Fixed budget sorting preferential method and system and storage medium

Abstract

The invention discloses a fixed budget sorting preferential method and system and a storage medium, and relates to the technical field of network architectures, and the method comprises the following steps: S1, dividing all candidate schemes into a plurality of groups, and carrying out the initialization setting; s2, based on a multi-round mechanism, independently executing a local selection process of a group of candidate schemes by each processor to obtain a local optimal candidate scheme; s3, each processor generates an additional observation value for the local optimal candidate scheme by using the remaining sampling budget; and then, calculating a sample mean value of the additional observation value of each local optimal candidate scheme, and selecting the optimal candidate scheme with the maximum sample mean value as the optimal candidate scheme. According to the method, the problem of large-scale fixed budget sorting preferential selection can be efficiently solved, the friendliness of user experience is ensured, and good expandability is achieved.

Description

Technical Field

[0001] The present invention relates to the field of network architecture technology, and in particular to a fixed budget sorting and optimization method, system and storage medium. Background Art

[0002] In the field of simulation and simulation, the Ranking and Selection (R&S) problem is a fundamental problem that aims to select the solution with the best average performance from a finite number of alternatives. Traditional R&S programs are mainly used to solve small-scale problems (e.g., k≤1000) and can run effectively when the number of candidates is small (e.g., k≤1000) because each alternative needs to be simulated multiple times to obtain a meaningful performance estimate during the selection process. However, when the number of candidate solutions increases, especially when the number reaches millions, the limited computing power of existing algorithms on a single processor will limit the problem from being solved in a reasonable time.

[0003] Although the emergence of parallel computing environments has made it possible to solve large-scale R&S problems, most of the existing parallel algorithms target fixed-precision problems, with few discussions on fixed-budget problems. In addition, the operations are complex and difficult for non-professional users to use.

[0004] There are currently two main types of fixed-budget R&S algorithms: the Optimal Computing Budget Allocation (OCBA) algorithm and its variants, and the Expected Improvement (EI) algorithm. The OCBA algorithm and its variants determine the allocation of the sampling budget by solving a static optimization problem. The OCBA algorithm attempts to maximize the probability of the correct choice. However, this type of algorithm usually needs to frequently update the sampling information of different alternatives during the calculation process to adjust the allocation scheme. The Expected Improvement (EI) algorithm is developed under the Bayesian framework of the EI algorithm. It minimizes the expected value of the mean difference between the selected solution and the best solution by solving a series of forward-looking optimization problems. Although these algorithms have made some progress in theory, there are still many limitations in their application in large-scale problems and parallel computing environments. In addition, the existing algorithm software for sorting and optimization is mostly professional research tools, and users need to have a deep algorithm and technical background to operate.

[0005] In addition, existing R&S algorithm software is usually designed as an experimental and research tool, and users must master the complex algorithm principles and operating details to use it effectively.

[0006] In summary, the above-mentioned prior art has obvious deficiencies in solving large-scale problems, especially in parallel computing environments, and lacks easy-to-use sorting and optimization algorithm software. The prior art has the following defects:

[0007] 1. High computational burden and difficult to parallelize: OCBA and EI algorithms frequently update sampling information or solve forward-looking optimization problems during the selection process. These steps cannot be parallelized, which greatly increases the computational burden when dealing with large-scale problems, weakening the advantages of the parallel computing environment.

[0008] 2. Inability to effectively handle the growing complexity of large-scale problems: The total sampling budget of existing algorithms needs to grow significantly with the number of alternatives to maintain a high PCS. This growth is usually on the order of klogk, which quickly becomes infeasible in large-scale problems.

[0009] 3. Complex configuration and high user threshold: Most existing sorting algorithm software is complex in design, and users must have in-depth algorithm and programming background to correctly configure and use these tools. Many software require manual parameter adjustment or additional code writing in a parallel computing environment, which increases the user's learning and operation costs.

[0010] 4. Poor scalability: Existing software usually has a fixed architecture and poor scalability. When users want to introduce new modules in large-scale problems or adapt to different computing environments, they often need to make a lot of modifications to the underlying code.

[0011] Therefore, there is an urgent need to provide a solution that is easy to parallelize, highly scalable, and can effectively handle large-scale fixed-budget sorting and optimization problems. Summary of the invention

[0012] The purpose of the present invention is to provide a fixed budget sorting and optimization method, system and storage medium, which significantly improves parallelization efficiency and reduces the demand for computing resources while maintaining efficient computing performance by optimizing algorithm structure and budget allocation strategy.

[0013] To achieve the above object, the present invention provides the following solutions:

[0014] A fixed budget sorting and optimization method comprises the following steps:

[0015] S1, divide all candidate solutions into several groups and perform initialization settings;

[0016] S2, based on the round-by-round mechanism, each processor independently performs a local selection process of a set of candidate solutions to obtain the local optimal candidate solution;

[0017] S3, each processor uses the remaining sampling budget to generate additional observations for the local optimal candidate solution; then, calculates the sample mean of the additional observations of each local optimal candidate solution, and selects the optimal candidate solution with the largest sample mean as the best candidate solution.

[0018] Furthermore, the S1 divides all candidate solutions into several groups and performs initialization settings, specifically including:

[0019] The total sampling budget N 0 +N is allocated to m processors, where N 0 =n 0 k, represents the initial allocated budget, n 0 is the budget allocated to each candidate solution in the initial allocation phase, k is the number of candidate solutions, and N is the total sampling budget for subsequent rounds;

[0020] Assign (N 0 +N) / m sampling budget;

[0021] Set the number of rounds each processor takes to The sampling budget N' for each round r The sum of all rounds does not exceed the total budget of each processor, that is

[0022] Furthermore, the S2, based on a round-by-round mechanism, each processor independently performs a local selection process of a group of candidate solutions to obtain the local optimal candidate solution, specifically including the following steps:

[0023] S201, initial selection: For each processor, all allocated candidate solutions are initially sampled, and each candidate solution obtains n 0 observations, and calculate the sample mean of these observations

[0024] S202, round selection: In each round r, the processor selects the sample mean. The candidate solutions are sorted and compared in pairs in a duel manner. The candidate solutions with lower sample means are eliminated and the winners are added to the candidate set for the next round. In each round, each candidate solution is additional observations and update the sample mean of the candidate solution;

[0025] S203, local optimal selection: after After a round of selection, each processor selects the best local candidate.

[0026] The present invention also provides a fixed-budget ranking and optimization system, which is used to execute the fixed-budget ranking and optimization method, comprising:

[0027] The sample generator is used to generate the initial sample data required by the sorting optimization algorithm according to the parameters input by the user or the default parameters;

[0028] An algorithm execution module, used for receiving sample data transmitted from a sample generator, generating initial sample data required for executing the fixed budget sorting and optimization method according to the sample data, and transmitting the initial sample data to an initialization module;

[0029] The visualization module is used to interact with the algorithm execution module to provide users with real-time feedback functions, display the progress and results of the current algorithm, and allow users to make adjustments according to their needs.

[0030] Furthermore, the algorithm execution module specifically includes:

[0031] The initialization module is used to divide all candidate solutions into several groups and perform initialization settings;

[0032] A local selection module is used to independently perform a local selection process of a set of candidate solutions on each processor based on a round-by-round mechanism to obtain the local optimal candidate solution;

[0033] The best candidate solution module is used for each processor to use the remaining sampling budget to generate additional observations for the local best candidate solution; then, calculate the sample mean of the additional observations of each local best candidate solution, and select the candidate solution with the largest sample mean as the best candidate solution.

[0034] Furthermore, the local selection module includes:

[0035] The initial selection unit is used to perform initial sampling of all candidate solutions for each processor, and each candidate solution obtains n 0 observations, and calculate the sample mean of these observations

[0036] The round selection unit is used to perform round selection: in each round r, the processor selects the round according to the sample mean The candidate solutions are sorted and compared in pairs in a duel manner. The candidate solutions with lower sample means are eliminated and the winners are added to the candidate set for the next round. In each round, each candidate solution is additional observations and update the sample mean of the candidate solution;

[0037] Local selection unit, used to After a round of selection, each processor selects the best local candidate.

[0038] The present invention also provides a computer-readable storage medium storing executable instructions, which, when executed, enable a processor to execute the fixed-budget sorting and optimization method.

[0039] The present invention also provides a computer program product, comprising a computer program, wherein the computer program is executed by a processor to implement the fixed budget sorting and optimization method.

[0040] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects: the fixed budget sorting and optimization method, system and storage medium provided by the present invention, the system parallelizes and implements the fixed budget sorting and optimization (FBKT) algorithm, allowing users to use it without understanding the complex details of the algorithm. The FBKT algorithm reduces the computing tasks in each round through a round-by-round elimination mechanism and implements the parallelization of computing tasks. In addition, the algorithm can make the total sampling budget grow only linearly with the number of alternatives (i.e., the order of magnitude of k) while maintaining a high PCS. This feature makes the software of the present invention more efficient and more adaptable in large-scale problems and parallel computing environments. The parallel implementation of the FBKT algorithm not only improves the efficiency of the algorithm in large-scale problems, but also ensures the friendliness and scalability of the user experience, and solves the main challenges of the prior art in dealing with large-scale fixed budget sorting and optimization problems, that is, it realizes more efficient processing capabilities, optimized sampling budget allocation strategies, and simplified operation procedures in a parallel computing environment. In summary, the present invention has the following advantages:

[0041] 1. Simplify user operations and enhance flexibility: Existing sorting algorithm software usually requires users to perform complex settings and algorithm configurations, which is cumbersome to operate and difficult to expand. The present invention is highly user-friendly and users can easily get started without having to master the details of the algorithm. The modular design of the software enables it to have good scalability, can be dynamically adjusted according to the computing needs of different scales, and can adapt to various parallel computing environments, providing a flexible and efficient solution.

[0042] 2. Improve parallelization efficiency: The traditional fixed-budget sorting optimization algorithm contains a large number of steps that need to be serialized during the calculation process, and cannot fully utilize the advantages of parallel computing. In order to solve this problem, the FBKT algorithm adopts a round-by-round elimination structural design. In each round of competition, only two candidate solutions are compared in a local range, and the worse solution is eliminated. This local comparison feature allows the computing tasks between rounds to be completely parallelized, thereby significantly reducing the proportion of non-parallel computing parts. This design ensures that even in a large-scale parallel computing environment, the overall operating efficiency of the algorithm remains at a high level. Moreover, the system described in the present invention has the ability to adapt to a variety of parallel computing environments, including multi-core processors and distributed computing modules. Users do not need to manually configure for different hardware architectures, and can automatically optimize parallel execution strategies to ensure efficient operation in different computing environments.

[0043] 3. Reduce the sampling budget growth rate: The total sampling budget of existing algorithms usually grows at the order of klogk as the number of candidate solutions increases. This will cause the budget to expand rapidly when dealing with large-scale problems, limiting the practicality of the algorithm. The FBKT algorithm, through its knockout structure, can reduce the number of candidate solutions after each round of competition, gradually narrowing the set of solutions that need to be processed. In this way, the FBKT algorithm achieves the effect that the total sampling budget only grows linearly with the number of candidate solutions k (that is, the growth rate of (O(k))). This optimization enables the algorithm to achieve a higher probability of correct selection (PCS) with a smaller budget in large-scale problems, greatly improving resource utilization efficiency and algorithm scalability.

[0044] 4. Enhance the efficiency of the selection structure: The core of the FBKT algorithm lies in its round-by-round elimination strategy, which allows about half of the candidate solutions to be eliminated in each round. By reducing the number of candidate solutions that need to be processed in each round, the algorithm effectively reduces the computational complexity, allowing it to quickly converge to the best solution in large-scale problems. In addition, this gradual elimination mechanism allows the algorithm to focus only on the current optimal solution in each round, improving the decision quality and accuracy of each round.

[0045] The present invention adopts an improved new sampling budget allocation strategy, which makes the growth rate of the total sampling budget only the linear growth of the number of candidate solutions while maintaining a high selection probability (PCS) and preferred probability (PGS). This strategy not only reduces the cost of computing resources, but also makes the algorithm more suitable for large-scale problems.

[0046] 5. The present invention has a high degree of modular design and good expansion capabilities, and can dynamically adjust the number of processors and resource allocation according to different computing requirements. Users can flexibly expand algorithm functions or integrate other modules without modifying the underlying code, so that it can adapt to larger-scale computing tasks or different parallel computing environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0048] Figure 1 A flow chart of the fixed budget sorting and optimization method of the present invention;

[0049] Figure 2 A schematic diagram of a fixed budget sorting and optimization method according to an embodiment of the present invention;

[0050] Figure 3 It is a structural diagram of a fixed budget sorting and optimization system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0051] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0052] The purpose of the present invention is to provide a fixed budget sorting and optimization method, system and storage medium, which significantly improves parallelization efficiency and reduces the demand for computing resources while maintaining efficient computing performance by optimizing algorithm structure and budget allocation strategy.

[0053] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0054] The present invention uses multiple processors in a parallel computing environment to first divide all alternatives into several groups, and each processor independently processes a group of alternatives to perform local optimal selection. Then, the remaining sampling budget is used to generate additional observation data of the local optimal solution, and the global optimal solution is selected by further analyzing these data. This method effectively avoids frequent synchronization operations, greatly reduces the non-parallel computing part, makes the utilization rate of the algorithm in the parallel computing environment close to 1, and gives full play to the advantages of parallel computing.

[0055] like Figure 1 As shown, a fixed budget sorting and optimization method provided by an embodiment of the present invention includes the following steps:

[0056] S1, divide all candidate solutions into several groups and perform initialization settings;

[0057] S2, based on the round-by-round mechanism, each processor independently performs a local selection process of a set of candidate solutions to obtain the local optimal candidate solution;

[0058] S3, each processor uses the remaining sampling budget to generate additional observations for the local optimal candidate solution; then, calculates the sample mean of the additional observations of each local optimal candidate solution, and selects the optimal candidate solution with the largest sample mean as the best candidate solution.

[0059] In this embodiment, the S1 divides all candidate solutions into several groups and performs initialization settings, which specifically includes:

[0060] The total sampling budget N 0 +N is allocated to m processors, where N 0 =n 0 k, represents the initial allocated budget, n 0 is the budget allocated to each candidate solution in the initial allocation phase, k is the number of candidate solutions, and N is the total sampling budget for subsequent rounds;

[0061] Assign (N 0 +N) / m sampling budget;

[0062] Set the number of rounds each processor takes to The sampling budget N' for each round r The sum of all rounds does not exceed the total budget of each processor, that is

[0063] In this embodiment, S2, based on a round-by-round mechanism, each processor independently performs a local selection process of a group of candidate solutions to obtain the local optimal candidate solution, specifically including the following steps:

[0064] S201, initial selection: For each processor, all allocated candidate solutions are initially sampled, and each candidate solution obtains n 0 observations, and calculate the sample mean of these observations

[0065] S202, round selection: In each round r, the processor selects the sample mean. The candidate solutions are sorted and compared in pairs in a duel manner. The candidate solutions with lower sample means are eliminated and the winners are added to the candidate set for the next round. In each round, each candidate solution is additional observations and update the sample mean of the candidate solution;

[0066] S203, local optimal selection: after After a round of selection, each processor selects the best local candidate.

[0067] like Figure 2 FIG. 4 is a specific embodiment of the fixed budget sorting and optimization method of the present invention, and the method includes the following steps:

[0068] S1, there are 12 candidate solutions in total, which are processed in parallel by three processors and all candidate solutions are divided into 6 groups for initial sampling;

[0069] S2, each processor compares and screens the candidate solutions based on their sample mean. For example, in processor 1, the sample mean of solution 1 is greater than that of solution 4, so solution 1 enters the next round, and solution 10 also enters the next round because its mean is greater than that of solution 7; similarly, in processor 2, solutions 5 and 11 win respectively; in processor 3, solutions 3 and 12 win. The winning solutions enter the next round and undergo further sampling and comparison;

[0070] S3, each processor samples the winning solutions in the previous round again and compares them based on the sample mean. In processor 1, solution 10 beats solution 1 and enters the next round; in processor 2, solution 11 wins, and in processor 3, solution 3 wins. The winning solutions enter the next round again;

[0071] S4, each processor outputs an optimal candidate solution, namely solution 10 for processor 1, solution 11 for processor 2, and solution 3 for processor 3. Finally, the three solutions are compared again, and solution 3 with the highest sample mean is selected as the final optimal candidate solution.

[0072] Example 2

[0073] like Figure 3 As shown, the present invention further provides a fixed budget sorting and optimization system, which is used to execute the fixed budget sorting and optimization method described in Example 1, including:

[0074] Sample Generator (SampleGen), the starting part of the system, is used to generate the initial sample data required by the sorting algorithm based on the parameters entered by the user or the default parameters; these data can be simulated data or real data from an external source. The sample generator can flexibly generate samples suitable for different algorithm requirements without the need for users to manually write complex generation rules, and the modular design makes it easily expandable and adaptable to different types of data sources.

[0075] The algorithm execution module (AlgoExec) is the core part of the system, which is used to receive the sample data transmitted from the sample generator, and generate the initial sample data required for executing the fixed budget sorting and optimization method described in Example 1 based on the sample data, and transmit it to the initialization module. In the algorithm execution module, the parallelization of the FBKT algorithm is realized, making full use of multi-core processors or distributed computing systems for efficient calculations. By grouping candidate solutions and independently performing elimination and selection operations on each processor, the algorithm execution module can automatically coordinate computing resources, reduce the proportion of non-parallel computing, and greatly improve the processing efficiency of the algorithm. The AlgoExec module also has flexible algorithm scalability. Users can add new algorithms or adjust algorithm parameters according to needs to adapt to different sorting problems.

[0076] The visualization module (UserInterface) is used to interact with the algorithm execution module, provide users with real-time feedback, display the progress and results of the current algorithm, and allow users to adjust according to their needs. The visualization module can be a user interface, which is concise and intuitive. Users do not need to master the complex details of the algorithm and can complete the setting and operation through simple operations. The user interface provides users with real-time feedback, displays the progress and results of the current algorithm, and allows users to adjust according to their needs. The modular design allows the user interface to work with different types of sorting algorithms, ensuring the scalability of the software.

[0077] In this embodiment, the algorithm execution module specifically includes:

[0078] The initialization module is used to divide all candidate solutions into several groups and perform initialization settings;

[0079] A local selection module is used to independently perform a local selection process of a set of candidate solutions on each processor based on a round-by-round mechanism to obtain the local optimal candidate solution;

[0080] The best candidate solution module is used for each processor to use the remaining sampling budget to generate additional observations for the local best candidate solution; then, calculate the sample mean of the additional observations of each local best candidate solution, and select the candidate solution with the largest sample mean as the best candidate solution.

[0081] In this embodiment, the local selection module includes:

[0082] The initial selection unit is used to perform initial sampling of all candidate solutions for each processor, and each candidate solution obtains n 0 observations, and calculate the sample mean of these observations

[0083] The round selection unit is used to perform round selection: in each round r, the processor selects the round according to the sample mean The candidate solutions are sorted and compared in pairs in a duel manner. The candidate solutions with lower sample means are eliminated and the winners are added to the candidate set for the next round. In each round, each candidate solution is additional observations and update the sample mean of the candidate solution;

[0084] Local selection unit, used to After a round of selection, each processor selects the best local candidate.

[0085] Through a large number of numerical experiments, the present invention verifies its high efficiency in large-scale fixed-budget R&S problems. Experimental results show that in a parallel computing environment, compared with the traditional R&S algorithm, the present method can maintain a higher probability of correct selection (PCS) at a lower sampling budget growth rate. Especially when the number of candidate items reaches the million level, the computational efficiency and accuracy of the method described in the present invention are significantly better than those of traditional methods. The parallel design of the system ensures that in a multi-processor environment, the processor utilization is close to maximum and the computing time is significantly shortened.

[0086] FBKTS realizes a complete sorting and optimization solution by integrating the above three core modules. The system solves the large-scale fixed budget sorting and optimization problem through automated processes and parallel processing. The system is highly user-friendly, and its modular design makes it highly scalable. Users can customize the sample generator, add new algorithms, or optimize the interface according to their needs to adapt to different scenarios. In general, the software prepared according to this system generates the required data through the sample generator, the algorithm execution module executes the parallelized FBKT algorithm, and the user interface provides a simple and easy-to-use operating environment. Finally, a complete, easy-to-use and efficient sorting and optimization system is realized, which is suitable for handling large-scale sorting and optimization problems.

[0087] Example 3

[0088] An embodiment of the present invention provides a computer-readable storage medium storing executable instructions, which, when executed, enable a processor to execute the fixed-budget sorting and optimization method described in Embodiment 1.

[0089] An embodiment of the present invention further provides a computer program product, including a computer program, wherein the computer program is executed by a processor to implement the fixed budget sorting and optimization method described in Embodiment 1.

[0090] In summary, the fixed budget knockout tournament algorithm (FBKT) provided by the present invention effectively solves the large-scale sorting and selection (R&S) problem in a parallel computing environment. By optimizing the algorithm structure and budget allocation strategy, the solution significantly improves the parallelization efficiency and reduces the demand for computing resources while maintaining efficient computing performance.

[0091] The present invention is applicable to a variety of industries and application scenarios, especially in large-scale data processing and optimization problems. Specific application areas include:

[0092] Production management: Optimize production line configuration and resource allocation.

[0093] Financial risk management: rank and prioritize risk investments and portfolio strategies.

[0094] Transportation: Optimizing logistics and transportation solutions.

[0095] Deep learning model hyperparameter optimization: Quickly select the best model hyperparameters in large-scale parameter adjustment scenarios.

[0096] Feature selection for large-scale data sets: used to select the best performing feature combinations to improve model accuracy and computational efficiency.

[0097] Through these improvements, the software of the present invention not only solves the computational bottleneck of the traditional fixed-budget R&S algorithm in large-scale problems, but also provides a user-friendly operating experience and powerful scalability, so that users from different backgrounds can easily and efficiently handle large-scale sorting and optimization problems.

[0098] The fixed-budget sorting and optimization method, system and storage medium provided by the present invention effectively solve the large-scale sorting and selection (R&S) problem in a parallel computing environment. The solution significantly improves the parallelization efficiency and reduces the demand for computing resources while maintaining efficient computing performance by optimizing the algorithm structure and budget allocation strategy.

[0099] 1. Efficient parallelism: The software of the present invention reduces the frequent synchronization operations in traditional algorithms by designing the FBKT algorithm for parallel execution, grouping candidate solutions, and independently performing local selection operations on multiple processors. This optimization significantly improves processor utilization in a parallel computing environment and avoids the problem of high synchronization costs, especially on large-scale parallel computing platforms (such as Apache Spark and Apache Hadoop). By reducing the non-parallel computing part, the algorithm can more efficiently play the advantages of parallel computing in large-scale problems.

[0100] 2. Simplify the operation process and lower the user threshold: Traditional sorting optimization algorithm software usually requires users to master complex algorithm configuration and computing environment settings. The software of the present invention greatly simplifies the operation process. Users do not need to deeply understand the details of parallel algorithms, and can automatically implement the parallel application of the FBKT algorithm through a friendly operation interface. This design lowers the user's usage threshold, allowing even users without technical background to easily handle large-scale sorting optimization problems.

[0101] 3. Shorten the computation time: By parallelizing the local selection process and avoiding global synchronization in each round, the software of the present invention significantly reduces the computation time. The local best solution selection is performed in each processor, and the remaining budget is used to perform additional sampling of the local best solution, which can reduce the dependence on global synchronization. Experimental results show that compared with the existing sorting optimization algorithm, the parallelized FBKT algorithm can identify the best alternative solution more quickly without using common random numbers (CRNs), and the computation time is greatly shortened.

[0102] 4. Highly scalable and adaptable to a variety of parallel computing environments: The software of the present invention has good scalability and can adapt to a variety of parallel computing environments, such as multi-core processors and distributed computing systems. Users can adapt to different hardware architectures without manually configuring the software. The software automatically adjusts the parallel execution strategy to achieve optimal resource allocation and computing efficiency. This design makes the algorithm highly adaptable in computing tasks of different scales and can flexibly respond to a variety of practical application scenarios.

[0103] The remaining technical features in this embodiment can be flexibly selected by those skilled in the art according to actual conditions to meet different specific practical needs. However, it is obvious to those skilled in the art that it is not necessary to adopt these specific details to implement the present invention. In other examples, in order to avoid confusing the present invention, the well-known components, structures or parts are not specifically described, which are all within the technical protection scope defined by the technical solution claimed for protection in the claims of the present invention.

[0104] Modifications and changes made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the scope of protection of the claims attached to the present invention. In the above description, a large number of specific details are set forth in order to provide a thorough understanding of the present invention. However, it is obvious to those of ordinary skill in the art that these specific details are not necessary to practice the present invention. In other examples, in order to avoid confusing the present invention, well-known technologies, such as specific construction details, operating conditions and other technical conditions, are not specifically described.

[0105] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A fixed budget sorting and optimization method, characterized in that: The following steps are involved: S1, divide all candidate solutions into several groups and perform initialization settings; S2, based on the round-by-round mechanism, each processor independently performs a local selection process of a set of candidate solutions to obtain the local optimal candidate solution; S3, each processor uses the remaining sampling budget to generate additional observations for the local optimal candidate solution; then, calculates the sample mean of the additional observations of each local optimal candidate solution, and selects the optimal candidate solution with the largest sample mean as the best candidate solution.

2. The fixed budget sorting and optimization method according to claim 1, characterized in that: S1 divides all candidate solutions into several groups and performs initialization settings, which specifically include: Allocate the total sampling budget N0+N to m processors, where N0=n0k, representing the initial allocated budget; n0 is the budget allocated to each candidate solution in the initial allocation phase; k is the number of candidate solutions; and N is the total sampling budget for subsequent rounds. Allocate a sampling budget of (N0+N) / m to each processor; Set the number of rounds each processor takes to The sampling budget N' for each round r The sum of all rounds does not exceed the total budget of each processor, that is 3. The fixed budget sorting and optimization method according to claim 1, characterized in that: The S2, based on the round-by-round mechanism, is performed by each processor independently performing a local selection process of a set of candidate solutions to obtain the local optimal candidate solution, which specifically includes the following steps: S201, initial selection: For each processor, perform initial sampling on all assigned candidate solutions, obtain n0 observations for each candidate solution, and calculate the sample mean of these observations. S202, round selection: In each round r, the processor selects the sample mean. The candidate solutions are sorted and compared in pairs in a duel manner. The candidate solutions with lower sample means are eliminated and the winners are added to the candidate set for the next round. In each round, each candidate solution is additional observations and update the sample mean of the candidate solution; S203, local optimal selection: after After a round of selection, each processor selects the best local candidate.

4. A fixed budget sorting and optimization system, used to execute the fixed budget sorting and optimization method according to any one of claims 1 to 3, characterized in that: include: The sample generator is used to generate the initial sample data required by the sorting optimization algorithm according to the parameters input by the user or the default parameters; An algorithm execution module, configured to receive sample data transmitted from a sample generator, and generate initial sample data required for executing the fixed budget sorting and optimization method according to any one of claims 1 to 3 according to the sample data, and transmit the initial sample data to an initialization module; The visualization module is used to interact with the algorithm execution module to provide users with real-time feedback functions, display the progress and results of the current algorithm, and allow users to make adjustments according to their needs.

5. The fixed budget ranking and optimization system according to claim 4, characterized in that: The algorithm execution module specifically includes: The initialization module is used to divide all candidate solutions into several groups and perform initialization settings; A local selection module is used to independently perform a local selection process of a set of candidate solutions on each processor based on a round-by-round mechanism to obtain the local optimal candidate solution; The best candidate solution module is used for each processor to use the remaining sampling budget to generate additional observations for the local best candidate solution; then, calculate the sample mean of the additional observations of each local best candidate solution, and select the candidate solution with the largest sample mean as the best candidate solution.

6. The fixed budget ranking and optimization system according to claim 4, characterized in that: The local selection module comprises: The initial selection unit is used to perform initial sampling of all assigned candidate solutions for each processor, obtain n0 observations for each candidate solution, and calculate the sample mean of these observations. The round selection unit is used to perform round selection: in each round r, the processor selects the round according to the sample mean The candidate solutions are sorted and compared in pairs in a duel manner. The candidate solutions with lower sample means are eliminated and the winners are added to the candidate set for the next round. In each round, each candidate solution is additional observations and update the sample mean of the candidate solution; Local selection unit, used to After a round of selection, each processor selects the best local candidate.

7. A computer-readable storage medium storing executable instructions, wherein when the instructions are executed, a processor executes the fixed-budget sorting and optimization method according to any one of claims 1 to 3.

8. A computer program product, comprising a computer program, wherein the computer program is executed by a processor to implement the fixed-budget sorting and optimization method according to any one of claims 1 to 3.

Images