A method and device for realizing interactiveness of target allocation decision model

Abstract

The present application discloses an interactive implementation method and device of a target allocation decision model, including: constructing a decision model according to a defined core target and set constraints; defining a benefit function for each participating model, the benefit function mapping all possible strategy combinations to the real number domain; analyzing the benefits of each participating model under different strategy combinations according to the benefit function, and finding the optimal strategy combination, thereby constructing a decision model; wherein the optimal strategy combination is defined as no participating model has the motivation to unilaterally change its own strategy and obtain better results alone; repeatedly executing the iterative steps until an allocation scheme that meets the needs of each participating model and achieves the core target of the system is found; and outputting the final allocation scheme. The problem that the target allocation of the prior art often lacks in-depth analysis and optimization of the interaction logic between models, resulting in possible conflicts and inconsistencies when multiple models work together, is solved.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a method and device for implementing the interactivity of a target allocation decision model. Background Art

[0002] The target allocation problem needs to consider the data and instruction transmission between multiple models. It involves how to allocate tasks or resources to multiple participants efficiently, fairly and cost-effectively, and requires the models to have a certain degree of interactivity.

[0003] In the prior art, the methods for achieving interactivity between models mainly include data exchange protocols, middleware, and distributed computing frameworks. Although the prior art has made some progress in the interactive realization of the target allocation decision model, there are still some shortcomings. First, data exchange protocols and middleware limit the communication efficiency and flexibility between models, especially when processing complex data and instructions. Second, although the distributed computing framework improves the parallel processing capability of the model, there are still challenges in data synchronization and consistency maintenance.

[0004] Therefore, the target allocation of existing technologies often lacks in-depth analysis and optimization of the interaction logic between models, resulting in possible conflicts and inconsistencies when multiple models work together. Summary of the invention

[0005] In an embodiment of the present application, by providing an interactive implementation method of a target allocation decision model, the problem that the target allocation in the prior art often lacks in-depth analysis and optimization of the interaction logic between models, which may lead to conflicts and inconsistencies when multiple models work together, is solved.

[0006] In the first aspect, an embodiment of the present application provides an interactive implementation method of a target allocation decision model, the method comprising: defining the core goal of the allocation system and setting constraints; constructing a decision model based on the defined core goal and the set constraints; the construction of the decision model based on the defined core goal and the set constraints comprises: defining a benefit function for each participating model, the benefit function mapping all possible strategy combinations to the real number domain; wherein the participating model is a model for participating target allocation; according to the benefit function, analyzing the benefits of each participating model under different strategy combinations, and finding the optimal strategy combination, thereby constructing a decision model; wherein the optimal strategy combination is defined as no participating model has the motivation to unilaterally change its own strategy and obtain better results alone; executing the iterative steps of the model solution; repeating the iterative steps until an allocation plan is found that meets the needs of each participating model and achieves the core goal of the system, or reaches a preset upper limit on the number of iterations; and outputting the allocation result after the iterative steps as the final allocation plan.

[0007] In one possible implementation, the constraints include: setting a time threshold for the system to complete the assigned task; using the number of parameters as an indicator to measure the complexity of the decision model and setting a reasonable upper limit on the number of parameters; evaluating the system's memory requirements and ensuring that they do not exceed the memory limit; and ensuring that the system can achieve the predetermined target effect when completing task assignment.

[0008] In a possible implementation, the memory requirement of the evaluation system is calculated by the following formula: M=Mmodel+Minput+Mbatch+Moptimizer+Mother; wherein Mmodel is the memory consumption of the model structure, Minput is the memory consumption of the input data, Mbatch is the memory consumption of the batch size, Moptimizer is the memory consumption of the optimizer, and Mother is the memory consumption caused by other factors; wherein other factors include data type and model parallelization.

[0009] In a possible implementation, before defining the profit function for each participating model, it also includes: representing all participating models in the form of a set so that each participating model is assigned a unique identifier; defining the strategy space and decision-making moment set of each participating model; wherein the strategy space includes all possible strategies; defining the information set of each participating model; wherein the information set includes complete information games and incomplete information games.

[0010] In a possible implementation, the profit function is defined for each participating model, and the profit function maps all possible strategy combinations to the real number domain, including: the profit function is defined as ;in, For each participating model i, the profit function is is the strategy space of different participation models, where T is the number of decision moments, R is a set of real numbers, indicating that the output of the profit function can be any real number, is the Cartesian product operation.

[0011] In a possible implementation, the method analyzes the benefits of each participating model under different strategy combinations according to the benefit function and finds the optimal strategy combination, including: listing the order in which the participating models make decisions and constructing a tree diagram according to the listed order; the root node in the tree diagram represents the initial state of all participating models, and each node represents a moment when a participating model needs to make a decision; wherein the moment when the participating model needs to make a decision is selected from the set of decision moments of the participating model; the edge on each node corresponds to an action that the participating model can choose; wherein the action that the participating model can choose is selected from the strategy space of the participating model; the nodes with the same information are grouped together to form an information set, indicating that the participating model cannot distinguish which node it is making a decision on at the nodes constituting the information set; analyzing all possible action sequences starting from each information set in the tree diagram; evaluating the benefits that each participating model can obtain when following a specific action sequence and reaching its end point; after analyzing all possible action sequences and corresponding benefits, finding an optimal strategy combination, thereby finding the optimal strategy combination; wherein the optimal strategy combination is defined as no participating model has the motivation to unilaterally change its own strategy and obtain better results alone.

[0012] In one possible implementation, the iterative steps of executing the model solution include: performing a preliminary solution to the decision model based on the tree diagram and the benefit function to obtain an allocation plan; analyzing the results obtained from the preliminary solution to check whether all constraints are met, and evaluating whether the optimal strategy combination is reached or close to it; adjusting the decision model based on the preliminary analysis results; wherein the adjustment method includes modifying the benefit function, adjusting the constraints, and modifying the strategy space; and re-solving the adjusted decision model.

[0013] In a possible implementation, it also includes: providing an interactive interface to allow users to view, analyze and adjust the allocation plan; allowing dynamic modification of constraints or goals, and re-executing the iterative steps of model solution.

[0014] In the second aspect, an embodiment of the present application provides an interactive implementation device of a target allocation decision model, which includes: a setting module, which is used to define the core goals of the allocation system and set constraints; a construction module, which is used to construct a decision model based on the defined core goals and the set constraints; the construction of the decision model based on the defined core goals and the set constraints includes: defining a benefit function for each participating model, and the benefit function maps all possible strategy combinations to the real number domain; wherein the participating model is a model for participating in target allocation; according to the benefit function, the benefits of each participating model under different strategy combinations are analyzed, and the optimal strategy combination is found to construct a decision model; wherein the optimal strategy combination is defined as no participating model has the motivation to unilaterally change its own strategy and obtain better results alone; an execution module, which is used to execute the iterative steps of the model solution; the iterative steps are repeatedly executed until an allocation plan that meets the needs of each participating model and achieves the core goals of the system is found, or the preset upper limit of the number of iterations is reached; an output module, which is used to output the allocation results after the iterative steps as the final allocation plan.

[0015] In the third aspect, an embodiment of the present application provides an interactive implementation server of a target allocation decision model, comprising a memory and a processor; the memory is used to store computer-executable instructions; the processor is used to execute the computer-executable instructions to implement the method described in the first aspect or any possible implementation method of the first aspect.

[0016] One or more technical solutions provided in the embodiments of the present application have at least the following technical effects:

[0017] The embodiment of the present application provides an interactive implementation method of a target allocation decision model. By constructing a decision model and defining a benefit function, the system can analyze the benefits under different strategy combinations to find the optimal strategy combination. This greatly improves the efficiency of the allocation process and ensures the optimization of the allocation results, so that the system can achieve the optimization of the core goals while meeting the needs of all participating models. By ensuring that no participating model has the motivation to unilaterally change its strategy to obtain better results, the fairness and stability of the allocation results are guaranteed, which helps to maintain a harmonious relationship within the system and reduce potential conflicts and disputes. Through the steps of iterative solution, the system can continuously adjust the allocation plan according to the actual situation until a distribution plan that meets the needs of each participating model and achieves the core goals of the system is found. This iterative solution method makes the system highly adaptable and flexible, and can cope with various complex and changeable situations. Due to the clear definition of the benefit function and the optimal strategy combination, the system can clearly show the benefits of each participating model under different strategy combinations, making the allocation process highly transparent and explainable, which helps to enhance the trust within the system and improve user satisfaction. This solves the problem that the target allocation of existing technologies often lacks in-depth analysis and optimization of the interaction logic between models, which may lead to conflicts and inconsistencies when multiple models work together. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments of the present application or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0019] Figure 1 A flowchart of a method for interactively implementing a target allocation decision model provided in an embodiment of the present application;

[0020] Figure 2 A specific flow chart for setting constraint conditions provided in an embodiment of the present application;

[0021] Figure 3 A specific flow chart for analyzing the benefits of each participating model under different strategy combinations and finding the optimal strategy combination according to the benefit function provided in the embodiment of the present application;

[0022] Figure 4 A specific flow chart of the iterative steps of solving the model provided in the embodiment of the present application;

[0023] Figure 5 A schematic diagram of an interactive implementation device of a target allocation decision model provided in an embodiment of the present application;

[0024] Figure 6 A schematic diagram of a server for implementing the interactivity of a target allocation decision model provided in an embodiment of the present application. DETAILED DESCRIPTION

[0025] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0026] The following describes some of the techniques involved in the embodiments of the present application to facilitate understanding, and they should be considered as merely exemplary. Therefore, it should be appreciated by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present application. Similarly, for the sake of clarity and conciseness, some descriptions of well-known functions and structures are omitted in the following description.

[0027] The present application embodiment provides an interactive implementation method of a target allocation decision model, such as Figure 1 As shown, the method includes steps S101 to S107. Among them, Figure 1 This is only an execution order shown in the embodiment of the present application, and does not represent the only execution order of an interactive implementation method of a target allocation decision model. In the case that the final result can be achieved, Figure 1 The steps shown may be performed in parallel or reversed.

[0028] S101: Define the core objectives of the allocation system and set constraints.

[0029] Specifically, when planning the allocation system, it is first necessary to clarify the core goals of the system design and the constraints that must be followed. These goals and constraints will guide the development process of the entire system and ensure that the final solution meets the actual needs of the project. The core goals of the system can be maximizing efficiency, minimizing costs, ensuring fairness, etc. This application chooses to maximize efficiency as the core goal of the allocation system for illustration, which means that the design of the system is committed to optimizing the process of resource allocation to achieve the ability to complete more tasks per unit time.

[0030] Figure 2 A specific flow chart for setting constraint conditions provided in the embodiment of the present application, such as Figure 2 As shown, it includes steps S201 to S204.

[0031] S201: Setting a time threshold for the system to complete the assigned task.

[0032] Specifically, when the system performs task allocation, it needs to ensure that decisions are made in the shortest possible time. A shorter time threshold will ensure that the system does not become a bottleneck in the entire process and improve overall efficiency.

[0033] S202: Use the number of parameters as an indicator to measure the complexity of the decision model and set a reasonable upper limit on the number of parameters.

[0034] Specifically, while pursuing maximum efficiency, the complexity of the decision model also needs to be considered. An overly complex decision model may be difficult to understand and implement, and may also consume a lot of computing resources. Therefore, the number of parameters will be used as an indicator to measure the complexity of the model to find a balance between maintaining efficiency and controlling model complexity.

[0035] S203: Evaluate the memory requirements of the system and ensure that they do not exceed the memory limit.

[0036] Furthermore, the calculation formula for evaluating the system's memory requirements is as follows:

[0037] M=Mmodel+Minput+Mbatch+Moptimizer+Mother. Mmodel is the memory consumption of the model structure, Minput is the memory consumption of the input data, Mbatch is the memory consumption of the batch size, Moptimizer is the memory consumption of the optimizer, and Mother is the memory consumption caused by other factors. Other factors include data types and model parallelization.

[0038] Specifically, the memory requirements of the system at runtime are an important consideration because it determines whether the model can run on a specific hardware and whether it will have adverse effects on other programs running in parallel. The memory consumption of the model structure mainly depends on the number and structure of the model's parameters. For example, a deep neural network with millions of parameters will consume more memory than a linear model with a small number of parameters. It also includes the structural information of the model, such as the number of layers, layer types (such as convolutional layers, fully connected layers, recurrent layers, etc.), activation functions, etc. The memory consumption of input data is directly related to the size and type of input data. For example, processing high-resolution images or long sequences of text will require more memory. During training, the model usually processes a batch of data at a time (i.e., mini-batch gradient descent). The larger the batch size, the more memory is required. Optimizers (such as SGD, Adam, RMSprop, etc.) are used to update the parameters of the model. They usually need to store some additional information, such as the first and second moments of the gradient. Therefore, the memory consumption of the optimizer is related to the optimizer type and state. Furthermore, other factors include data type and model parallelism. For example, when the data type uses double-precision floating point numbers instead of single-precision floating point numbers, memory consumption may also increase if the model uses some form of parallelism (such as data parallelism or model parallelism). Therefore, other factors need to be considered when evaluating the memory requirements of the system.

[0039] S204: Ensure that the system can achieve the predetermined target effect when completing task allocation.

[0040] Specifically, the system must ensure that it can achieve the intended target effect when completing task assignments, including meeting user needs, complying with business logic, and achieving expected benefits. Continuous testing and adjustment are required to ensure that the system can achieve the expected results in actual applications.

[0041] S102: Construct a decision model based on the defined core objectives and set constraints.

[0042] Specifically, a decision model is constructed based on the defined core objectives and set constraints, including:

[0043] 1. Use a set to represent all participating models, so that each participating model is assigned a unique identifier. Specifically, N = {1, 2, ..., n} can be used to represent the set of all participating target allocation models, where each participating model represents an entity participating in target allocation.

[0044] 2. Define the strategy space and decision moment set of each participating model. The strategy space includes all possible strategies. Specifically, for each participating model i, define a strategy space S i, which contains all possible strategies or action plans for the participating model. Strategies can be simple actions or complex decision rules, such as plans to take specific actions in specific situations. The set of decision moments can be represented by t={0,1,2,...,T}, where each element represents a specific time point, and the order in the set reflects the passage of time, where 0 usually represents the initial moment and T is the last decision moment. The set of decision moments can also be understood as a time structure, which can be discrete or continuous. When the time structure is discrete, it means that decisions are made at a series of discontinuous time points. For example, if t={0,1,2}, then the model will make decisions at three different time points (0, 1, and 2). This structure is suitable for situations where there are obvious intervals between decisions or discrete steps are required to execute. It should be noted that, unlike discrete time structures, continuous time structures allow models to make decisions at any time point.

[0045] 3. Define the information set of each participating model. The information set includes complete information games and incomplete information games. Specifically, in complete information games, the information set of each participating model contains the complete history of behaviors (or strategy choices) at all previous decision moments, which means that at each decision moment, each participating model knows all the decisions of all other participating models at all past time points. In incomplete information games, the information set of each participating model only contains part of the history, and they may not know the specific decisions of other participating models at certain time points in the past, which is usually due to information privacy, communication delays, perception limitations or other reasons.

[0046] 4. Define a profit function for each participating model, which maps all possible strategy combinations to the real number domain. The participating model is the model for participating in the target allocation. Specifically, the defined profit function is .in, For each participating model i, the profit function is is the strategy space of different participation models, where T is the number of decision moments, R is a set of real numbers, indicating that the output of the profit function can be any real number, is the Cartesian product operation.

[0047] 5. According to the profit function, analyze the profit of each participating model under different strategy combinations, and find the optimal strategy combination to build a decision model. Among them, the optimal strategy combination is defined as no participating model has the motivation to unilaterally change its own strategy to obtain better results alone.

[0048] Figure 3The specific flow chart of analyzing the benefits of each participating model under different strategy combinations and finding the optimal strategy combination according to the benefit function provided in the embodiment of the present application is as follows: Figure 3 As shown, it includes steps S301 to S306.

[0049] S301: List the order in which the participating models make decisions, and construct a tree diagram according to the listed order.

[0050] S302: The root node in the tree diagram represents the initial state of all participating models, and each node represents a moment when a participating model needs to make a decision. The moment when a participating model needs to make a decision is selected from the set of decision moments of the participating model.

[0051] S303: Each edge on a node corresponds to an action that the participating model can choose, wherein the action that the participating model can choose is selected from the strategy space of the participating model.

[0052] S304: Nodes with the same information are grouped together to form an information set, indicating that the participating model cannot distinguish which node it is making a decision on at the nodes constituting the information set.

[0053] Specifically, the participating models may not know exactly which node in the tree they are making decisions on. Therefore, nodes with the same information need to be grouped together to form an information set.

[0054] S305: Analyze all possible action sequences starting from each information set in the tree diagram.

[0055] S306: After analyzing all possible action sequences and corresponding benefits, an optimal strategy combination is found, thereby finding the optimal strategy combination. The optimal strategy combination is defined as a combination in which no participating model has the motivation to unilaterally change its strategy and obtain a better result alone.

[0056] An implementation method is listed below to illustrate the process of constructing a decision model based on the defined core objectives and set constraints. Of course, other implementation methods are also possible, which are not intended to limit the scope of protection of the present application.

[0057] Assume that there are two participating models, and label them as participating model 1 and participating model 2 respectively. N={1,2} can be used to represent the set of all participating target allocation models. The strategy space of participating model 1 is S1={A,B}, which means that participating model 1 can choose strategy A or strategy B when making decisions. The strategy space of participating model 2 is S2={X,Y}, which means that participating model 2 can choose strategy X or strategy Y when making decisions. Use t={0,1} to represent the set of decision moments, that is, the structure of time. At time point t=0, the two participating models make decisions at the same time, but they do not know each other's strategy choices at this time. At time point t=1, all participating models can see each other's strategy choices at time point 0 and have the opportunity to adjust their own strategies based on this information. Define the information set of each participating model. At time point t=0, both participating models only know the current time point, but do not know each other's strategy choices. Therefore, they make decisions under incomplete information. At time point t=1, if the game continues to this step, all strategy choices will become public knowledge.

[0058] Specifically, the payoff function defines the payoff of each participating model under each strategy combination. For participating model 1: Assuming the strategy combination is (A, X), u1(A, X) = 5; Assuming the strategy combination is (A, Y), u1(A, Y) = 1; Assuming the strategy combination is (B, X), u1(B, X) = 2; Assuming the strategy combination is (B, Y), u1(B, Y) = 4. For participating model 2: Assuming the strategy combination is (A, X), u2(A, X) = 3; Assuming the strategy combination is (A, Y), u2(A, Y) = 2; Assuming the strategy combination is (B, X), u2(B, X) = 1; Assuming the strategy combination is (B, Y), u2(B, Y) = 4.

[0059] Furthermore, the two participating models need to consider their own interests and the possible reactions of their opponents at the same time. Each participating model will try to predict the other party's strategy choice and choose its own optimal strategy accordingly.

[0060] For the above example, we find the optimal strategy combination: (A, X) is an optimal strategy combination, because when model 1 chooses A, the optimal response of model 2 is X; and when model 2 chooses X, the optimal response of model 1 is also A. (B, Y) is also an optimal strategy combination, because when model 1 chooses B, the optimal response of model 2 is Y; and when model 2 chooses Y, the optimal response of model 1 is also B. The optimal strategy combination that meets the definition is that no model has the motivation to unilaterally change its strategy and obtain better results alone.

[0061] S103: Execute iterative steps of model solution.

[0062] Figure 4 A specific flow chart of the iterative steps of the model solution provided in the embodiment of the present application, such as Figure 4 As shown, it includes steps S401 to S404.

[0063] S401: Perform a preliminary solution to the decision model based on the tree diagram and the benefit function to obtain an allocation plan.

[0064] S402: Analyze the results obtained from the preliminary solution, check whether all constraints are met, and evaluate whether the optimal strategy combination is reached or close to it.

[0065] S403: According to the preliminary analysis results, the decision model is adjusted, wherein the adjustment methods include modifying the benefit function, adjusting the constraint conditions and modifying the strategy space.

[0066] S404: Re-solve the adjusted decision model.

[0067] An implementation is listed below to illustrate the process of the iterative steps of executing the model solution. Of course, other implementations are also possible, which are not intended to limit the scope of protection of the present application.

[0068] Select a suitable solution algorithm based on the characteristics of the decision model. This application uses the branch and bound method to perform a preliminary solution to the decision model to obtain a feasible allocation plan.

[0069] Set the bounds L(S) and U(S). L(S) and U(S) represent the lower and upper bounds of the solution space subset S, respectively. The bounds provide an estimated range of the objective function value for the node currently being searched. In this way, during the search process, the potential of the current search path can be evaluated based on these bounds.

[0070] Set pruning conditions. L(S)>U(S): indicates that the lower bound of the solution space subset S has exceeded its upper bound, which means that no solution in S will be better than the currently known optimal solution. Therefore, this subset S and all its child nodes can be safely pruned because they cannot produce better solutions. L(S)>Current Best Solution: This condition is often used to dynamically update pruning conditions. If the lower bound of the subset S has exceeded the currently known best solution, then S and its child nodes can also be pruned because they also cannot produce better solutions. The purpose of the pruning condition is to reduce unnecessary search work and only retain those search paths that have the potential to produce better solutions. Through pruning, we can significantly reduce the number of nodes that need to be explored, thereby improving search efficiency.

[0071] The basic steps of the branch and bound method are: 1. Construct a solution space tree. Represent the problem as a tree diagram, where each node represents a decision state and each branch represents a decision choice. 2. Branching: Select a node for branching and generate all successor nodes. 3. Bounding: Calculate the boundary value of the new node and evaluate its potential value. 4. Pruning: If the node meets the pruning condition, prune the node and its subtree. 5. Memoization: Store the calculated boundary values ​​to avoid repeated calculations. 6. Strategy: Select an appropriate search strategy for searching. 7. Termination condition: The algorithm terminates when a solution that meets all constraints is found or the search space is fully explored. Specifically, the gap between the objective function value of the current solution and the known optimal upper and lower bounds can also be calculated. If the objective function value of the current solution is very close to the optimal lower bound, then it can be considered that the current solution is close to the optimal strategy combination.

[0072] S104: Repeat the iteration steps until an allocation scheme is found that satisfies the requirements of each participating model and achieves the core goal of the system, or the preset upper limit of the number of iterations is reached.

[0073] Specifically, in order to avoid infinite loops, the system usually sets a preset upper limit on the number of iterations. When the number of iterations reaches this upper limit, the iteration process will stop regardless of whether a satisfactory allocation solution is found.

[0074] S105: Output the allocation result after the iterative steps as the final allocation solution.

[0075] S106: Provide an interactive interface to allow the user to view, analyze and adjust the allocation plan.

[0076] Specifically, the system can also provide analysis tools or functions to help users conduct in-depth analysis of allocation plans. These tools may include data visualization, sensitivity analysis, plan comparison, etc. If the user finds that the allocation plan is unreasonable or needs further optimization, the system should allow the user to adjust the allocation plan. The adjustment methods may include modifying the resource allocation amount, adjusting the time allocation, changing the priority, etc.

[0077] S107: Allows dynamic modification of constraints or objectives and re-execution of iterative steps of model solution.

[0078] Specifically, after the user completes the modification, the system automatically updates the decision model to reflect these changes. This usually involves steps such as rebuilding the tree diagram and updating the boundary function.

[0079] The present application also provides an interactive implementation device 500 of a target allocation decision model, such as Figure 5 As shown, the device includes: a setting module 501, a construction module 502, an execution module 503 and an output module 504.

[0080] The setting module 501 is used to define the core objectives of the allocation system and set constraints.

[0081] Construction module 502 is used to construct a decision model according to the defined core objectives and set constraints; constructing a decision model according to the defined core objectives and set constraints, including: defining a benefit function for each participating model, the benefit function mapping all possible strategy combinations to the real number domain; wherein the participating model is a model assigned to the participating objectives; according to the benefit function, analyzing the benefits of each participating model under different strategy combinations, and finding the optimal strategy combination, thereby constructing a decision model; wherein the optimal strategy combination is defined as no participating model has the motivation to unilaterally change its own strategy and obtain better results alone.

[0082] The execution module 503 is used to execute the iterative steps of the model solution; the iterative steps are repeatedly executed until an allocation scheme that satisfies the requirements of each participating model and achieves the core goal of the system is found, or the preset upper limit of the number of iterations is reached.

[0083] The output module 504 is used to output the allocation result after the iterative steps as the final allocation solution.

[0084] Some modules in the apparatus described in the present application can be described in the general context of computer executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, classes, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0085] The devices or modules described in the above application embodiments can be implemented by computer chips or entities, or by products with certain functions. For the convenience of description, the above devices are described in various modules according to their functions. When implementing the embodiments of the present application, the functions of each module can be implemented in the same or multiple software and / or hardware. Of course, the module that implements a certain function can also be implemented by combining multiple sub-modules or sub-units.

[0086] The methods, devices or modules described in this application can be implemented in the form of computer-readable program codes. The controller can be implemented in any appropriate manner. For example, the controller can take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program codes (such as software or firmware) that can be executed by the (micro)processor, logic gates, switches, application-specific integrated circuits (English: Application Specific Integrated Circuit, referred to as: ASIC), programmable logic controllers and embedded microcontrollers. Examples of controllers include but are not limited to the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20 and Silicone Labs C8051F320. The memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art also know that in addition to implementing the controller in the form of pure computer-readable program codes, the controller can be implemented in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers and embedded microcontrollers by logically programming the method steps. Therefore, this controller can be considered as a hardware component, and the devices included in it for implementing various functions can also be regarded as structures within the hardware component. Or even, the means for realizing various functions may be regarded as both a software module for realizing the method and a structure within a hardware component.

[0087] like Figure 6 As shown, an embodiment of the present application also provides an interactive implementation server of a target allocation decision model, including a memory 601 and a processor 602; the memory 601 is used to store computer executable instructions; the processor 602 is used to execute computer executable instructions to implement an interactive implementation method of a target allocation decision model described above in an embodiment of the present application.

[0088] An embodiment of the present application also provides a computer-readable storage medium, which stores executable instructions. When a computer executes the executable instructions, it can implement an interactive implementation method of a target allocation decision model described above in the embodiment of the present application.

[0089] It can be seen from the description of the above implementation methods that those skilled in the art can clearly understand that the present application can be implemented by means of software plus necessary hardware. Based on this understanding, the technical solution of the present application can be essentially or partly contributed to the prior art in the form of a software product, or it can be reflected in the implementation process of data migration. The computer software product can be stored in a storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a mobile terminal, a server, or a network device, etc.) to execute the method described in the embodiment of the present application.

[0090] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. All or part of this application can be used in many general or special computer system environments or configurations.

[0091] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit the present application. Although the present application has been described in detail with reference to the aforementioned embodiments, a person of ordinary skill in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some or all of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the present application.

Claims

1. An interactive implementation method of a target allocation decision model, characterized in that: include: Define the core objectives of the allocation system and set constraints; The constraints include: setting a time threshold for the system to complete the assigned task; using the number of parameters as an indicator of the complexity of the decision model and setting a reasonable upper limit on the number of parameters; evaluating the system's memory requirements and ensuring that they do not exceed the memory limit; ensuring that the system can achieve the predetermined target effect when completing the task assignment; Construct decision models based on defined core objectives and set constraints; The decision model is constructed according to the defined core objectives and set constraints, including: A profit function is defined for each participating model, wherein the profit function maps all possible strategy combinations to a real number domain; wherein the participating model is a model for participating in target allocation; The profit function is defined for each participating model, and the profit function maps all possible strategy combinations to the real number domain, including: the profit function defined is ;in, For each participating model i, the profit function is is the strategy space of different participation models, where T is the number of decision moments, R is a set of real numbers, indicating that the output of the profit function can be any real number, is the Cartesian product operation; According to the profit function, the profit of each participating model under different strategy combinations is analyzed, and the optimal strategy combination is found to build a decision model; wherein the optimal strategy combination is defined as the one in which no participating model has the motivation to unilaterally change its strategy and obtain better results alone; The method of analyzing the benefits of each participating model under different strategy combinations according to the benefit function and finding the optimal strategy combination includes: listing the order in which the participating models make decisions and constructing a tree diagram according to the listed order; the root node in the tree diagram represents the initial state of all participating models, and each node represents the moment when a participating model needs to make a decision; wherein the moment when the participating model needs to make a decision is selected from the set of decision moments of the participating model; the edge on each node corresponds to an action that the participating model can choose; wherein the action that the participating model can choose is selected from the strategy space of the participating model; the nodes with the same information are grouped together to form an information set, indicating that the participating model cannot distinguish which node it is making a decision on at the nodes constituting the information set; analyzing all possible action sequences starting from each information set in the tree diagram; evaluating the benefits that each participating model can obtain when following a specific action sequence and reaching its end point; after analyzing all possible action sequences and corresponding benefits, finding an optimal strategy combination, thereby finding the optimal strategy combination; Perform iterative steps to solve the model; The iterative steps of solving the execution model include: preliminarily solving the decision model based on the tree diagram and the benefit function to obtain the allocation plan; analyzing the results obtained from the preliminary solution to check whether all constraints are met and evaluate whether the optimal strategy combination is reached or close to it; adjusting the decision model according to the preliminary analysis results; wherein the adjustment method includes modifying the benefit function, adjusting the constraints and modifying the strategy space; and re-solving the adjusted decision model; Repeat the iteration steps until an allocation scheme is found that meets the requirements of each participating model and achieves the core goal of the system, or the preset upper limit of the number of iterations is reached; The allocation results after the iterative steps are output as the final allocation solution.

2. The interactive implementation method of the target allocation decision model according to claim 1 is characterized in that: The memory requirement of the evaluation system is calculated as follows: M=Mmodel+Minput+Mbatch+Moptimizer+Mother; Mmodel is the memory consumption of the model structure, Minput is the memory consumption of the input data, Mbatch is the memory consumption of the batch size, Moptimizer is the memory consumption of the optimizer, and Mother is the memory consumption caused by other factors; other factors include data type and model parallelization.

3. The interactive implementation method of the target allocation decision model according to claim 1 is characterized in that: Before defining the profit function for each participating model, the following is also included: Use a set to represent all participating models, so that each participating model is assigned a unique identifier; Defining a strategy space and a set of decision moments for each participating model; wherein the strategy space includes all possible strategies; An information set for each participating model is defined; wherein the information set includes a complete information game and an incomplete information game.

4. The interactive implementation method of the target allocation decision model according to claim 1 is characterized in that: Also includes: Provide an interactive interface that allows users to view, analyze and adjust the allocation plan; Allows dynamic modification of constraints or objectives and re-execution of model solution iterations.

5. An interactive implementation device for a target allocation decision model, characterized in that: include: The setting module is used to define the core objectives of the allocation system and set constraints; The constraints include: setting a time threshold for the system to complete the assigned task; using the number of parameters as an indicator of the complexity of the decision model and setting a reasonable upper limit on the number of parameters; evaluating the system's memory requirements and ensuring that they do not exceed the memory limit; ensuring that the system can achieve the predetermined target effect when completing the task assignment; A construction module is used to construct a decision model according to the defined core objectives and the set constraints; the construction of the decision model according to the defined core objectives and the set constraints includes: defining a benefit function for each participating model, the benefit function mapping all possible strategy combinations to the real number domain; wherein the participating model is a model for the distribution of participating objectives; the definition of a benefit function for each participating model, the benefit function mapping all possible strategy combinations to the real number domain, includes: the defined benefit function is ;in, For each participating model i, the profit function is is the strategy space of different participation models, where T is the number of decision moments, R is a set of real numbers, indicating that the output of the profit function can be any real number, is a Cartesian product operation; according to the profit function, the profit of each participating model under different strategy combinations is analyzed, and the optimal strategy combination is found, so as to construct a decision model; wherein the optimal strategy combination is defined as no participating model has the motivation to unilaterally change its own strategy and obtain better results alone; according to the profit function, the profit of each participating model under different strategy combinations is analyzed, and the optimal strategy combination is found, including: listing the order in which the participating models make decisions, and constructing a tree diagram according to the listed order; the root node in the tree diagram represents the initial state of all participating models, and each node represents the moment when a participating model needs to make a decision; wherein the moment when the participating model needs to make a decision is selected from the set of decision moments of the participating model; the edge on each node corresponds to an action that the participating model can choose; wherein the action that the participating model can choose is selected from the strategy space of the participating model; nodes with the same information are grouped together to form an information set, indicating that the participating models cannot distinguish at the nodes constituting the information set. which node it is making a decision on; analyzing all possible action sequences starting from each information set in the tree diagram; evaluating the benefits that each participating model can obtain when following a specific action sequence and reaching its end point; after analyzing all possible action sequences and corresponding benefits, finding an optimal strategy combination, thereby finding the optimal strategy combination; an execution module is used to execute the iterative steps of model solution; the iterative steps of executing model solution include: performing a preliminary solution to the decision model based on the tree diagram and the benefit function to obtain an allocation plan; analyzing the results obtained from the preliminary solution to check whether all constraints are met, and evaluating whether the optimal strategy combination is reached or close to it; adjusting the decision model according to the preliminary analysis results; wherein the adjustment method includes modifying the benefit function, adjusting the constraints and modifying the strategy space; re-solving the adjusted decision model; repeating the iterative steps until an allocation plan that meets the needs of each participating model and achieves the core goal of the system is found, or the preset upper limit of the number of iterations is reached; The output module is used to output the allocation results after the iterative steps as the final allocation plan.

6. An interactive implementation server of a target allocation decision model, characterized in that: including memory and processor; The memory is used to store computer executable instructions; The processor is used to execute the computer executable instructions to implement the method according to any one of claims 1 to 4.

Images