Electronic device and method for generating a reference configuration of a computing device

Abstract

An electronic device and method for generating a reference configuration for a computing device are provided. The method includes: obtaining a first neural network model, wherein the first neural network model includes multiple cluster centers, wherein the multiple cluster centers correspond to multiple features; obtaining a first configuration requirement; determining whether the first configuration requirement corresponds to a first cluster center among the multiple cluster centers; generating a reference configuration based on multiple first eigenvalues ​​of the first cluster center, wherein the multiple first eigenvalues ​​respectively correspond to the multiple features; and outputting the reference configuration.

Description

Technical Field

[0001] The present invention relates to an electronic device and a method for generating a reference configuration of a computing device. Background Art

[0002] To fairly evaluate computing devices with different configurations (e.g., hardware, firmware, or software), manufacturers typically use testing software to calculate device scores for reference. For example, manufacturers can use software such as SPECpower_ssj2008 to calculate the device's score (e.g., the SP value output by SPECpower_ssj2008). To test computing devices using testing software, manufacturers must complete processes such as preparing materials, assembling the devices, and setting up a testing environment (e.g., maintaining a constant temperature) based on customer requirements. This process is time-consuming. As orders for computing devices increase, manufacturers often cannot quickly test all devices, resulting in delayed product shipments.

[0003] Furthermore, different customers care about different features. Because testing software primarily evaluates a device's computing power, it's difficult for manufacturers to use the software's test results to assess a device's performance on specific features. For example, if a customer prioritizes a device's storage capacity over its computing power, it would be difficult for manufacturers to identify a device with the best storage capacity based on the software's test results. Summary of the Invention

[0004] Embodiments of the present invention provide an electronic device and method for generating a reference configuration of a computing device, which can estimate a score of the device configuration.

[0005] An electronic device for generating a reference configuration of a computing device according to an embodiment of the present invention includes a processor and a storage medium. The storage medium stores a plurality of modules and a first neural network model, wherein the first neural network model includes a plurality of cluster centers, wherein the plurality of cluster centers correspond to a plurality of features. The processor is coupled to the storage medium, and accesses and executes a plurality of modules, wherein the plurality of modules include a data collection module, a computing module, and an output module. The data collection module obtains a first configuration requirement. The computing module is configured to execute: determining whether the first configuration requirement corresponds to a first cluster center among a plurality of cluster centers; and generating a reference configuration based on a plurality of first feature values ​​of the first cluster center, wherein the plurality of first feature values ​​respectively correspond to a plurality of features. The output module outputs the reference configuration.

[0006] In one embodiment of the present invention, the data collection module obtains a plurality of labeled data, wherein each of the plurality of labeled data includes a label score and a plurality of label feature values ​​corresponding to a plurality of features, and the plurality of modules further include a training module. The training module generates a first neural network model based on the plurality of labeled data.

[0007] In one embodiment of the present invention, the above-mentioned first configuration requirement includes a first orientation, wherein each of the multiple label data further includes a label orientation, wherein the operation module counts at least one label orientation of at least one label data among the multiple label data corresponding to the first cluster center to determine the first application orientation corresponding to the first cluster center, wherein the operation module determines that the first configuration requirement corresponds to the first cluster center in response to the first orientation matching the first application orientation.

[0008] In one embodiment of the present invention, the above-mentioned multiple cluster centers further include a second cluster center, wherein in response to both the first cluster center and the second cluster center corresponding to the first application orientation, the operation module calculates a first score corresponding to the first cluster center and a second score corresponding to the second cluster center based on multiple label data, wherein the operation module determines that the first configuration requirement corresponds to the first cluster center in response to the first orientation matching the first application orientation and the first score being greater than the second score.

[0009] In one embodiment of the present invention, the above-mentioned multiple label data reports contain first training data, wherein the multiple cluster centers further include a second cluster center and a third cluster center, wherein the first neural network model includes a first hidden layer, wherein the first hidden layer updates the multiple cluster centers according to the multiple label data, including: defining the third cluster center to be associated with the second cluster center; generating multiple second eigenvalues ​​of the second cluster center and multiple third eigenvalues ​​of the third cluster center; and determining that the first training data corresponds to the second cluster center, and updating the multiple second eigenvalues ​​and the multiple third eigenvalues ​​according to the first training data in response to the first training data corresponding to the second cluster center.

[0010] In one embodiment of the present invention, the above-mentioned first configuration requirement includes multiple feature values ​​corresponding to multiple features respectively, wherein the operation module calculates multiple distances between the first configuration requirement and multiple cluster centers based on the multiple feature values, and judges that the first configuration requirement corresponds to the first cluster center in response to the first distance corresponding to the first cluster center being the minimum distance among the multiple distances.

[0011] In one embodiment of the present invention, the above-mentioned multiple cluster centers further include a second cluster center, wherein the multiple distances further include a second distance corresponding to the second cluster center, wherein in response to the first distance being equal to the second distance, the operation module calculates a first score corresponding to the first cluster center and a second score corresponding to the second cluster center based on multiple label data corresponding to the first cluster center and the second cluster center respectively, wherein the operation module determines that the first configuration requirement corresponds to the first cluster center in response to the first score being greater than the second score.

[0012] In one embodiment of the present invention, the above-mentioned multiple label data reports contain multiple training data corresponding to multiple label scores respectively, wherein the first neural network model includes a second hidden layer, wherein the second hidden layer is configured to perform: generating a distance matrix corresponding to multiple cluster centers and multiple training data; generating an imaginary inverse matrix of the distance matrix; generating a weight matrix based on the imaginary inverse matrix and multiple label scores; generating a second distance matrix corresponding to multiple cluster centers and a first configuration requirement; and generating a first score corresponding to the first configuration requirement based on the second distance matrix and the weight matrix, wherein the output module outputs the first score.

[0013] In one embodiment of the present invention, the above-mentioned data collection module obtains multiple configuration requirements, wherein the multiple configuration requirements include a first configuration requirement and a second configuration requirement, wherein the operation module inputs the multiple configuration requirements into the first neural network to generate multiple scores corresponding to the multiple configuration requirements respectively, wherein the operation module selects the first configuration requirement to generate a reference configuration in response to the first score being the maximum score among the multiple scores, wherein the operation module generates a difference analysis report based on the first configuration requirement and the second configuration requirement in response to the first score being the maximum score and the second score corresponding to the second configuration requirement being the minimum score among the multiple scores, wherein the output module outputs the difference analysis report.

[0014] In one embodiment of the present invention, the above-mentioned multiple label data reports contain multiple training data and multiple test data, wherein the training module generates a first neural network model and a second neural network model based on the multiple training data, wherein the second neural network model includes multiple second cluster centers, wherein the first number of the multiple cluster centers is different from the second number of the multiple second cluster centers, wherein the operation module calculates a first loss function value of the first neural network model and a second loss function value of the second neural network model based on the multiple test data, wherein the operation module selects the first neural network model from the first neural network model and the second neural network model in response to the first loss function value being less than the second loss function value to generate a reference configuration.

[0015] A method for generating a reference configuration of a computing device according to an embodiment of the present invention includes: obtaining a first neural network model, wherein the first neural network model includes multiple cluster centers, wherein the multiple cluster centers correspond to multiple features; obtaining a first configuration requirement; determining that the first configuration requirement corresponds to a first cluster center among the multiple cluster centers; generating a reference configuration based on multiple first eigenvalues ​​of the first cluster center, wherein the multiple first eigenvalues ​​respectively correspond to multiple features; and outputting the reference configuration.

[0016] In one embodiment of the present invention, the above method further includes: obtaining a plurality of label data, wherein each of the plurality of label data includes a label score and a plurality of label feature values ​​corresponding to a plurality of features respectively; and generating a first neural network model based on the plurality of label data.

[0017] In one embodiment of the present invention, the above-mentioned first configuration requirement includes a first orientation, wherein each of the multiple label data further includes a label orientation, wherein the step of determining whether the first configuration requirement corresponds to a first cluster center among the multiple cluster centers includes: counting at least one label orientation of at least one label data among the multiple label data corresponding to the first cluster center to determine a first application orientation corresponding to the first cluster center; and determining that the first configuration requirement corresponds to the first cluster center in response to the first orientation matching the first application orientation.

[0018] In one embodiment of the present invention, the above-mentioned multiple cluster centers further include a second cluster center, wherein the step of determining whether the first configuration requirement corresponds to the first cluster center among the multiple cluster centers includes: in response to both the first cluster center and the second cluster center corresponding to the first application orientation, calculating a first score corresponding to the first cluster center and a second score corresponding to the second cluster center based on multiple label data; and in response to the first orientation matching the first application orientation and the first score being greater than the second score, determining that the first configuration requirement corresponds to the first cluster center.

[0019] In one embodiment of the present invention, the above-mentioned multiple label data reports contain first training data, wherein the multiple cluster centers further include a second cluster center and a third cluster center, wherein the first neural network model includes a first hidden layer, wherein the first hidden layer updates the multiple cluster centers according to the multiple label data, including: defining the third cluster center to be associated with the second cluster center; generating multiple second eigenvalues ​​of the second cluster center and multiple third eigenvalues ​​of the third cluster center; and determining that the first training data corresponds to the second cluster center, and updating the multiple second eigenvalues ​​and the multiple third eigenvalues ​​according to the first training data in response to the first training data corresponding to the second cluster center.

[0020] In one embodiment of the present invention, the above-mentioned first configuration requirement includes a plurality of feature values ​​corresponding to a plurality of features respectively, wherein the step of determining whether the first configuration requirement corresponds to a first cluster center among a plurality of cluster centers includes: calculating a plurality of distances between the first configuration requirement and the plurality of cluster centers based on the plurality of feature values; and determining that the first configuration requirement corresponds to the first cluster center in response to the first distance corresponding to the first cluster center being the minimum distance among the plurality of distances.

[0021] In one embodiment of the present invention, the above-mentioned multiple cluster centers further include a second cluster center, wherein the multiple distances further include a second distance corresponding to the second cluster center, wherein the step of determining whether the first configuration requirement corresponds to the first cluster center among the multiple cluster centers further includes: in response to the first distance being equal to the second distance, calculating a first score corresponding to the first cluster center and a second score corresponding to the second cluster center based on multiple label data corresponding to the first cluster center and the second cluster center respectively; and in response to the first score being greater than the second score, determining that the first configuration requirement corresponds to the first cluster center.

[0022] In one embodiment of the present invention, the above-mentioned multiple label data reports contain multiple training data corresponding to multiple label scores respectively, wherein the first neural network model includes a second hidden layer, wherein the second hidden layer is configured to perform: generating a distance matrix corresponding to multiple cluster centers and multiple training data; generating an imaginary inverse matrix of the distance matrix; generating a weight matrix based on the imaginary inverse matrix and multiple label scores; generating a second distance matrix corresponding to multiple cluster centers and the first configuration requirement; and generating a first score corresponding to the first configuration requirement based on the second distance matrix and the weight matrix, wherein the method further includes: outputting the first score.

[0023] In one embodiment of the present invention, the above-mentioned step of generating a reference configuration based on multiple first eigenvalues ​​of the first cluster center includes: obtaining multiple configuration requirements, wherein the multiple configuration requirements include a first configuration requirement and a second configuration requirement; inputting the multiple configuration requirements into a first neural network to generate multiple scores corresponding to the multiple configuration requirements respectively; selecting the first configuration requirement to generate a reference configuration in response to the first score being the maximum score among the multiple scores; and generating a difference analysis report based on the first configuration requirement and the second configuration requirement in response to the first score being the maximum score and the second score corresponding to the second configuration requirement being the minimum score among the multiple scores, and outputting the difference analysis report.

[0024] In one embodiment of the present invention, the above-mentioned multiple label data reports contain multiple training data and multiple test data, wherein the step of generating a reference configuration based on multiple first eigenvalues ​​of the first cluster center includes: generating a first neural network model and a second neural network model based on the multiple training data, wherein the second neural network model includes multiple second cluster centers, wherein the first number of the multiple cluster centers is different from the second number of the multiple second cluster centers; calculating a first loss function value of the first neural network model and a second loss function value of the second neural network model based on the multiple test data; and selecting the first neural network model from the first neural network model and the second neural network model to generate a reference configuration in response to the first loss function value being less than the second loss function value.

[0025] Based on the above, embodiments of the present invention can train a neural network model based on historical data to estimate device configuration scores. Compared to traditional testing software, the neural network model of embodiments of the present invention can calculate device configuration scores in a very short time. When customer needs change, embodiments of the present invention can estimate the scores of updated computing devices without re-executing the testing software. This can save a lot of manpower or material resources. On the other hand, the present invention can generate reference configurations of computing devices that focus on specific functions based on user needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 A schematic diagram of an electronic device for generating a reference configuration of a computing device is shown according to an embodiment of the present invention.

[0027] Figure 2A A flowchart of generating a neural network model is shown according to an embodiment of the present invention.

[0028] Figure 2B Another embodiment of the present invention illustrates a flowchart for generating a neural network model.

[0029] Figure 3 A schematic diagram of a neural network model is shown according to an embodiment of the present invention.

[0030] Figure 4A 、 4B and 4C is a schematic diagram illustrating generating cluster centers according to an embodiment of the present invention.

[0031] Figure 5 A schematic diagram illustrating the relationship between cluster centers according to an embodiment of the present invention is shown.

[0032] Figure 6A A schematic diagram of generating a weight matrix according to an embodiment of the present invention is shown.

[0033] Figure 6BA schematic diagram illustrating generating scores corresponding to test data using a weight matrix according to an embodiment of the present invention is shown.

[0034] Figure 7 A schematic diagram illustrating the loss function value of a neural network model according to an embodiment of the present invention is shown.

[0035] Figure 8 A flowchart of a method for generating a reference configuration of a computing device is shown according to an embodiment of the present invention. DETAILED DESCRIPTION

[0036] Figure 1 According to an embodiment of the present invention, a schematic diagram of an electronic device 100 for generating a reference configuration of a computing device is shown. The electronic device 100 may include a processor 110 , a storage medium 120 , and a transceiver 130 .

[0037] The processor 110 may be, for example, a central processing unit (CPU), or other programmable general-purpose or special-purpose microcontroller unit (MCU), microprocessor, digital signal processor (DSP), programmable controller, application-specific integrated circuit (ASIC), graphics processing unit (GPU), image signal processor (ISP), image processing unit (IPU), arithmetic logic unit (ALU), complex programmable logic device (CPLD), field programmable gate array (FPGA), or other similar components or combinations thereof. The processor 110 may be coupled to the storage medium 120 and the transceiver 130 to access and execute multiple modules and various applications stored in the storage medium 120.

[0038] The storage medium 120 is, for example, any type of fixed or removable random access memory (RAM), read-only memory (ROM), flash memory, hard disk drive (HDD), solid state drive (SSD), or similar device, or a combination thereof, and is used to store multiple modules or various applications executable by the processor 110. In this embodiment, the storage medium 120 can store multiple modules including a data collection module 121, a training module 122, a calculation module 123, an output module 124, and a database 125. The functions of these modules will be described later.

[0039] The transceiver 130 transmits and receives signals wirelessly or wired. The transceiver 130 may also perform operations such as low-noise amplification, impedance matching, frequency mixing, up- or down-conversion, filtering, amplification, and the like. In another embodiment, the transceiver 130 is coupled to an input / output interface circuit (not shown). The transceiver 130 receives user input data or instructions through the input / output interface circuit and outputs the results through the input / output interface circuit.

[0040] Database 125 may pre-store one or more neural network models. For example, database 125 may pre-store neural network model 1000. In one embodiment, database 125 may further pre-store one or more neural network models different from neural network model 1000, such as neural network models 2000, 3000, 4000, and 5000. Neural network model 1000 may be used to generate a score for configuration requirements. Users may evaluate configuration requirements based on the score output by neural network model 1000. Configuration requirements may be associated with various features such as hardware configuration, firmware configuration, or software configuration. For example, configuration requirements may include information related to features such as motherboard model, CPU power, memory capacity, solid-state drive capacity, M.2 solid-state drive capacity, hard disk drive capacity, FPGA card, Hyper-Threading (HT) configuration, operating system energy configuration, or temperature, although the present invention is not limited thereto.

[0041] The neural network model 1000 may be generated by the training module 122 . Figure 2AAccording to one embodiment of the present invention, a flowchart for generating a neural network model 1000 is shown. In step S201, the data collection module 121 may obtain a plurality of label data through the transceiver 130, wherein each of the plurality of label data may include a label score and a plurality of label feature values ​​corresponding to a plurality of features. The label data may be a historical configuration requirement of a label. Table 1 is an example of a piece of label data. In one embodiment, the label data may also include a label orientation. The label orientation may indicate an application scenario of the configuration requirement corresponding to the label data. For example, the label orientation may indicate that the configuration requirement corresponds to a product oriented towards high computing performance, a product oriented towards high storage capacity, a product oriented towards edge devices of a network architecture, or a product oriented towards high energy efficiency. In one embodiment, before using the label data to train the neural network model 1000, the training module 122 may first normalize the plurality of feature values ​​of the label data.

[0042] Table 1

[0043]

[0044]

[0045] In one embodiment, after obtaining a plurality of label data, the training module 122 may perform a correlation coefficient analysis between each feature and the label score based on the plurality of label data. The training module 122 may delete features that are less correlated with the label score based on the results of the correlation coefficient analysis. Table 2 is an example of the correlation coefficient between each feature of the label data and the label score. For example, a correlation coefficient threshold may be set for screening. For example, the absolute value of the correlation coefficient is 0.7, and the absolute value of the correlation coefficient between feature 1 and feature 2 and the label score is greater than or equal to the correlation coefficient threshold. Therefore, the training module 122 may determine that the correlation between feature 1 and feature 2 and the label score is more significant. Therefore, the training module 122 may retain feature 1 and feature 2, and may delete features 3 to 10 from the label data. In another embodiment, the correlation coefficient threshold is set to the absolute value of the correlation coefficient of 0.1, and the absolute value of the correlation coefficient between features 1 to 9 is greater than or equal to the correlation coefficient threshold. The training module 122 determines that features 1 to 9 are retained and feature 10 is deleted from the label data.

[0046] Table 2

[0047]

[0048]

[0049] In step S202, the training module 122 may generate multiple neural network models based on the multiple label data. The number of the multiple neural network models may be any positive integer. In this embodiment, the multiple neural network models may include neural network models 1000, 2000, 3000, 4000, and 5000. Figure 3 According to one embodiment of the present invention, a schematic diagram of a neural network model 1000 is shown. Taking neural network model 1000 as an example, neural network model 1000 may include an input layer 1100, a hidden layer 1200, and an output layer 1300. Hidden layer 1200 may include a first hidden layer 1201 and a second hidden layer 1202. Input layer 1100 may be configured to receive input data (e.g., configuration requirements), and output layer 1300 may be configured to generate a score corresponding to the input data.

[0050] The training method of the neural network model 1000 can be referred to Figures 4A to 6A Specifically, the neural network model 1000 may include multiple cluster centers, and each cluster center may correspond to multiple features. The number of cluster centers may be preset. For example, the neural network model 1000 may include 9 cluster centers, such as Figure 4A shown. Figure 4A 、 4B and 4C is a schematic diagram illustrating generating cluster centers according to an embodiment of the present invention. Figure 4A Nine cluster centers of the neural network model 1000 are depicted, namely cluster centers P1, P2, P3, P4, P5, P6, P7, P8, and P9. A cluster center may correspond to multiple features. In this embodiment, a cluster center may correspond to feature x1 and feature x2, where feature x1 is, for example, feature 1 (i.e., motherboard model) as shown in Table 2, and feature x2 is, for example, feature 2 (i.e., central processing unit power) as shown in Table 2. The initial feature values ​​of each feature of the cluster center may be generated by the training module 122. For example, the training module 122 may randomly generate the initial feature values ​​of the cluster center.

[0051] The first hidden layer 1201 can be used to execute a self-organizing map (SOM) algorithm. First, the first hidden layer 1201 can define the relationship between the cluster centers of the neural network model 1000. Figure 5A schematic diagram illustrating the relationship between cluster centers is shown according to one embodiment of the present invention. In this embodiment, the first hidden layer 1201 may distribute nine cluster centers in a nine-square grid and define adjacent cluster centers as associated cluster centers. For example, the first hidden layer 1201 may define cluster center P4 as associated with cluster centers P7, P5, and P1. The first hidden layer 1201 may define cluster center P3 as associated with cluster centers P2 and P6. The first hidden layer 1201 may define cluster center P5 as associated with cluster centers P2, P4, P6, and P8.

[0052] Next, the training module 122 may update the characteristic values ​​of each cluster center according to the plurality of label data. The training module 122 may divide the plurality of label data into a training data set, a test data set, and a validation data set. The first hidden layer 1201 may update the characteristic values ​​of each cluster center according to the training data in the training data set. Specifically, the first hidden layer 1201 may determine that the training data corresponds to a specific cluster center according to the characteristic values ​​of the training data, and update the characteristic values ​​of the specific cluster center and the characteristic values ​​of the cluster centers associated with the specific cluster center according to the training data. Figure 4A As shown, assuming that the training module 122 inputs the training data t1 into the neural network model 1000, the first hidden layer 1201 can determine (for example, based on the Euclidean distance) that the training data t1 corresponds to the cluster center P4. Accordingly, the first hidden layer 1201 can update the feature values ​​of the cluster centers P4, P7, P5, and P1 according to the training data t1. The cluster centers updated using the training data t1 are shown in FIG. Figure 4B shown.

[0053] The first hidden layer 1201 can continuously update the characteristic value of the cluster center according to the multiple training data in the training data set until the stopping condition is met. The stopping condition is, for example, related to the number of updates or the convergence error, etc., but the embodiment of the present invention is not limited thereto. The cluster center after the update can be as follows Figure 4C After the cluster center is updated, the training of the first hidden layer 1201 is completed.

[0054] During the training process of the first hidden layer 1201, the processor 110 may record the relationship between the training data and the cluster centers in the storage medium 120. For example, if the first hidden layer 1201 determines that the training data t1 corresponds to the cluster center P4, the processor 110 may record "training data t1 corresponds to cluster center P4" in the storage medium 120. In other words, after completing the update of the feature values ​​of each cluster center, the storage medium 120 may record one or more training data corresponding to the specific cluster center.

[0055] In one embodiment, the calculation module 123 may calculate the score of a specific cluster center based on the label scores of one or more training data corresponding to the specific cluster center. For example, if the storage medium 120 records three training data items, namely, training data t1, training data t2, and training data t3, corresponding to cluster center P4, the calculation module 123 may calculate the score corresponding to cluster center P4 based on the label score s1 of training data t1, the label score s2 of training data t2, and the label score s3 of training data t3. For example, the calculation module 123 may calculate the average of the label scores s1, s2, and s3 as the score corresponding to cluster center P4.

[0056] In one embodiment, the operation module 123 may count at least one label orientation of at least one training data among the plurality of training data corresponding to the first cluster center to determine the application orientation corresponding to the first cluster center. For example, if the storage medium 120 records three pieces of data, namely, training data t1, training data t2, and training data t3, corresponding to cluster center P4. The operation module 123 may determine the application orientation corresponding to cluster center P4 based on the label orientation o1 of training data t1, the label orientation o2 of training data t2, and the label orientation o3 of training data t3. For example, the operation module 123 may count the number of label orientations corresponding to the high computing efficiency orientation among the label orientations o1, o2, and o3. If the number of label orientations corresponding to the high computing efficiency orientation is the majority, the operation module 123 may determine that the application orientation of cluster center P4 is the high computing efficiency orientation.

[0057] After completing the update of the cluster centers, the second hidden layer 1202 may generate a weight matrix according to the cluster centers based on a radial basis function (RBF) algorithm. Figure 6A According to one embodiment of the present invention, a schematic diagram of generating a weight matrix W is shown. Assume that N is the number of features, M is the number of cluster centers, and K is the number of training data, where N, M, and K are positive integers. In stage 1, the second hidden layer 1202 may generate a matrix A and a matrix B, where the element A of the matrix A is i,j (1≤i≤N,1≤j≤M) represents the eigenvalue of the i-th feature of the j-th cluster center, and the element B of the matrix B i,j (1≤i≤N,1≤j≤K) represents the eigenvalue of the i-th feature of the j-th training data. The second hidden layer 1202 can calculate the distance (e.g., Euclidean distance) between the K-piece training data and the M cluster centers based on the matrix A and the matrix B, thereby generating a distance matrix D, where the element D of the distance matrix D is i,j(1≤i≤K, 1≤j≤M) represents the distance between the i-th training data and the j-th cluster center. Then, the second hidden layer 1202 can generate a pseudo inverse matrix I of the distance matrix D.

[0058] In stage 2, the second hidden layer 1202 may generate a matrix S, where the elements S i,1 (1≤i≤K) represents the label score of the i-th training data. The training module 122 can multiply the virtual inverse matrix I with the matrix S to generate a weight matrix W, where the element W of the weight matrix W is i,1 (1≤i≤M) represents the weight corresponding to the i-th cluster center.

[0059] After obtaining the weight matrix W, the training of the second hidden layer 1202 is completed. In one embodiment, the training module 122 may further determine whether the generated weight matrix W is usable. Specifically, in stage 3, the training module 122 may multiply the distance matrix D corresponding to the K training data with the weight matrix W to generate a matrix E1, where the element E1 of the matrix E1 is i,1 (1≤i≤K) represents the estimated score corresponding to the i-th training data. Training module 122 can determine whether weight matrix W is available based on matrix E1. For example, training module 122 can determine that weight matrix W is available in response to the distance (e.g., Euclidean distance) between matrix E1 and matrix S being less than a threshold. If weight matrix W is available, training module 122 can determine that training of second hidden layer 1202 is complete.

[0060] Returning to FIG. 2 , in step S202 , in addition to generating neural network model 1000 , training module 122 may also generate neural network models 2000 , 3000 , 4000 , and 5000 in a manner similar to that of generating neural network model 1000 . The difference between different neural network models may lie in the number of cluster centers. For example, neural network model 1000 may include 9 cluster centers, and neural network model 2000 may include 1 cluster center. The number of cluster centers may be any positive integer, and embodiments of the present invention are not limited thereto. Table 3 shows an example of the number of cluster centers corresponding to neural network models 1000 , 2000 , 3000 , 4000 , and 5000 .

[0061] Table 3

[0062]

[0063] After generating multiple neural network models, in step S203, the calculation module 123 can analyze the effectiveness of the multiple neural network models. Taking neural network model 1000 as an example, the calculation module 123 can calculate the loss function value of neural network model 1000 based on the test data set or the validation data set in the multiple labeled data, and thus determine the effectiveness of neural network model 1000 based on the loss function value. The loss function value is, for example, the root mean square error (RMSE) between the score calculated by neural network model 1000 and the label score.

[0064] In order to calculate the loss function value of the neural network model 1000, the operation module 123 may input data from the validation dataset or the test dataset into the neural network model 1000. Assuming that the operation module 123 inputs the test data into the neural network model 1000, the second hidden layer 1202 of the neural network model 1000 may calculate a score corresponding to the test data using the weight matrix W. Figure 6B A schematic diagram illustrating using a weight matrix W to generate scores corresponding to test data according to an embodiment of the present invention is shown.

[0065] In stage 1, the second hidden layer 1202 may calculate the distance between the matrix A and the matrix T1 to generate a distance matrix D2, where the element A of the matrix A is i,j (1≤i≤N,1≤j≤M) represents the eigenvalue of the i-th feature of the j-th cluster center, and the element T1 of the matrix T1 i,1 (1≤i≤N) represents the eigenvalue of the i-th feature of the test data, and the element D2 of the matrix D2 1,j (1≤j≤M) represents the distance between the test data and the jth cluster center. After obtaining the distance matrix D2, in stage 2, the second hidden layer 1202 can multiply the distance matrix D2 by the weight matrix W to generate a score E2. The operation module 123 can calculate the loss function value of the neural network model 1000 based on the score E2. The loss function value is, for example, the label score of the test data and the root mean square error of the score E2. As shown in Table 3, the operation module 123 can calculate the loss function value of the neural network model 1000 as 120. Based on a similar method, the operation module 123 can calculate the loss function values ​​of the neural network models 2000, 3000, 4000 and 5000, as shown in Table 3.

[0066] 2 , in step S204 , the operation module 123 may select the best neural network model from the multiple neural network models. For example, the operation module 123 may select the best neural network model based on the loss function value. Figure 7 A schematic diagram illustrating the loss function value of a neural network model according to an embodiment of the present invention is shown. Figure 7As shown in Table 3, the neural network model with 9 cluster centers (i.e., neural network model 1000) has the lowest loss function value compared to the neural network models with 1, 4, 16, and 25 cluster centers (i.e., neural network models 2000, 3000, 4000, and 5000). Therefore, the operation module 123 can select neural network model 1000 as the optimal neural network model.

[0067] Figure 2B According to another embodiment of the present invention, a flowchart for generating a neural network model 1000 is shown. In step S21, the data collection module 121 may obtain a plurality of labeled data via the transceiver 130. Each of the plurality of labeled data may include a label score and a plurality of label feature values ​​corresponding to a plurality of features. In step S22, the training module 122 may generate the neural network model 1000 based on the plurality of labeled data. In step S23, the calculation module 123 may analyze the performance of the neural network model 1000.

[0068] In step S24, the operation module 123 may determine whether the performance of the neural network model 1000 is acceptable. For example, the operation module 123 may determine that the performance of the neural network model 1000 is acceptable in response to the loss function value of the neural network model 1000 being less than a threshold, and may determine that the performance of the neural network model 1000 is unacceptable in response to the loss function value being greater than or equal to the threshold. If the performance of the neural network model 1000 is acceptable, the process proceeds to step S26. If the performance of the neural network model 1000 is unacceptable, the process proceeds to step S25.

[0069] In step S25, the training module 122 may change the architecture of the neural network model 1000 and return to step S22 to retrain the neural network model 1000. Specifically, the training module 122 may change the number of cluster centers of the neural network model 1000. For example, the training module 122 may change the number of cluster centers of the neural network model 1000 from 1 to 4. In step S26, the operation module 123 may determine that the training of the neural network model 1000 has been completed.

[0070] In one embodiment, the electronic device 100 may use the trained neural network model 1000 to calculate the score of the configuration requirement. For example, the data collection module 121 may obtain the configuration requirements corresponding to multiple features (e.g., features shown in Table 2) through the transceiver 130. The operation module 123 may input the configuration requirements into the neural network model 1000, and the neural network model 1000 may output the score corresponding to the configuration requirements. Specifically, the second hidden layer 1202 of the neural network model 1000 may be based on the following example: Figure 6BThe illustrated method calculates a second distance matrix D2 corresponding to multiple cluster centers and configuration requirements, and calculates a score for the configuration requirement based on the second distance matrix D2 and the weight matrix W. The output module 124 can output the score for user reference via the transceiver 130 or a coupled display device (not shown). The user can evaluate the configuration requirement based on the score.

[0071] In one embodiment, the data collection module 121 may obtain multiple configuration requirements, including a first configuration requirement, via the transceiver 130 and input the multiple configuration requirements into the neural network model 1000 to generate multiple scores corresponding to the multiple configuration requirements. The calculation module 123 may determine that the performance of the first configuration requirement is superior to the performance of the other configuration requirements in response to the first score of the first configuration requirement being the highest score among the multiple scores. For example, if the first score of the first configuration requirement is higher than the second score of the second configuration requirement, the user may determine that the performance of the first configuration requirement is superior to the performance of the second configuration requirement. In one embodiment, the calculation module 123 may determine that the performance of the second configuration requirement is inferior to the performance of the other configuration requirements in response to the second score of the second configuration requirement being the lowest score among the multiple scores. The calculation module 123 may perform a difference analysis based on the first configuration requirement corresponding to the highest score and the second configuration requirement corresponding to the lowest score to generate a difference analysis report. The output module 124 may output the difference analysis report via the transceiver 130 for user reference. In this way, users can determine which features (eg, motherboard model or CPU power) can significantly affect performance based on the variance analysis report.

[0072] The electronic device 100 can use the trained neural network model 1000 to generate a reference configuration. If a user wants to design a computing device with a specific function, the user can collect historical data related to the specific function and collect relevant configuration requirements through the historical data. The electronic device 100 can generate a reference configuration of the computing device based on the configuration requirements related to the historical data. For example, if a user wants to design a computing device with high computing power, the electronic device 100 can use configuration requirements corresponding to high computing performance orientation to generate a reference configuration. Similarly, if a user wants to design a computing device with high storage capacity, the electronic device 100 can use configuration requirements corresponding to high storage capacity orientation to generate a reference configuration.

[0073] In one embodiment, the data collection module 121 of the electronic device 100 may obtain multiple configuration requirements via the transceiver 130, where the multiple configuration requirements correspond to multiple features (e.g., features shown in Table 2). The calculation module 123 may then input the multiple configuration requirements into the neural network model 1000. The neural network model 1000 may generate multiple scores corresponding to the multiple configuration requirements. The calculation module 123 may select a first configuration requirement from the multiple configuration requirements based on the multiple scores. For example, the calculation module 123 may select the first configuration requirement from the multiple configuration requirements in response to the first score of the first configuration requirement being the maximum score among the multiple scores.

[0074] In another embodiment, after the data collection module 121 obtains the first configuration requirement through the transceiver 130 , the operation module 123 may determine that the first configuration requirement corresponds to one of the plurality of cluster centers in the neural network model 1000 .

[0075] In one embodiment, the computing module 123 may determine a cluster center corresponding to the first configuration requirement based on the orientation of the first configuration requirement. Specifically, the first configuration requirement may include a first orientation. The computing module 123 may determine that the first configuration requirement corresponds to the first cluster center in response to the first orientation matching the application orientation of a first cluster center among the plurality of cluster centers. For example, if the first configuration requirement corresponds to a high computing performance orientation and the first cluster center corresponds to a high computing performance orientation, the computing module 123 may determine that the first configuration requirement corresponds to the first cluster center.

[0076] If there are multiple cluster centers corresponding to the same application orientation, and the application orientation matches the first orientation of the first configuration requirement, the operation module 123 cannot directly determine the cluster center corresponding to the first configuration requirement based on the orientation. Accordingly, the operation module 123 can select the cluster center with the largest score from the multiple cluster centers and determine that the first configuration requirement corresponds to the cluster center. For example, assume that among the multiple cluster centers, cluster center P4 and cluster center P5 correspond to a high computing performance orientation. If the first configuration requirement also corresponds to a high computing performance orientation, the operation module 123 can determine that the first configuration requirement corresponds to cluster center P4 in response to the score of cluster center P4 being greater than the score of cluster center P5.

[0077] In one embodiment, the operation module 123 may determine that the first configuration requirement corresponds to the first cluster center in response to the first configuration requirement being closest to the first cluster center among the multiple cluster centers (for example, determined based on the Euclidean distance). In one embodiment, the operation module 123 may determine the cluster center corresponding to the first configuration requirement based on the distance between the first configuration requirement and the multiple cluster centers. Specifically, the first configuration requirement may include multiple feature values ​​corresponding to multiple features, and each of the multiple cluster centers may include multiple feature values ​​corresponding to the multiple features. The operation module 123 may calculate the distance between the first configuration requirement and each of the multiple cluster centers based on the multiple feature values ​​of the first configuration requirement and the multiple feature values ​​of each of the multiple cluster centers, thereby generating multiple distances. The operation module 123 may determine that the first configuration requirement corresponds to the first cluster center in response to the first distance corresponding to the first cluster center being the minimum distance among the multiple distances.

[0078] If there are multiple cluster centers closest to the first configuration requirement, the calculation module 123 cannot directly determine the cluster center corresponding to the first configuration requirement based on distance. Therefore, the calculation module 123 may select the cluster center with the highest score from the multiple cluster centers and determine that the first configuration requirement corresponds to that cluster center. For example, assume that among the multiple cluster centers, cluster centers P4 and P5 are closest to the first configuration requirement. In response to the score of cluster center P4 being greater than the score of cluster center P5, the calculation module 123 may determine that the first configuration requirement corresponds to cluster center P4.

[0079] After determining the first cluster center corresponding to the first configuration requirement, the computation module 123 may generate a reference configuration based on multiple feature values ​​of the first cluster center, where the multiple feature values ​​correspond to multiple features. For example, based on the feature value of "memory capacity" of the first cluster center being "16 megabytes," the computation module 123 may set the feature value of the reference configuration's "memory capacity" to "16 megabytes." After generating the reference configuration, the output module 124 may output the reference configuration via the transceiver 130 for user reference.

[0080] In one embodiment, the data collection module 121 of the electronic device 100 may obtain multiple configuration requirements via the transceiver 130, where the multiple device configurations may correspond to multiple features (e.g., features shown in Table 2). The computation module 123 may then input the multiple configuration requirements into the neural network model 1000. The neural network model 1000 may generate multiple scores corresponding to the multiple configuration requirements. The computation module 123 may select multiple configuration requirements from the multiple configuration requirements based on the multiple scores. For example, the computation module 123 may select the second and third configuration requirements from the multiple configuration requirements in response to the second score of the second configuration requirement and the third score of the third configuration requirement being greater than a score threshold. The computation module 123 may then determine the cluster centers corresponding to the second and third configuration requirements. For example, the computation module 123 may determine that the second configuration requirement corresponds to the second cluster center in response to the second configuration requirement being closest to the second cluster center among the multiple cluster centers, and may determine that the third configuration requirement corresponds to the third cluster center in response to the third configuration requirement being closest to the third cluster center among the multiple cluster centers. Accordingly, the computation module 123 can generate a reference configuration based on the multiple feature values ​​of the second cluster center and the multiple feature values ​​of the third cluster center, where the multiple feature values ​​correspond to the multiple features. Each feature value of the multiple features of the cluster center is the most representative data in the cluster, so the computation module 123 can generate a reference configuration based on the multiple feature values ​​of the cluster centers.

[0081] Figure 8 According to an embodiment of the present invention, a flow chart of a method for generating a reference configuration of a computing device is shown, wherein the method may be performed as follows: Figure 1 The electronic device 100 shown is implemented. In step S801, a first neural network model is obtained, wherein the first neural network model includes multiple cluster centers, wherein the multiple cluster centers correspond to multiple features. In step S802, a first configuration requirement is obtained. In step S803, it is determined whether the first configuration requirement corresponds to a first cluster center among the multiple cluster centers. In step S804, a reference configuration is generated based on multiple first eigenvalues ​​of the first cluster center, wherein the multiple first eigenvalues ​​respectively correspond to the multiple features. In step S805, the reference configuration is output.

[0082] In summary, the embodiment of the present invention can train a neural network model for estimating the score of the device configuration based on historical data. Compared with traditional testing software, the neural network model of the embodiment of the present invention can calculate the score of the device configuration in a very short time and save a lot of manpower or material resources. The neural network model can include multiple cluster centers, and different cluster centers can represent device configurations that focus on different functions. The embodiment of the present invention can use historical data to determine the function corresponding to each cluster center, and use the cluster center to generate a reference configuration of the computing device that focuses on a specific function according to the needs of the user. In addition to being a reference for users when assembling computing devices, the reference configuration can also be used for user analysis to help users understand the impact of each component on the function of the computing device, thereby improving the user's ability to design device configurations.

[0083]

Explanation of symbols

[0084] 100: Electronic devices

[0085] 1000, 2000, 3000, 4000, 5000: neural network model

[0086] 110: Processor

[0087] 1100: Input layer

[0088] 120: Storage medium

[0089] 1200: Hidden layer

[0090] 1201: first hidden layer

[0091] 1202: Second hidden layer

[0092] 121:Data collection module

[0093] 122: Training Module

[0094] 123: Operation module

[0095] 124: Output module

[0096] 125: Database

[0097] 130: transceiver

[0098] 1300: Output layer

[0099] A, B, E1, S, T1: Matrix

[0100] D, D2: distance matrix

[0101] E2: Score

[0102] I: Virtual Inverse Matrix

[0103] P1, P2, P3, P4, P5, P6, P7, P8, P9: cluster centers

[0104] S201, S202, S203, S204, S21, S22, S23, S24, S25, S801, S802, S803, S804, S805: Steps

[0105] t1: training data

[0106] W: weight matrix

[0107] x1, x2: features.

Claims

1. An electronic device for generating a reference configuration of a computing device, comprising: A storage medium storing a plurality of modules and a first neural network model, wherein the first neural network model includes a plurality of cluster centers, wherein the plurality of cluster centers correspond to a plurality of features; as well as A processor is coupled to the storage medium and accesses and executes the plurality of modules, wherein the plurality of modules include: A data collection module obtains a first configuration requirement, wherein the data collection module obtains a plurality of label data, wherein each of the plurality of label data includes a label score and a plurality of label feature values ​​corresponding to the plurality of features; A training module, generating the first neural network model according to the plurality of label data; A computation module configured to perform: determining whether the first configuration requirement corresponds to a first cluster center among the plurality of cluster centers; and generating the reference configuration according to a plurality of first eigenvalues ​​of the first cluster center, wherein the plurality of first eigenvalues ​​respectively correspond to the plurality of features; and An output module outputs the reference configuration based on the trained first neural network model.

2. An electronic device as described in claim 1, wherein the first configuration requirement includes a first orientation, wherein each of the multiple label data also includes a label orientation, wherein the operation module counts at least one label orientation of at least one label data among the multiple label data corresponding to the first cluster center to determine a first application orientation corresponding to the first cluster center, wherein the operation module determines that the first configuration requirement corresponds to the first cluster center in response to the first orientation matching the first application orientation.

3. The electronic device as claimed in claim 2, wherein the plurality of cluster centers further comprises a second cluster center, wherein In response to both the first cluster center and the second cluster center corresponding to the first application orientation, the operation module calculates a first score corresponding to the first cluster center and a second score corresponding to the second cluster center according to the plurality of label data, wherein The operation module determines that the first configuration requirement corresponds to the first cluster center in response to the first orientation matching the first application orientation and the first score being greater than the second score.

4. The electronic device of claim 1 , wherein the plurality of label data includes first training data, wherein the plurality of cluster centers further includes a second cluster center and a third cluster center, wherein the first neural network model includes a first hidden layer, wherein the first hidden layer updates the plurality of cluster centers according to the plurality of label data, comprising: defining the third cluster center to be associated with the second cluster center; generating a plurality of second eigenvalues ​​of the second cluster center and a plurality of third eigenvalues ​​of the third cluster center; as well as It is determined that the first training data corresponds to the second cluster center, and in response to the first training data corresponding to the second cluster center, the plurality of second eigenvalues ​​and the plurality of third eigenvalues ​​are updated according to the first training data.

5. An electronic device as described in claim 1, wherein the first configuration requirement includes a plurality of feature values ​​corresponding to the plurality of features respectively, wherein the operation module calculates a plurality of distances between the first configuration requirement and the plurality of cluster centers based on the plurality of feature values, and judges that the first configuration requirement corresponds to the first cluster center in response to the first distance corresponding to the first cluster center being the minimum distance among the plurality of distances.

6. The electronic device of claim 5, wherein the plurality of cluster centers further include a second cluster center, wherein the plurality of distances further include a second distance corresponding to the second cluster center, wherein In response to the first distance being equal to the second distance, the operation module calculates a first score corresponding to the first cluster center and a second score corresponding to the second cluster center according to a plurality of label data corresponding to the first cluster center and the second cluster center respectively, wherein The operation module determines that the first configuration requirement corresponds to the first cluster center in response to the first score being greater than the second score.

7. The electronic device of claim 1 , wherein the plurality of label data comprises a plurality of training data respectively corresponding to a plurality of label scores, wherein the first neural network model comprises a second hidden layer, wherein the second hidden layer is configured to perform: generating a distance matrix corresponding to the plurality of cluster centers and the plurality of training data; generating an imaginary inverse matrix of the distance matrix; Generate a weight matrix according to the virtual inverse matrix and the multiple label scores; generating a second distance matrix corresponding to the plurality of cluster centers and the first configuration requirement; as well as A first score corresponding to the first configuration requirement is generated according to the second distance matrix and the weight matrix, wherein The output module outputs the first score.

8. The electronic device as claimed in claim 7, wherein The data collection module obtains a plurality of configuration requirements, wherein the plurality of configuration requirements include the first configuration requirement and the second configuration requirement, wherein The operation module inputs the plurality of configuration requirements into the first neural network to generate a plurality of scores corresponding to the plurality of configuration requirements, wherein The operation module selects the first configuration requirement to generate the reference configuration in response to the first score being the maximum score among the multiple scores, wherein The operation module generates a difference analysis report according to the first configuration requirement and the second configuration requirement in response to the first score being the maximum score and the second score corresponding to the second configuration requirement being the minimum score among the multiple scores, wherein The output module outputs the difference analysis report.

9. The electronic device as claimed in claim 1, wherein the plurality of label data comprises a plurality of training data and a plurality of test data, wherein The training module generates the first neural network model and the second neural network model according to the plurality of training data, wherein the second neural network model includes a plurality of second cluster centers, wherein a first number of the plurality of cluster centers is different from a second number of the plurality of second cluster centers, The operation module calculates a first loss function value of the first neural network model and a second loss function value of the second neural network model according to the plurality of test data, wherein The operation module selects the first neural network model from the first neural network model and the second neural network model to generate the reference configuration in response to the first loss function value being less than the second loss function value.

10. A method for generating a reference configuration of a computing device, comprising: Obtaining a first neural network model, wherein obtaining the first neural network model includes: obtaining a plurality of label data, wherein each of the plurality of label data includes a label score and a plurality of label feature values ​​corresponding to a plurality of features; and generating the first neural network model according to the plurality of label data, wherein the first neural network model includes a plurality of cluster centers, wherein the plurality of cluster centers correspond to the plurality of features; Get the first configuration requirement; determining that the first configuration requirement corresponds to a first cluster center among the plurality of cluster centers; generating the reference configuration according to a plurality of first eigenvalues ​​of the first cluster center, wherein the plurality of first eigenvalues ​​respectively correspond to the plurality of features; and The reference configuration is output based on the trained first neural network model.

11. The method of claim 10 , wherein the first configuration requirement includes a first orientation, wherein each of the plurality of tag data further includes a tag orientation, wherein the step of determining whether the first configuration requirement corresponds to the first cluster center among the plurality of cluster centers comprises: Counting at least one label orientation of at least one label data among the plurality of label data corresponding to the first cluster center to determine a first application orientation corresponding to the first cluster center; as well as In response to the first orientation matching the first application orientation, it is determined that the first configuration requirement corresponds to the first cluster center.

12. The method according to claim 11, wherein the plurality of cluster centers further include a second cluster center, and wherein the step of determining whether the first configuration requirement corresponds to the first cluster center among the plurality of cluster centers comprises: In response to both the first cluster center and the second cluster center corresponding to the first application orientation, calculating a first score corresponding to the first cluster center and a second score corresponding to the second cluster center according to the plurality of label data; as well as In response to the first orientation matching the first application orientation and the first score being greater than the second score, it is determined that the first configuration requirement corresponds to the first cluster center.

13. The method of claim 10, wherein the plurality of label data includes first training data, wherein the plurality of cluster centers further includes a second cluster center and a third cluster center, wherein the first neural network model includes a first hidden layer, wherein the first hidden layer updates the plurality of cluster centers according to the plurality of label data, comprising: defining the third cluster center to be associated with the second cluster center; generating a plurality of second eigenvalues ​​of the second cluster center and a plurality of third eigenvalues ​​of the third cluster center; as well as It is determined that the first training data corresponds to the second cluster center, and in response to the first training data corresponding to the second cluster center, the plurality of second eigenvalues ​​and the plurality of third eigenvalues ​​are updated according to the first training data.

14. The method of claim 10, wherein the first configuration requirement comprises a plurality of feature values ​​corresponding to the plurality of features, wherein the step of determining whether the first configuration requirement corresponds to the first cluster center among the plurality of cluster centers comprises: Calculating a plurality of distances between the first configuration requirement and the plurality of cluster centers according to the plurality of eigenvalues; as well as In response to a first distance corresponding to the first cluster center being a minimum distance among the plurality of distances, it is determined that the first configuration requirement corresponds to the first cluster center.

15. The method of claim 14, wherein the plurality of cluster centers further include a second cluster center, wherein the plurality of distances further include a second distance corresponding to the second cluster center, wherein the step of determining whether the first configuration requirement corresponds to the first cluster center among the plurality of cluster centers further includes: In response to the first distance being equal to the second distance, calculating a first score corresponding to the first cluster center and a second score corresponding to the second cluster center based on a plurality of label data corresponding to the first cluster center and the second cluster center respectively; and In response to the first score being greater than the second score, it is determined that the first configuration requirement corresponds to the first cluster center.

16. The method of claim 10, wherein the plurality of label data comprises a plurality of training data respectively corresponding to a plurality of label scores, wherein the first neural network model comprises a second hidden layer, wherein the second hidden layer is configured to perform: generating a distance matrix corresponding to the plurality of cluster centers and the plurality of training data; generating an imaginary inverse matrix of the distance matrix; Generate a weight matrix according to the virtual inverse matrix and the multiple label scores; generating a second distance matrix corresponding to the plurality of cluster centers and the first configuration requirement; as well as Generating a first score corresponding to the first configuration requirement according to the second distance matrix and the weight matrix, wherein the method further comprises: The first score is output.

17. The method of claim 16, wherein the step of generating the reference configuration according to the plurality of first eigenvalues ​​of the first cluster center comprises: Obtaining a plurality of configuration requirements, wherein the plurality of configuration requirements include the first configuration requirement and the second configuration requirement; inputting the plurality of configuration requirements into the first neural network to generate a plurality of scores respectively corresponding to the plurality of configuration requirements; selecting the first configuration requirement to generate the reference configuration in response to the first score being a maximum score among the plurality of scores; as well as In response to the first score being the maximum score and the second score corresponding to the second configuration requirement being the minimum score among the multiple scores, a difference analysis report is generated according to the first configuration requirement and the second configuration requirement, and the difference analysis report is output.

18. The method of claim 10, wherein the plurality of label data includes a plurality of training data and a plurality of test data, and wherein the step of generating the reference configuration according to the plurality of first eigenvalues ​​of the first cluster center comprises: generating the first neural network model and the second neural network model based on the plurality of training data, wherein the second neural network model includes a plurality of second cluster centers, wherein a first number of the plurality of cluster centers is different from a second number of the plurality of second cluster centers; Calculate a first loss function value of the first neural network model and a second loss function value of the second neural network model according to the plurality of test data; as well as In response to the first loss function value being less than the second loss function value, the first neural network model is selected from the first neural network model and the second neural network model to generate the reference configuration.

Images