problem. For ease of introduction, the following text will mainly focus on the standardized eigenvalue problem.

[0042] Feature parameters: In this embodiment, the feature parameters of a matrix refer to the parameters used to represent the characteristics of the matrix. The feature parameters of a matrix can describe its characteristics. If the feature parameters of two matrices match, it can be considered that there is a certain similarity between the two matrices. It is important to emphasize that the feature parameters, eigenvalues, and eigenvectors of a matrix are three different concepts. Eigenvalues ​​and eigenvectors are concepts from the field of mathematics. Feature parameters are the parameters proposed in this embodiment to represent the characteristics of a matrix.

[0043] Initial reference vector: In the process of solving for the eigenvalues ​​and eigenvectors of a matrix using the iterative method, the reference vector can be gradually adjusted through multiple iterations to make it approximate the actual eigenvectors of the matrix. Before solving, an initial reference vector needs to be set. This initial reference vector can be called the initial reference vector.

[0044] Spectral shift: The spectral shift method is an iterative method for solving for the eigenvalues ​​of a matrix. It primarily improves numerical stability and convergence by shifting the eigenvalues. The basic idea is to transform the original matrix A into the form A-σI, where σ is the spectral shift and I is the identity matrix. By using the spectral shift method, the eigenvalues ​​of the matrix can be moved to a more computationally suitable region, thereby improving the efficiency of the solution.

[0045] Finding the eigenvectors and eigenvalues ​​of a matrix is ​​a common practice in engineering. For example, in industrial simulation, Computer-Aided Engineering (CAE) techniques can be used to analyze the structural mechanical properties of engineering projects and products. For instance, in the numerical simulation of a structural component, the component can be modeled first, and then the boundary conditions for the simulation can be set. After mesh generation, discretization methods can be used to transform the numerical simulation into a problem of finding the eigenvalues ​​of the matrix. In engineering, the generalized eigenvalue problem is often simplified to a standardized eigenvalue problem by converting matrix B into an identity matrix.

[0046] Currently, methods for finding the eigenvalues ​​of a matrix include the transformation method and the iterative method. For matrices with high dimensionality, the iterative method is often used. In the iterative process, a reference vector is used as a basis, and multiple iterations are performed to bring the reference vector closer to the actual eigenvectors of the matrix. After the iterations are complete, the final reference vector can be output as the eigenvectors of the matrix, thus completing the eigenvalue calculation.

[0047] Before starting the iteration, an initial reference vector (also known as the initial reference vector) needs to be set. Since the final eigenvector is obtained by iterating from the initial reference vector, the smaller the difference between the initial reference vector and the actual eigenvector of the matrix, the fewer iterations may be required, and the fewer calculations may be needed during the iteration process. In this way, the amount of computation in the solution process is reduced, and the speed of solving for eigenvalues ​​and eigenvectors is improved.

[0048] It is evident that increasing the similarity between the initial reference vector and the actual eigenvectors of the matrix can improve the efficiency of solving for eigenvalues ​​and eigenvectors. However, the eigenvectors of the matrix cannot be determined before they are obtained. Therefore, random vectors are often used as initial reference vectors. Alternatively, in some scenarios, the initial reference vector can be manually set by technicians. However, manually setting the reference vector relies heavily on the experience and expertise of the technicians. This increases the demands on the technicians and may also lead to inaccurate initial reference vectors.

[0049] Based on this, this application provides a data processing method. This method can be implemented by a data processing device, improving the efficiency of solving for the eigenvalues ​​and eigenvectors of a matrix by increasing the similarity between the initial vector and the actual eigenvectors. Specifically, if it is necessary to solve for the eigenvalues ​​and eigenvectors of a target matrix, the target matrix can first be obtained. Next, the eigenvalues ​​of the target matrix can be determined, and then a first database is searched based on the eigenvalues ​​of the target matrix to determine at least one target reference eigenvector. The first database stores multiple sets of first reference data. Each set of first reference data corresponds to a first reference matrix. The first reference data includes multiple eigenvalues ​​and eigenvectors of the corresponding first reference matrix. The target reference eigenvector is the eigenvector of the target first reference matrix. The eigenvalues ​​of the target first reference matrix match the multiple eigenvalues ​​of the target matrix. After determining at least one target reference eigenvector, an initial reference vector can be determined based on the at least one target reference eigenvector, so that the initial reference vector can be used as the initial vector to solve for the eigenvalues ​​and eigenvectors of the target matrix. That is, the initial reference vector used to solve for the eigenvalues ​​and eigenvectors of the target matrix is ​​neither a randomly generated vector nor a vector manually set by a technician, but is obtained based on the eigenvectors of the target first reference matrix. Since the eigenvalues ​​and eigenvectors of the first reference matrix match those of the target matrix, it can be assumed that the eigenvectors of the first reference matrix and the target matrix have a high degree of similarity. Therefore, obtaining an initial reference vector from at least one target reference eigenvector can yield an initial reference vector with a high degree of similarity to the eigenvectors of the target matrix. This improves the similarity between the initial reference vector and the eigenvectors when solving for eigenvalues ​​and eigenvectors, thus increasing the efficiency of solving for eigenvalues ​​and eigenvectors.

[0050] Next, various non-limiting specific implementation methods of the feature solving process will be described in detail.

[0051] First, an exemplary application scenario is introduced. The data processing method provided in this application can be applied to both the client and the server.

[0052] See Figure 1a , Figure 1a This is a schematic diagram illustrating an application scenario of the data processing method provided in an embodiment of this application. Figure 1a In the application scenario shown, the data processing device 111 runs on the client 11. User A can perform feature solving operations through the client 11. The data processing device 111 includes an acquisition unit 1111, a search unit 1112, and a vector determination unit 1113.

[0053] Specifically, user A using client 11 can perform operations on client 11 to allow acquisition unit 111 to acquire the target matrix. Next, search unit 1112 can search a first database based on the feature parameters of the target matrix to determine at least one target reference feature vector. Vector determination unit 1113 can determine an initial reference vector based on multiple target reference feature vectors. The first database may or may not be located on client 11.

[0054] In practical applications, client 11 can be a client of software that needs to solve for the eigenvectors and eigenvalues ​​of a matrix during operation. For example, client 11 can be CAE software. User A can import the model to be simulated into the CAE software client 11. The CAE software client 11 can determine the target matrix based on the model to be simulated and call the data processing device 111 to solve for the eigenvalues ​​and eigenvectors of the target matrix. Optionally, the data processing device 111 can be integrated into client 11 in the form of a library.

[0055] See Figure 1b , Figure 1b This is a schematic diagram illustrating another application scenario of the data processing method provided in the embodiments of this application. Figure 1b In the application scenario shown, the data processing device 121 runs on the server 12. User B can perform feature solving operations through the client 12. The data processing device 121 includes an acquisition unit 1211, a search unit 1212, and a vector determination unit 1213.

[0056] User B can input matrix-related data through client 13. Client 13 can send this data to server 12 via the network or process the data and then send it to server 12. Server 12 can further process the data to determine the target matrix. After the acquisition unit 1211 acquires the target matrix, the search unit 1212 can search the first database 140 based on the feature parameters of the target matrix to determine at least one target reference feature vector. The vector determination unit 1213 can determine an initial reference vector based on at least one target reference feature vector.

[0057] In practical applications, server 12 can be a cloud service server, and client 12 can be software such as a browser. Specifically, user B can access the cloud service client through a browser and call cloud services related to matrix solving. For example, server 12 can provide feature solving services, and user B can call the feature solving service through client 13 and send the matrix to be solved to server 12. Alternatively, server 12 can also provide model simulation services, and user B can upload the model to be simulated to server 12 through client 13. Server 12 can process the model, determine the target matrix to be solved, and perform feature solving through data processing device 121. After determining the eigenvectors and eigenvalues ​​of the target matrix, server 12 can determine the simulation results based on the eigenvectors and eigenvalues ​​and return the simulation results to client 13.

[0058] It should be noted that the above two application scenarios are only examples, and the data processing method provided in this application embodiment can be applied to any application scenario that requires feature solving.

[0059] The following section provides a detailed introduction to the specific implementation methods in the data processing process.

[0060] See Figure 2 , Figure 2 A flowchart illustrating the data processing method provided in this application. This method can be applied to... Figure 1a or Figure 1b The application scenarios shown can also be applied to other applicable application scenarios.

[0061] Specifically, Figure 2 The data processing methods shown may specifically include:

[0062] S201: Obtain the target matrix.

[0063] Before solving for the features, the matrix to be solved can first be obtained. Figure 2 In the implementation shown, the matrix to be solved is called the target matrix. That is, after obtaining the target matrix, the data processing device can use... Figure 2 The data processing method shown solves for the eigenvectors and eigenvalues ​​of the target matrix.

[0064] As discussed earlier, in some implementations, feature solving can be user-triggered. Correspondingly, the target matrix can be obtained based on user actions. For example, if a user needs the eigenvalues ​​and eigenvectors of a matrix, they can call a cloud service for matrix solving through a client and upload the matrix to be solved to the cloud service's server. The data processing device running on the server can then obtain the target matrix.

[0065] In this embodiment, the target matrix can be a matrix obtained directly from user-uploaded data, or it can be a preprocessed matrix. For example, if the spectral shift method is used for feature extraction, the target matrix can be a matrix that has undergone spectral shifting.

[0066] Let's illustrate this with a practical application scenario. If a user needs to simulate the mechanical properties of a structural component, they can first analyze and model the mechanical properties of that component, and then establish a matrix based on the results of the analysis and modeling. Optionally, the target matrix can be this matrix established based on the analysis results. Alternatively, to better solve for features, the spectral displacement parameters can be determined first, and then the results obtained from the analysis and modeling can be spectrally shifted using the spectral displacement parameters. The matrix after spectral shifting can then be used as the target matrix for feature solving.

[0067] For a detailed description of this part, please refer to Figure 3 This will not be elaborated upon here.

[0068] S202: Search the first database based on multiple feature parameters of the target matrix to determine at least one target reference feature vector.

[0069] In this embodiment, to improve the efficiency of feature solving, it is necessary to increase the similarity between the initial reference vector and the actual eigenvectors of the target matrix. Matrixes with similar eigenvectors often have similar eigenparameters. Therefore, the data processing device can first determine multiple eigenparameters of the target matrix. Then, based on these eigenparameters, it searches a first database to find matrices whose eigenparameters are similar to those of the target matrix. The initial vector in the matrix solving process is then determined based on the eigenvectors of the found matrices.

[0070] For ease of explanation, the eigenvectors determined from the first database based on the eigenparameters of the target matrix will be referred to as target reference eigenvectors, the eigenparameters corresponding to the target reference eigenvectors will be referred to as target reference eigenparameters, and the matrix whose eigenparameters are similar to those of the target matrix will be referred to as the target first reference matrix. Accordingly, the target reference eigenparameters are the eigenparameters of the target first reference matrix, and the target reference eigenvectors are the eigenvectors of the target first reference matrix.

[0071] Feature parameters are parameters that represent the characteristics of a matrix. For example, the feature parameters of a target matrix may include one or more of the following: the number of columns, the rank, the number of non-zero elements, and the bandwidth of the diagonal elements. By matching the feature parameters of the target matrix, eigenvectors of matrices similar to the target matrix can be found from a first database.

[0072] The first database stores multiple sets of first reference data. Each set of first reference data corresponds to a matrix with known eigenvectors. These matrices with known eigenvectors can be called first reference matrices. Each set of first reference data includes multiple eigenparameters and eigenvectors of the same first reference matrix. For example, it may include some or all of the eigenvectors of the first reference matrix. Since they come from the first reference matrix, the eigenparameters of the first reference matrix can be called reference eigenparameters, and the eigenvectors of the first reference matrix can be called reference eigenvectors.

[0073] Optionally, the first database can be a relational database or a vector database. If the first database is a vector database, then the first reference data can be a vector, where one element includes multiple feature parameters and the other element includes multiple reference feature vectors. That is, in the first database, each set of first reference data can correspond to a vector, where one element is a vector composed of multiple feature parameters of the first reference matrix, and the other element is a vector composed of multiple feature vectors of the first reference matrix.

[0074] In other words, for a matrix A with known eigenvectors, matrix A can be used as a first reference matrix. Multiple eigenparameters of matrix A can be extracted, and these eigenparameters and eigenvectors can be integrated into a set of first reference data stored in a first database. Thus, when eigenvalues ​​of a new matrix need to be calculated, the first database can be searched based on the eigenparameters of the matrix to be solved, thereby determining whether there exists a first reference matrix with known eigenvectors whose eigenparameters are similar to those of the matrix to be solved.

[0075] Accordingly, when solving for the features of the target matrix, a search can be performed in the first database based on the feature parameters of the target matrix to determine whether there exists a set of reference feature parameters in the first database that match the feature parameters of the target matrix. If such a set exists, this first reference data can be identified as the target first reference data, and the reference feature vector in the target first reference data can be identified as the target reference feature vector.

[0076] In practical applications, matrices often have high dimensionality and many elements. Storing matrices in a database can lead to wasted storage space. Therefore, in some implementations, the first reference data may not include the corresponding first reference matrix. Accordingly, when determining the target reference eigenvector, the data processing device can match the target reference eigenparameters based on the eigenparameters of the target matrix, and then determine the corresponding target reference eigenvector based on the target reference eigenparameters. During this process, the data processing device is unaware of the target first reference matrix.

[0077] Alternatively, in some other possible implementations, the first reference data may also include a first reference matrix corresponding to the reference feature vector. Accordingly, when determining the target reference feature vector, the data processing device can determine the target first reference matrix based on the target reference feature parameters, and then use the target first reference matrix for verification, comparing the similarity between the target first reference matrix and the target matrix to determine the target reference feature vector. Optionally, if the first reference data includes the first reference matrix corresponding to the reference feature vector, the first reference data may not include the reference feature parameters. When matching the feature parameters of the target matrix, the data processing device can determine the reference feature parameters based on the first reference matrix in the first reference data, thereby determining the target reference feature vector.

[0078] In this embodiment, the target reference feature vector can be determined by matching the feature parameters. The feature parameters of the target matrix are matched with the target reference feature parameters. In this embodiment, the target matrix has multiple feature parameters. For a portion of the multiple feature parameters, the target reference feature parameters may be required to be consistent with the feature parameters of the target matrix. For another portion of the multiple feature parameters, the target reference feature parameters may be required to have a high degree of similarity with the feature parameters of the target matrix.

[0079] In other words, multiple feature parameters can be classified into one or more first-class feature parameters and / or one or more second-class feature parameters. The first-class feature parameters of the target matrix (if they exist) are the same as the first-class feature parameters of the target first reference matrix. The second-class feature parameters of the target matrix (if they exist) have a high degree of similarity to the second-class feature parameters of the target first reference matrix.

[0080] For example, according to the definition of eigenvectors, the number of rows in the eigenvectors of a matrix is ​​the same as the number of columns in the matrix. Therefore, the number of columns in the eigenvectors of the target matrix is ​​the same as the number of columns in the target matrix, and the number of rows in the target reference eigenvectors is the same as the number of columns in the target first reference matrix. Correspondingly, the number of columns in the target matrix is ​​the same as the number of columns in the target first reference matrix. Therefore, "the number of rows in a matrix" can be considered a "first type of eigenparameter" as described above. However, two matrices with different eigenparameters, such as "the number of non-zero elements in the target matrix," may have the same eigenvectors. Therefore, eigenparameters such as "the number of non-zero elements in the target matrix" can be considered "second type of eigenparameters" as described above.

[0081] Optionally, when determining the target reference feature vector, a preliminary screening can be performed in a first database based on multiple first-type feature parameters of the target matrix to identify first reference data whose first-type feature parameters are identical to those of the target matrix. Next, the similarity between the second-type feature parameters of the preliminarily selected first reference data and the second-type feature parameters of the target matrix can be compared, and the reference feature vector corresponding to the second-type feature parameter with the highest similarity can be determined as the target reference feature vector. Alternatively, in some other possible implementations, a preliminary screening can be performed first based on the second-type feature parameters, followed by screening based on the first-type feature parameters. Alternatively, multiple screening methods can be used. Alternatively, screening can be performed solely based on the first-type feature parameters or the second-type feature parameters to determine the target first reference matrix. This application does not limit the implementation method of matching feature parameters.

[0082] In some possible implementations, the target reference feature vector can be determined based on vector matching. For example, in an application scenario where the first database is a vector database and one element in the first reference data is a vector composed of the feature parameters of the first reference matrix, the target reference feature vector can be determined based on vector matching. Specifically, multiple feature parameters can be combined into a vector, and then by matching the vector composed of the feature parameters of the target matrix with the vector composed of the reference feature parameters, the target reference data set that matches the feature parameters of the target matrix can be determined, thereby determining the target reference feature vector.

[0083] Specifically, when establishing the first database, the feature parameters of the first reference matrix can be converted into vectors. The vector composed of the feature parameters of the first reference matrix can be called the reference feature parameter vector. When determining the target reference feature vector, the data processing device can first determine the feature parameter vector based on multiple feature parameters of the target matrix. Then, the feature parameter vector and the reference feature parameter vector can be matched, and the first reference data that matches the feature parameter vector is determined as the target first reference data, thereby determining the target reference feature vector.

[0084] Optionally, the feature parameter vector can include multiple elements. For example, each feature parameter can be included as an element in the feature parameter vector. Alternatively, the first type of feature parameters can be combined into a single element of the feature parameter vector. During matching, the elements formed by combining the first type of feature parameters can be compared to see if they are identical. After determining that they are identical, the matching degree between the feature parameter vector and the reference feature parameter vector can be further assessed. Specifically, the feature parameter vector can include two elements. The first element is obtained by combining the first type of feature parameters, and the second element is obtained by combining the second type of feature parameters. During matching, it can be first determined whether the first element of the feature parameter vector is consistent with the first element of the reference feature parameter vector. If they are consistent, the distance between the second element of the feature parameter vector and the second element of the reference feature parameter vector can be calculated. Thus, the reference feature parameter vector with the same first element and the shortest distance between the second elements can be determined as the reference feature parameter vector for the target first reference data.

[0085] S203: Determine an initial reference vector based on at least one target reference feature vector.

[0086] After determining at least one target reference eigenvector, an initial reference vector can be determined based on it. For example, in some implementations, one of the at least one target reference eigenvectors, such as the one with the most non-zero elements, can be chosen as the initial reference vector. Alternatively, in some implementations, the mean vector of multiple target reference eigenvectors can be calculated and used as the initial reference vector. Yet another example is that the initial reference vector can be obtained by linearly combining multiple target reference eigenvectors. Specifically, if the target first reference data includes all the eigenvectors of the target first reference matrix, then the initial reference vector can be obtained by linearly combining all the eigenvectors of the target first reference matrix.

[0087] It's important to clarify that the technical features "initial reference vector" and "reference eigenvector" may be confused in their description, but they are distinct technical features. The initial reference vector provides a reference during the feature solving process and is not an eigenvector of a specific matrix. Logically, the initial reference vector and the eigenvector of the target matrix are different vectors (although they may be numerically equal). Therefore, the initial reference vector is essentially a "reference vector." The "reference eigenvector," on the other hand, is the "eigenvector" of the first reference matrix, serving only as a reference in determining the initial reference vector; it is, in essence, still an "eigenvector."

[0088] Specifically, when linearly combining multiple target reference feature vectors, the weight of each target reference feature vector can be obtained first, and the multiple target reference feature vectors can be linearly combined based on the weights to obtain the target reference vector. The weights of the target reference feature vectors can be pre-set non-zero numbers or randomly generated non-zero numbers.

[0089] After determining the initial reference vector, feature extraction can be performed using it. The initial reference vector is the vector used as the initial vector for iteration during the feature extraction process, equivalent to the initial vector mentioned earlier. Optionally, multiple iterations can be performed using the user-selected algorithm and the initial reference vector to determine the eigenvectors of the target matrix. Alternatively, a preset algorithm can be used for feature extraction. Or, a recommended target extraction algorithm can be used. An introduction to the recommended target extraction algorithms can be found below and will not be repeated here.

[0090] As can be seen from the above introduction, the initial reference vectors used to solve for the eigenvalues ​​and eigenvectors of the target matrix are neither randomly generated vectors nor manually set by technicians. Instead, they are obtained based on the eigenvectors of matrices whose eigenparameters match the eigenparameters of the target matrix. Since the eigenparameters of the matrix corresponding to the target reference eigenvector match the eigenparameters of the target matrix, the eigenvectors of this matrix have a high similarity to the eigenvectors of the target matrix. Therefore, obtaining initial reference vectors from multiple target reference eigenvectors can yield initial reference vectors with high similarity to the eigenvectors of the target matrix. This improves the similarity between the initial reference vector and the eigenvectors when solving for eigenvalues ​​and eigenvectors, thus improving the efficiency of solving for eigenvalues ​​and eigenvectors.

[0091] In the implementation described above, a suitable initial reference matrix can be determined based on the eigenvalues ​​of the target matrix. In practical applications, if the spectral shift method is used to solve for the eigenvalues, the matrix to be solved needs to be adjusted first based on the spectral shift parameters, and then the initial reference vector is used for eigenvalue solving. In this implementation, the target matrix is ​​not the original matrix to be solved, but rather a matrix adjusted based on the original matrix and the spectral shift parameters.

[0092] Similar to the initial reference vector, when using the spectral shift method for eigenvalue solving, the closer the spectral shift parameter is to the eigenvalues ​​of the original matrix, the higher the efficiency of the eigenvalue solving. Therefore, in some implementations, the data processing device can also determine the spectral shift parameter based on the eigenvalues ​​of the original matrix.

[0093] See Figure 3 , Figure 3 Another flowchart illustrating the data processing method provided in this application. Specifically, Figure 3The data processing methods shown may specifically include:

[0094] S301: Obtain the original matrix.

[0095] When determining spectral shift parameters, the original matrix can be obtained first. "Original" in "original matrix" means the original matrix has not undergone spectral shifting. For example, if a user uploads a matrix to be solved to the server, the server can first preprocess the matrix, then perform spectral shifting on the preprocessed matrix, and finally perform eigenvalue solving. In this process, the "preprocessed matrix" is the original matrix.

[0096] S302: Search the second database based on multiple feature parameters of the original matrix to determine at least one target reference feature value.

[0097] After obtaining the original matrix, at least one target reference eigenvalue can be determined by searching the second database based on multiple feature parameters of the original matrix. An explanation of the feature parameters can be found above and will not be repeated here.

[0098] The second database stores multiple sets of second reference data. Each second reference data set includes multiple reference feature parameters and reference eigenvalues ​​corresponding to the same matrix; for example, it may include all eigenvalues ​​of the same matrix. The matrix corresponding to the reference eigenvalues ​​can be called the second reference matrix. The reference feature parameters corresponding to the target reference eigenvalue match the reference feature parameters of the original matrix. At least one target reference eigenvalue corresponds to multiple reference feature parameters that match multiple feature parameters of the original matrix. The second reference matrix corresponding to at least one target reference eigenvalue can be called the target second reference matrix. Optionally, at least one target reference eigenvalue can be all eigenvalues ​​of the target second reference matrix.

[0099] Optionally, the second database and the first database can be the same database, and the first reference data and the second reference data can also be the same data. Specifically, for a reference matrix whose eigenvalues ​​and eigenvectors are known, the eigenvalues, eigenparameters, and eigenvectors of the reference matrix can be associated and stored in the database. In this way, when it is necessary to determine the spectral shift parameters, multiple target reference eigenvalues ​​can be determined based on the eigenparameters of the original matrix. When it is necessary to determine the initial reference vector, multiple target reference eigenvectors can be determined based on the eigenparameters of the target matrix.

[0100] For a detailed explanation of how to search the second database, please refer to the section on searching the first database above; it will not be repeated here.

[0101] S303: Determine the spectral shift parameter based on at least one target reference eigenvalue.

[0102] After determining the target reference eigenvalue, the spectral shift parameter can be determined based on at least one target reference eigenvalue.

[0103] For example, the largest target reference eigenvalue among at least one target reference eigenvalue can be determined as the spectral shift parameter. As another example, the smallest target reference eigenvalue among at least one target reference eigenvalue can be determined as the spectral shift parameter. Yet another example is calculating the average value of at least one target reference eigenvalue and determining this average value as the spectral shift parameter.

[0104] Alternatively, in some possible implementations, the spectral shift parameter can be determined based on at least one target reference eigenvalue according to user configuration. Specifically, the user can set configuration information before triggering the feature solving process. The configuration information indicates the requirements for the spectral shift parameter. For example, it may include information about the standard for the spectral shift parameter selected by the user, or it may include the target solving algorithm selected by the user, indicating the target solving algorithm's requirements for the spectral shift parameter. After determining at least one target reference eigenvalue, a suitable target reference eigenvalue can be selected as the spectral shift parameter based on the configuration information.

[0105] For example, in some implementations, users can configure the feature solving method through configuration information. Accordingly, the configuration information can include the user-defined solving algorithm. When determining spectral shift parameters based on multiple target reference eigenvalues, the target reference eigenvalues ​​suitable for the solving algorithm can be selected as the spectral shift parameters according to the characteristics of the solving algorithm.

[0106] If the user does not specify a target algorithm, a random algorithm or a default algorithm can be used for feature solving. Alternatively, in some possible implementations, a target algorithm can be recommended to the user based on the actual situation. Optionally, the user can trigger an algorithm recommendation operation. In response to the user-triggered algorithm recommendation operation, an algorithm recommendation model can be used to analyze relevant information in the feature solving process, and then select a suitable algorithm as the target algorithm. Accordingly, the configuration information can correspond to the target algorithm, indicating the target algorithm's requirements for spectral displacement parameters.

[0107] The aforementioned "related information in the feature solving process" may include, for example, multiple feature parameters of the original matrix. That is, after obtaining the original matrix, multiple feature parameters of the original matrix can be determined, and the target solving algorithm can be determined based on these feature parameters using an algorithm recommendation model. Optionally, the multiple feature parameters used to determine the target solving algorithm and the multiple feature parameters used to determine the target reference eigenvalues ​​can be completely identical, partially identical, or completely different.

[0108] S304: Perform spectral shift on the original matrix based on the spectral shift parameter to obtain the target matrix.

[0109] After determining the spectral displacement parameters, the original matrix can be spectrally shifted based on the spectral displacement parameters to obtain the target matrix. Specifically, this can be calculated using the formula B = A - σI, where A is the original matrix, B is the target matrix, I is the identity matrix, and σ is the spectral displacement parameter.

[0110] Alternatively, if the spectral shift parameter is zero, the original matrix and the target matrix can be the same matrix. Accordingly, when determining the target reference eigenvector, multiple eigenvalues ​​of the original matrix can be used as multiple eigenvalues ​​of the target matrix.

[0111] exist Figure 3 In the implementation shown, the spectral shift parameters are neither randomly generated nor manually set by technicians. Instead, they are obtained based on the eigenvalues ​​of a matrix whose eigenvalues ​​match those of the original matrix. Since the eigenvalues ​​of the matrix corresponding to the target reference eigenvalues ​​match those of the original matrix, the eigenvalues ​​of this matrix have a high similarity to the eigenvalues ​​of the target matrix. Therefore, spectral shift parameters with high similarity to the eigenvalues ​​of the original matrix can be obtained based on multiple target reference eigenvalues. This improves the similarity between the spectral shift parameters and the eigenvalues ​​of the original matrix, enhances the effectiveness of the spectral shift, and increases the efficiency of eigenvalue solving.

[0112] In the implementation described above, eigenvalues ​​or eigenvectors of similar matrices can be searched from a database based on the matrix's eigenparameters to determine the spectral shift parameters or initial reference matrix. The data in the database is derived from matrices whose eigenvalues ​​and eigenvectors are known. In some scenarios, matrices similar to the target or original matrix with known eigenvalues ​​and eigenvectors may exist, allowing the data processing device to find the target reference eigenvector or eigenvalue. However, in other scenarios, matrices similar to the target or original matrix with known eigenvalues ​​and eigenvectors may not exist. In such cases, the data processing device may be unable to determine the target reference eigenvector from the first database or the target reference eigenvalue from the second database.

[0113] To address this issue, some implementations can generate initial reference vectors or spectral shift parameters using a generation-time model.

[0114] Specifically, when determining the initial reference vector, the data processing device can first search the first database based on multiple feature parameters of the target matrix to determine whether the first database includes multiple target reference feature parameters that match the multiple feature parameters of the target matrix. If they exist, multiple target reference feature vectors are determined according to step S202 above. If they do not exist, the data processing device can input the multiple feature parameters of the target matrix into the generative model, for example, by calling the service corresponding to the generative model through the network and sending the multiple feature parameters of the target matrix. The generative model can generate an initial reference vector suitable for solving the feature of the target matrix based on the multiple feature parameters of the target matrix and return it to the data processing device. The data processing device can obtain the initial reference vector generated by the generative model based on the multiple feature parameters, and then solve for the eigenvalues ​​and eigenvectors of the target matrix based on the initial reference vector. In this way, even if no feature parameter of the first reference matrix matches the feature parameter of the target matrix, a suitable initial reference vector can still be generated by the generative model.

[0115] Similarly, when determining the spectral shift parameters, the data processing device can first search the first database based on the multiple eigenvalues ​​of the original matrix to determine whether the first database includes multiple target reference eigenvalues ​​that match the multiple eigenvalues ​​of the original matrix. If they exist, the multiple target reference eigenvalues ​​are determined according to step S302 above. If they do not exist, the data processing device can input the multiple eigenvalues ​​of the original matrix into the generative model, for example, by calling the service corresponding to the generative model through the network and sending the multiple eigenvalues ​​of the original matrix. The generative model can generate spectral shift parameters suitable for solving the eigenvalues ​​of the original matrix based on the multiple eigenvalues ​​of the original matrix and return them to the data processing device. The data processing device can obtain the spectral shift parameters generated by the generative model based on the multiple eigenvalues, and then solve for the eigenvalues ​​of the original matrix based on the spectral shift parameters. In this way, even if there is no eigenvalue of a first reference matrix that matches the eigenvalue of the original matrix, a suitable spectral shift parameter can still be generated by the generative model.

[0116] Alternatively, the above process can be as follows: Figure 4 As shown. See also Figure 4 , Figure 4 This is a flowchart illustrating the feature solving process provided in this application. Specifically, Figure 4 The feature solving process shown specifically includes steps S401-S408.

[0117] S401: Obtain matrix A and configuration information.

[0118] exist Figure 4In the implementation shown, matrix A is the matrix for which feature extraction is required. The configuration information indicates the user's requirements for feature extraction. The configuration information does not include the feature extraction algorithm.

[0119] S402: Based on the feature parameters of matrix A, perform algorithm recommendation and determine the recommendation algorithm.

[0120] exist Figure 4 The implementation shown can analyze the user's configuration information and recommend suitable feature solving algorithms to the user.

[0121] S403: Determine multiple target reference eigenvalues ​​based on the eigenparameters of matrix A.

[0122] For an explanation of step S403, please refer to the above text; it will not be repeated here.

[0123] S404: Determine the spectral shift parameters from multiple target reference feature values ​​according to the requirements of the recommendation algorithm.

[0124] After determining multiple target reference feature values, the most suitable target reference feature value can be selected as the spectral shift parameter according to the needs of the recommendation algorithm.

[0125] S405: Perform spectral shift on matrix A according to the spectral shift parameter to obtain matrix B.

[0126] After determining the spectral shift parameters, matrix A can be spectrally shifted according to the spectral shift parameters to obtain matrix B after spectral shift.

[0127] S406: Determine multiple target reference eigenvectors based on the eigenparameters of matrix B.

[0128] S407: Perform a linear combination of multiple target reference feature vectors to obtain an initial reference vector.

[0129] For a description of steps S406 and S407, please refer to the above text; they will not be repeated here.

[0130] S408: Solve for the characteristics of matrix A.

[0131] Step S402 above determines the algorithm used in the feature solving process. Step S404 above determines the spectral shift parameters. Step S407 above determines the initial reference vector. Thus, following the recommended algorithm, the initial reference vector is used as the initial vector, and the spectral shift parameters are used to perform spectral shifting on matrix A to solve for the features. In this way, the recommended algorithm is determined based on the user's needs, the spectral shift parameters are selected based on the feature parameters of matrix A, and the initial reference vector is selected based on the feature parameters of matrix B. Therefore, the algorithm and parameters used in the feature solving process not only meet the user's needs but are also closely related to the matrix A to be solved. This improves the efficiency of feature solving.

[0132] This application also provides a data processing apparatus. The data processing apparatus can be applied to... Figure 1a Client 11 or in the implementation shown Figure 1b The server 120 in the implementation shown. Specifically, as... Figure 5 As shown, the data processing device 500 includes:

[0133] Acquisition unit 510 is used to acquire the target matrix;

[0134] The search unit 520 is used to search a first database based on multiple feature parameters of the target matrix to determine multiple target reference feature vectors. The first database stores at least one set of first reference data. Each set of first reference data corresponds to a first reference matrix. Each set of first reference data includes multiple feature parameters and feature vectors of the first reference matrix. The at least one target reference feature vector is a feature vector of the target first reference matrix. The multiple feature parameters of the target first reference matrix match the multiple feature parameters of the target matrix.

[0135] The vector determination unit 530 is used to determine an initial reference vector based on the at least one target reference feature vector, wherein the initial reference vector is used to solve for the eigenvalues ​​and eigenvectors of the target matrix.

[0136] The acquisition unit 510, the search unit 520, and the vector determination unit 530 can all be implemented in software or in hardware. For example, the implementation of the search unit 520 will be described below. Similarly, the implementation of the acquisition unit 510 and the vector determination unit 530 can refer to the implementation of the search unit 520.

[0137] As an example of a software functional unit, the lookup unit 520 may include code running on a computing instance. The computing instance may include at least one of a physical host (computing device), a virtual machine, or a container. Further, the aforementioned computing instance may be one or more. For example, the lookup unit 520 may include code running on multiple hosts / virtual machines / containers. It should be noted that the multiple hosts / virtual machines / containers used to run the code may be distributed within the same region or in different regions. Further, the multiple hosts / virtual machines / containers used to run the code may be distributed within the same availability zone (AZ) or in different AZs, each AZ including one or more geographically proximate data centers. Typically, a region may include multiple AZs.

[0138] Similarly, multiple hosts / virtual machines / containers used to run this code can be distributed within the same Virtual Private Cloud (VPC) or across multiple VPCs. Typically, a VPC is set up within a region. Communication between two VPCs within the same region, as well as between VPCs in different regions, requires a communication gateway to be set up within each VPC to enable interconnection between VPCs.

[0139] As an example of a hardware functional unit, the lookup unit 520 may include at least one computing device, such as a server. Alternatively, the lookup unit 520 may also be a device implemented using an application-specific integrated circuit (ASIC) or a programmable logic device (PLD). The PLD may be implemented using a complex programmable logical device (CPLD), a field-programmable gate array (FPGA), generic array logic (GAL), or any combination thereof.

[0140] The multiple computing devices included in the lookup unit 520 can be distributed in the same region or in different regions. Similarly, the multiple computing devices included in the lookup unit 520 can be distributed in the same Availability Zone (AZ) or in different AZs. Likewise, the multiple computing devices included in the lookup unit 520 can be distributed in the same Virtual Private Cloud (VPC) or in multiple VPCs. These multiple computing devices can be any combination of computing devices such as servers, ASICs, PLDs, CPLDs, FPGAs, and GALs.

[0141] It should be noted that, in other embodiments, the acquisition unit 510 is used to execute any step in the data processing method, the search unit 520 is used to execute any step in the data processing method, and the vector determination unit 530 can be used to execute any step in the data processing method. The steps implemented by the acquisition unit 510, the search unit 520, and the vector determination unit 530 can be specified as needed. Thus, the acquisition unit 510, the search unit 520, and the vector determination unit 530 respectively implement all the functions of the data processing device based on different steps in the data processing method.

[0142] This application also provides a computing device. For example... Figure 6 As shown, the computing device 100 includes a bus 102, a processor 104, a memory 106, and a communication interface 108. The processor 104, the memory 106, and the communication interface 108 communicate with each other via the bus 102. The computing device 100 can be a server or a terminal device. It should be understood that this application does not limit the number of processors and memories in the computing device 100.

[0143] Bus 102 can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, Figure 6 The bus 102 may be represented by a single line, but this does not mean that there is only one bus or one type of bus. The bus 102 may include a path for transmitting information between various components of the computing device 100 (e.g., memory 106, processor 104, communication interface 108).

[0144] The processor 104 may include any one or more processors such as a central processing unit (CPU), a graphics processing unit (GPU), a microprocessor (MP), or a digital signal processor (DSP).

[0145] Memory 106 may include volatile memory, such as random access memory (RAM). Processor 104 may also include non-volatile memory, such as read-only memory (ROM), flash memory, hard disk drive (HDD), or solid state drive (SSD).

[0146] The memory 106 stores executable program code, which the processor 104 executes to implement the functions of the aforementioned acquisition unit 510, search unit 520, and vector determination unit 530, thereby realizing the data processing method. In other words, the memory 106 stores instructions for executing this storage method.

[0147] Alternatively, the memory 106 stores executable code, which the processor 104 executes to implement the functions of the aforementioned data processing device, thereby implementing the data processing method. That is, the memory 106 stores instructions for executing the data processing method.

[0148] The communication interface 108 uses transceiver modules such as, but not limited to, network interface cards and transceivers to enable communication between the computing device 100 and other devices or communication networks.

[0149] This application also provides a computing device cluster. The computing device cluster includes at least one computing device. The computing device can be a server, such as a central server, an edge server, or a local server in a local data center. In some embodiments, the computing device can also be a terminal device such as a desktop computer, a laptop computer, or a smartphone.

[0150] like Figure 7 As shown, the computing device cluster includes at least one computing device 100. The memory 106 of one or more computing devices 100 in the computing device cluster may store the same instructions for executing data processing methods.

[0151] In some possible implementations, the memory 106 of one or more computing devices 100 in the computing device cluster may also store partial instructions for executing data processing methods. In other words, a combination of one or more computing devices 100 can jointly execute instructions for performing basic data processing methods.

[0152] It should be noted that the memories 106 in different computing devices 100 within the computing device cluster can store different instructions, each used to execute a portion of the functions of the data processing device 500. That is, the instructions stored in the memories 106 of different computing devices 100 can implement the functions of one or more units among the acquisition unit 510, the search unit 520, and the vector determination unit 530.

[0153] In some possible implementations, one or more computing devices in a computing device cluster can be connected via a network. This network can be a wide area network (WAN) or a local area network (LAN), etc. Figure 8 One possible implementation is shown. For example... Figure 8 As shown, the two computing devices 100A and 100B are connected via a network. Specifically, they are connected to the network through communication interfaces in each computing device. In this possible implementation, the memory 106 in computing device 100A stores instructions for executing the acquisition unit 510. Meanwhile, the memory 106 in computing device 100B stores instructions for executing the functions of the lookup unit 520 and the vector determination unit 530.

[0154] Figure 8 The connection method between the computing device clusters shown can be based on the fact that the data processing method provided in this application can be divided into two parts: calling the decision model and interacting with the computing cluster. Therefore, it is considered that the function of calling the decision model is executed by computing device 100A, and the function of interacting with the computing cluster is executed by computing device 100B.

[0155] It should be understood that Figure 8 The functions of the computing device 100A shown can also be performed by multiple computing devices 100. Similarly, the functions of the computing device 100B can also be performed by multiple computing devices 100.

[0156] This application also provides another computing device cluster. The connection relationships between the computing devices in this computing device cluster can be similarly referred to... Figure 7 and Figure 8 The connection method of the computing device cluster. The difference is that the memory 106 of one or more computing devices 100A in the computing device cluster can store the same instructions for executing data processing methods.

[0157] In some possible implementations, the memories of one or more computing devices 100B in the computing device cluster may also each store a portion of the instructions for executing the data processing method. In other words, a combination of one or more computing devices can jointly execute the instructions for executing the data processing method.

[0158] It should be noted that the memory 106 in different computing devices 100A within the computing device cluster can store different instructions for executing some functions of the data processing device. That is, the instructions stored in the memory 106 of different computing devices 100A can implement the functions of one or more units in the data processing device.

[0159] This application also provides a computer program product containing instructions. The computer program product may be a software or program product containing instructions, capable of running on a computing device or stored on any usable medium. When the computer program product is run on at least one computing device, it causes the at least one computing device to perform a data processing method.

[0160] This application also provides a computer-readable storage medium. The computer-readable storage medium can be any available medium capable of being stored by a computing device, or a data storage device such as a data center containing one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive). The computer-readable storage medium includes instructions that instruct the computing device to perform a data processing method, or instruct the computing device to perform a data processing method.

[0161] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the protection scope of the technical solutions of the embodiments of the present invention.

Claims

1. A data processing method, characterized in that, The method includes: Obtain the target matrix; The first database is searched based on multiple feature parameters of the target matrix to determine at least one target reference feature vector. The first database stores multiple sets of first reference data, each set of first reference data corresponds to a first reference matrix, and each set of first reference data includes multiple feature parameters and feature vectors of the first reference matrix. The multiple target reference feature vectors are feature vectors of the target first reference matrix, and the multiple feature parameters of the target first reference matrix match the multiple feature parameters of the target matrix. Based on the plurality of target reference feature vectors, an initial reference vector is determined, which is used to solve for the eigenvalues ​​and eigenvectors of the target matrix.

2. The method according to claim 1, characterized in that, The target matrix has at least one or more of the following feature parameters: The number of columns of the target matrix, the number of non-zero elements of the target matrix, and the bandwidth of the diagonal elements of the target matrix.

3. The method according to claim 1 or 2, characterized in that, The step of searching the first database based on multiple feature parameters of the target matrix to determine at least one target reference feature vector includes: The feature parameter vector is determined based on multiple feature parameters of the target matrix; Based on the feature parameter vector, vector matching is performed from the first database to determine the at least one target reference feature vector, wherein the vector composed of the feature parameters of the target first reference matrix is ​​matched with the feature parameter vector.

4. The method according to any one of claims 1 to 3, characterized in that, The plurality of feature parameters includes at least one first-type feature parameter and at least one second-type feature parameter; The first type of feature parameters of the target matrix are the same as the first type of feature parameters of the target first reference matrix; or... The similarity between the second type of feature parameters of the target matrix and the second type of feature parameters of the target first reference matrix meets the preset conditions.

5. The method according to any one of claims 1 to 4, characterized in that, The acquisition of the target matrix includes: Obtain the original matrix; The second database is searched based on multiple feature parameters of the original matrix to determine at least one target reference feature value. The second database stores multiple sets of second reference data. Each set of second reference data corresponds to a second reference matrix, including multiple feature parameters and feature values ​​of the second reference matrix. The multiple target reference feature values ​​are feature values ​​of the target second reference matrix. The multiple feature parameters corresponding to the target first reference matrix match the multiple feature parameters of the original matrix. The spectral shift parameters are determined based on the at least one target reference eigenvalue; The target matrix is ​​obtained by performing spectral shift on the original matrix based on the spectral shift parameter.

6. The method according to claim 5, characterized in that, Before searching the second database, the method further includes: Obtain configuration information, which indicates the requirements for the spectral shift parameters; Determining the spectral shift parameter based on the at least one target reference eigenvalue includes: Based on the configuration information, a target reference feature value is selected from the at least one target reference feature value as the spectral shift parameter.

7. The method according to claim 5 or 6, characterized in that, After obtaining the original matrix, the method further includes: The target solution algorithm is determined based on multiple feature parameters of the original matrix through an algorithm recommendation model. The target solution algorithm is used to solve the eigenvalues ​​and eigenvectors of the target matrix. The configuration information is the configuration information corresponding to the target solution algorithm.

8. A data processing apparatus, characterized in that, The device includes: The acquisition unit is used to acquire the target matrix; The search unit is used to search a first database based on multiple feature parameters of the target matrix to determine at least one target reference feature vector. The first database stores multiple sets of first reference data, each set of first reference data corresponds to a first reference matrix, and each set of first reference data includes multiple feature parameters and feature vectors of the first reference matrix. The multiple target reference feature vectors are feature vectors of the target first reference matrix, and the multiple feature parameters of the target first reference matrix match the multiple feature parameters of the target matrix. The vector determination unit is used to determine an initial reference vector based on the plurality of target reference feature vectors, wherein the initial reference vector is used to solve for the eigenvalues ​​and eigenvectors of the target matrix.

9. The apparatus according to claim 8, characterized in that, The target matrix has multiple characteristic parameters, including at least one or more of the following: the number of columns of the target matrix, the number of non-zero elements of the target matrix, and the bandwidth of the diagonal elements of the target matrix.

10. The apparatus according to claim 8 or 9, characterized in that, The search unit is specifically used to determine a feature parameter vector based on multiple feature parameters of the target matrix; perform vector matching from the first database based on the feature parameter vector to determine at least one target reference feature vector, wherein the vector composed of the feature parameters of the target first reference matrix matches the feature parameter vector.

11. The apparatus according to any one of claims 8 to 10, characterized in that, The plurality of feature parameters include at least one first type of feature parameter and at least one second type of feature parameter; the first type of feature parameter of the target matrix is ​​the same as the first type of feature parameter of the target first reference matrix; or, the similarity between the second type of feature parameter of the target matrix and the second type of feature parameter of the target first reference matrix satisfies a preset condition.

12. The apparatus according to any one of claims 8 to 11, characterized in that, The device also includes a spectral shift unit; The acquisition unit is specifically used to acquire the original matrix; The search unit is further configured to search a second database based on multiple feature parameters of the original matrix to determine at least one target reference feature value. The second database stores multiple sets of second reference data, each set of second reference data corresponding to a second reference matrix, including multiple feature parameters and feature values ​​of the second reference matrix. The multiple target reference feature values ​​are feature values ​​of the target second reference matrix, and the multiple feature parameters corresponding to the target first reference matrix match the multiple feature parameters of the original matrix. The spectral shift unit is used to determine the spectral shift parameters based on the at least one target reference feature value; The target matrix is ​​obtained by performing spectral shift on the original matrix based on the spectral shift parameter.

13. The apparatus according to claim 12, characterized in that, The acquisition unit is also used to acquire configuration information, which indicates the requirements for the spectral shift parameters; The spectral shift unit is further configured to select a target reference feature value from the at least one target reference feature value as the spectral shift parameter according to the configuration information.

14. The apparatus according to claim 12 or 13, characterized in that, The device also includes an algorithm recommendation unit; The algorithm recommendation unit is used to determine a target solution algorithm based on multiple feature parameters of the original matrix through an algorithm recommendation model. The target solution algorithm is used to solve for the eigenvalues ​​and eigenvectors of the target matrix. The configuration information is the configuration information corresponding to the target solution algorithm.

15. A computing device, characterized in that, The computing device includes a processor and memory; The processor is configured to execute instructions stored in the memory to cause the computing device to perform the operational steps of the method as described in any one of claims 1 to 7.

16. A computing device cluster, characterized in that, The computing device cluster includes at least one computing device, each computing device including a processor and memory: The memory is used to store instructions; The processor is configured to, according to the instructions, cause the computing device cluster to perform the operational steps of the method according to any one of claims 1 to 7.

17. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed on a computing device, cause the computing device to perform the operational steps of the method as described in any one of claims 1 to 7.

18. A computer program product comprising instructions that, when run on a computing device, cause the computing device to perform the operational steps of the method as described in any one of claims 1 to 7.

Images