CN115344386A - Method, device and equipment for predicting cloud simulation computing resources based on sequencing learning - Google Patents
Method, device and equipment for predicting cloud simulation computing resources based on sequencing learning Download PDFInfo
- Publication number
- CN115344386A CN115344386A CN202210997240.8A CN202210997240A CN115344386A CN 115344386 A CN115344386 A CN 115344386A CN 202210997240 A CN202210997240 A CN 202210997240A CN 115344386 A CN115344386 A CN 115344386A
- Authority
- CN
- China
- Prior art keywords
- simulation
- cloud
- ranking
- feature data
- learning model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000004088 simulation Methods 0.000 title claims abstract description 181
- 238000000034 method Methods 0.000 title claims abstract description 73
- 238000012163 sequencing technique Methods 0.000 title claims abstract description 59
- 238000000605 extraction Methods 0.000 claims abstract description 28
- 230000006870 function Effects 0.000 claims description 36
- 230000008569 process Effects 0.000 claims description 18
- 238000004590 computer program Methods 0.000 claims description 15
- 238000012545 processing Methods 0.000 claims description 11
- 238000004364 calculation method Methods 0.000 claims description 10
- 238000012549 training Methods 0.000 claims description 10
- 230000002596 correlated effect Effects 0.000 claims description 6
- 238000007781 pre-processing Methods 0.000 claims description 5
- 230000009467 reduction Effects 0.000 claims description 5
- 238000004140 cleaning Methods 0.000 claims description 3
- 238000010276 construction Methods 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 230000008859 change Effects 0.000 claims description 2
- 238000010586 diagram Methods 0.000 description 7
- 238000004422 calculation algorithm Methods 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 6
- 230000003068 static effect Effects 0.000 description 6
- 230000000875 corresponding effect Effects 0.000 description 4
- 238000011156 evaluation Methods 0.000 description 4
- 238000010801 machine learning Methods 0.000 description 4
- 238000004891 communication Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000007726 management method Methods 0.000 description 2
- 238000013468 resource allocation Methods 0.000 description 2
- 230000001360 synchronised effect Effects 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 230000001186 cumulative effect Effects 0.000 description 1
- 238000013075 data extraction Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000012804 iterative process Methods 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000007637 random forest analysis Methods 0.000 description 1
- 230000001172 regenerating effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5005—Allocation of resources, e.g. of the central processing unit [CPU] to service a request
- G06F9/5027—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5005—Allocation of resources, e.g. of the central processing unit [CPU] to service a request
- G06F9/5011—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resources being hardware resources other than CPUs, Servers and Terminals
- G06F9/5016—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resources being hardware resources other than CPUs, Servers and Terminals the resource being the memory
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
Landscapes
- Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The application relates to a method, a device and equipment for predicting cloud simulation computing resources based on sequencing learning, wherein the method comprises the following steps: acquiring a feature data set of simulation application on a cloud computing node; the data in the feature data set comprises pre-run parameters and run-time parameters of the simulation application; extracting relevant feature data which obviously influences the operation time of the simulation application from the feature data set based on an SHAP feature extraction method; inputting the related characteristic data into the constructed sequencing learning model, and predicting the priority ranking of the use amount of different cloud computing resources; and taking the cloud computing resource usage with the highest ranking output by the sequencing learning model as an optimal cloud simulation computing resource prediction result. Relevant feature data which obviously influences the operation time of the simulation application is extracted by a feature extraction method based on SHAP interpretability and used as model input, the use amount of the highest-ranking cloud computing resources is predicted by using the constructed sequencing learning model, and the resource prediction performance is greatly improved.
Description
Technical Field
The invention belongs to the technical field of cloud simulation, and relates to a method, a device and equipment for predicting cloud simulation computing resources based on sequencing learning.
Background
The complex system is a whole formed by a plurality of nodes which are mutually connected, interacted and complemented in function, the relationship among the systems is complex, the interaction is obvious, the mutual restriction is large, and the mutual influence is large. Simulation technology is one of the most important methods for studying complex systems at present. Due to the complexity, uncertainty, emerging performance of overall behaviors and the like of the internal relations of the complex system, the simulation application of the complex system is large in running calculation amount and long in running time, and therefore high requirements are put forward on computing resources and communication capacity. The traditional serial simulation or the distributed simulation on a common network cluster has the problems of relatively solidified calculation communication resource allocation, low simulation operation efficiency and the like, and the requirement of large sample operation of the simulation of a complex system on strong timeliness is difficult to meet.
The Parallel Discrete Event Simulation (PDES) technique supports high-performance parallel simulation operation by cooperating with multiple computing units, and is often used to improve the execution efficiency of complex system simulation applications. Before a complex system simulation application based on the PDES runs, the number of computing resources (usually referred to as the number of CPU cores) required by the running is generally required to be specified for the application, and a simulation task is divided into different CPU cores for parallel computing. The cloud computing technology can realize cooperative management and demand allocation of resources such as computing/storage, and the like, so that an efficient resource allocation means can be provided for the operation of the simulation application of the complex system.
In recent years, a large number of researchers have studied cloud environment-oriented simulation resource prediction methods based on classification/regression machine learning techniques, which predict the running time of simulation applications under allocation of different computing resources by collecting simulation application feature data using a classification/regression algorithm, then rank the simulation computing resources based on the prediction result (simulation running time), and finally obtain computing resources capable of minimizing the running time of the simulation applications. However, in the process of implementing the present invention, the inventor finds that the conventional simulation resource prediction method has a technical problem of low resource prediction performance.
Disclosure of Invention
In view of the problems in the conventional methods, the present invention provides a method for predicting resources in cloud-simulated computing based on rank learning, a device for predicting resources in cloud-simulated computing based on rank learning, a computer device, and a computer-readable storage medium, which can greatly improve the resource prediction performance.
In order to achieve the above purpose, the embodiment of the invention adopts the following technical scheme:
in one aspect, a cloud simulation computing resource prediction method based on ranking learning is provided, and includes the following steps:
acquiring a feature data set of simulation application on a cloud computing node; the data in the feature data set comprises pre-run parameters and run-time parameters of the simulation application;
extracting relevant feature data which obviously influences the operation time of the simulation application from the feature data set based on a SHAP feature extraction method;
inputting the related characteristic data into the constructed sequencing learning model, and predicting the priority ranking of the use amount of different cloud computing resources;
and taking the highest-ranking cloud computing resource usage amount output by the sequencing learning model as an optimal cloud simulation computing resource prediction result.
In another aspect, a cloud simulation computing resource prediction apparatus based on rank learning is also provided, including:
the data acquisition module is used for acquiring a characteristic data set of the simulation application on the cloud computing node; the data in the feature dataset comprises pre-run parameters and run-time parameters of the simulation application;
the system comprises a characteristic extraction module, a simulation application running time calculation module and a simulation application running time calculation module, wherein the characteristic extraction module is used for extracting relevant characteristic data which obviously influence the simulation application running time from a characteristic data set based on a SHAP characteristic extraction method;
the ranking prediction module is used for inputting the related characteristic data into the constructed sequencing learning model and predicting the priority ranking of the use amount of different cloud computing resources;
and the result output module is used for taking the highest-ranking cloud computing resource usage output by the sequencing learning model as an optimal cloud simulation computing resource prediction result.
In another aspect, a computer device is further provided, which includes a memory and a processor, where the memory stores a computer program, and the processor implements the steps of the above method for predicting cloud simulation computing resources based on rank learning when executing the computer program.
In still another aspect, a computer readable storage medium is provided, on which a computer program is stored, and the computer program, when executed by a processor, implements the steps of the above method for predicting cloud simulation computing resources based on rank learning.
One of the above technical solutions has the following advantages and beneficial effects:
according to the cloud simulation computing resource prediction method, device and equipment based on sequencing learning, after the parameters before operation and the parameters during operation of the simulation application on the cloud computing nodes are obtained, relevant feature data which obviously affect the operation time of the simulation application are extracted based on the SHAP feature extraction method, then the priority ranking of the usage amount of different cloud computing resources is predicted by inputting the relevant feature data into the constructed sequencing learning model, and finally the highest ranked cloud computing resource usage amount is used as the optimal cloud simulation computing resource prediction result.
Therefore, the importance of the influence factors of the simulation running time of the complex system is quantitatively evaluated by adopting a SHAP interpretable feature extraction method, namely, a factor set influencing the running efficiency of the simulation application is firstly analyzed, and static features before the running of the simulation application and dynamic features in the running of the simulation application are extracted as training sample feature data. And then, calculating the important contribution degree of each feature based on the SHAP value, extracting relevant feature data which obviously influences the operation time of the simulation application as model input, and predicting the use amount of the highest-ranking cloud computing resource by using the constructed sequencing learning model, thereby greatly improving the cloud computing resource prediction performance.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments or the conventional technologies of the present application, the drawings used in the descriptions of the embodiments or the conventional technologies will be briefly introduced below, it is obvious that the drawings in the following descriptions are only some embodiments of the present application, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
FIG. 1 is a schematic flow diagram of a method for cloud-emulated computing resource prediction based on rank learning in one embodiment;
FIG. 2 is a schematic flow chart illustrating a method for predicting cloud-emulated computing resources based on rank learning in another embodiment;
FIG. 3 is a schematic diagram of an embodiment of a related feature data extraction process;
FIG. 4 is a flow diagram that illustrates ranking prediction of cloud computing resource usage in one embodiment;
FIG. 5 is a block diagram of a flow diagram of a method for cloud-emulated computing resource prediction based on rank learning in one embodiment;
fig. 6 is a block diagram of a cloud-emulated computing resource prediction device based on rank learning in an embodiment.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.
Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein in the description of the present application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application.
It should be noted that reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the invention. The appearances of the phrase in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments.
One skilled in the art will appreciate that the embodiments described herein can be combined with other embodiments. The term "and/or" as used in this specification and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.
In research work, the inventors found that in the prediction process, a traditional resource prediction method generally constructs a prediction model based on high-precision prediction indexes, such as Root Mean Square Error (RMSE) or Mean Absolute Error (MAE), and the prediction goal of the prediction model is to minimize the difference between the predicted running time and the actual running time. However, the goal of the simulation resource prediction task is actually to predict the optimal amount of computing resources, thereby ensuring the shortest simulation application run time. Machine learning predictive models based on high precision predictors tend to result in a worse ranking. Therefore, in the simulation computing resource prediction task, compared with a method for predicting simulation running time under different computing resources and ranking again, the method for directly learning the ranking of the resource usage amount can explain the relationship between different resource usage amounts and simulation application running time more deeply.
The ranking Learning (Learning to Rank) method uses the relative ranking between two samples in a data set as feedback information, and ensures that the probability of wrong ranking between sample pairs is minimized through iterative Learning, thereby generating the optimal ranking of all samples. During the training process, the rank learning model generally considers that each wrong rank has the same error cost. However, during execution of the simulation application, the amount of computational resources that are first in the prediction ranking tends to be of greater concern and it is expected that the prediction result will minimize the simulation application runtime. Therefore, aiming at the technical problem of low resource prediction performance in the traditional resource prediction method, the application provides a cloud simulation computing resource prediction method based on sequencing learning, and by collecting static feature data and dynamic monitoring data of simulation application, the SHAP interpretable feature extraction method is used for analyzing the influence degree of each factor on the operation efficiency of the simulation application, so that the significant factors influencing the operation efficiency of the simulation application are extracted. Meanwhile, the computational resources that can minimize the runtime of the simulation application are predicted quickly and accurately by improving the cost-loss function of the design to ensure that more penalties are added when the ranking of the best resources is incorrect.
The following detailed description of the embodiments of the invention will be made with reference to the accompanying drawings.
The cloud computing technology provides an efficient resource management mode for the simulation application of the complex system. Compared with the traditional method for directly predicting the resource usage of the simulation application by adopting the classification/regression technology, the method for learning the ranking of the resource usage can deeply explain the relationship between different resource usage and the operation efficiency of the simulation application. Therefore, the cloud simulation computing resource prediction method based on the sequencing learning provided by the application has the design concept that the sequencing learning method is used for predicting the priority of the computing resources used before the simulation application runs. The method aims to minimize the number of resource pairs which are not sorted correctly and obtain the use amount of the cloud computing resources which are ranked at the top, so that the execution performance of the simulation application is improved.
Referring to fig. 1, in an embodiment, the present application provides a method for predicting cloud simulation computing resources based on rank learning, including the following processing steps S12 to S18:
s12, acquiring a feature data set of simulation application on the cloud computing node; the data in the feature dataset includes pre-runtime parameters and runtime parameters of the simulation application.
It can be appreciated that in practical applications, a resource monitor may be deployed on a cloud computing node to accurately monitor and collect real-time running information of a simulation application, such as pre-running parameters and runtime parameters of the simulation application. The pre-run parameters are determined before the simulation application executes, and may include known parameters such as the number of cloud computing nodes executing the simulation application, the number of simulation application entities, a look-ahead value, and a simulation time, for example. The runtime parameters reflect performance differences of different parameter simulation applications under specific cloud computing resources, and such parameters may be collected during execution of the simulation applications, and may include parameters such as CPU usage, memory usage, network throughput, network latency, and file system usage, for example. The collected data set of pre-run parameters and run-time parameters will be used as a feature data set for subsequent processing steps.
In some embodiments, the resource monitor may collect parameter information once every set time (e.g., 5 seconds) and store the parameter information in the deployed cloud application feature database, where specific parameters may be as shown in table 1 below.
TABLE 1
And S14, extracting relevant feature data which obviously influences the running time of the simulation application from the feature data set based on the SHAP feature extraction method.
It can be understood that the SHAP is a unified method for explaining the output of the machine learning model, and based on the SHAP feature extraction method, the influence factors of the feature dimensions of each data in the feature data set on the operation efficiency of the simulation application can be analyzed, so that the influence degree of each influence factor on the operation efficiency of the simulation application can be determined. The specific processing procedure based on the SHAP feature extraction method can be understood by referring to the same process based on the SHAP calculation method in the prior art. And finally, sorting according to the degree of influence of the feature dimensions of each datum on the running efficiency of the simulation application, and extracting feature data which obviously influence the running time of the simulation application from the feature data set, wherein the feature data are called related feature data. The number of extracted relevant feature data may be determined according to a decision threshold that significantly affects the runtime of the simulation application, which may be given by a distribution setting for different simulation applications.
And S16, inputting the related characteristic data into the constructed sequencing learning model, and predicting the priority ranking of the use amount of different cloud computing resources.
It will be appreciated that the purpose of the order learning model is to use the last run data to predict the resource priority required for the next run. The model takes the relative sequencing between every two samples in a training sample set as feedback information, minimizes the probability of wrong sequencing between sample pairs through iterative learning, and accordingly generates the optimal ranking of all samples. Thus, the present embodiment employs the built ranking learning model to predict the overall ranking of simulation computing resources and to predict the optimal number of resources needed for the shortest runtime of a simulation application. After the extracted relevant feature data are input into the sequencing learning model for iterative learning, priority ranks of the usage amounts of different cloud computing resources can be output, namely, the ranks of the usage amounts of the cloud computing resources corresponding to the sequencing of the simulation application running time from long to short (or from short to long) are output under the usage amounts of the cloud computing resources.
And S18, taking the highest-ranking cloud computing resource usage amount output by the sequencing learning model as an optimal cloud simulation computing resource prediction result.
It can be understood that after the ranking of the usage amount of each cloud computing resource is predicted, the highest-ranked usage amount of the cloud computing resources can be output as an optimal cloud simulation computing resource prediction result, and the simulation application has the shortest running time and the highest running efficiency under the optimal cloud simulation computing resource prediction result.
According to the cloud simulation computing resource prediction method based on sequencing learning, after the pre-operation parameters and the operation parameters of the simulation application on the cloud computing nodes are obtained, the relevant feature data which obviously influences the operation time of the simulation application are extracted based on the SHAP feature extraction method, the priority ranking of the usage amount of different cloud computing resources is predicted by inputting the relevant feature data into the constructed sequencing learning model, and finally the highest ranked cloud computing resource usage amount is used as the optimal cloud simulation computing resource prediction result.
Therefore, the importance of the influence factors of the simulation running time of the complex system is quantitatively evaluated by adopting a SHAP interpretable feature extraction method, namely, a factor set influencing the running efficiency of the simulation application is firstly analyzed, and static features before the running of the simulation application and dynamic features in the running of the simulation application are extracted as training sample feature data. And then, calculating the important contribution degree of each feature based on the SHAP value, extracting relevant feature data which obviously influences the operation time of the simulation application as model input, and predicting the use amount of the highest-ranked cloud computing resources by using the constructed sequencing learning model, thereby greatly improving the cloud computing resource prediction performance.
In one embodiment, as shown in fig. 2, the method further includes step S13:
s13, preprocessing the data in the characteristic data set; the preprocessing includes data cleaning, error screening, and data normalization.
It can be understood that after the required characteristic data set is acquired, in order to reduce the possibility of data deviation, common data preprocessing operations such as data cleaning, error screening and data standardization can be performed on the data first to eliminate the deviated data.
By reducing the possibility of data deviation in step S13, the prediction accuracy and prediction efficiency can be further improved.
In one embodiment, as shown in fig. 3, regarding step S14, the following processing steps S141 to S143 may be specifically included:
s141, analyzing the influence degree of each feature dimension in the feature data set on the simulation application running time based on a SHAP feature extraction method, and determining the correlation of each feature dimension on the simulation application running time;
s142, removing feature data irrelevant to the operation time of the simulation application in the feature data set;
s143, selecting relevant feature data which obviously influences the operation time of the simulation application; the correlated feature data includes strongly correlated feature data and weakly correlated and non-redundant feature data.
Specifically, after analyzing the influence degree of each feature dimension in the feature data set on the simulation application running time based on the SHAP feature extraction method, the relevance of each feature data on the simulation application running time can be determined, irrelevant features are removed, and strong relevant feature data and weak relevant but non-redundant feature data are selected, so that the occurrence of errors is minimized, and a more accurate prediction model is favorably established, namely the prediction accuracy of the sequencing learning model is improved.
Through the above processing steps, it is possible to minimize the occurrence of errors in the prediction process and improve the prediction accuracy.
In an embodiment, the process of constructing the rank learning model may specifically include the following steps:
determining a feedback function of the sequencing learning model according to a sequencing learning target of the cloud computing resource usage;
based on a feedback function, respectively endowing each sample pair in the characteristic data set with a sample weight and determining the weight distribution of the sample pairs;
and establishing a utility function and a loss function of the sequencing learning model based on the feedback function and the weight distribution to complete the construction of the sequencing learning model.
It can be appreciated that with respect to the construction process of the rank learning model, the problem is modeled first:
consider that the ordering goal for Set X indicates that it is necessary to apply S to a Set of simulations having the same simulation static parameters (except for the number of cores used) i And ranking is carried out. The optimal computing resource requirement for a simulation application in the cloud may be denoted as P i ＝f(S i (x i ,y i ) Where x) is i ＝(x 1 ,x 2 ,…x d Y) as a d-dimensional feature vector for simulation applications i Representing the runtime of the simulation application, f is the ranking function. The goal of the sequencing learning model is to learn a sequencing function f from the feature data set so as to correctly sequence the amount of cloud computing resources required to be used. Thus, a feedback function is defined:
for a pair of samples S in any set of simulation applications j And S k When S is j (x j ,y j )＞S k (x k ,y k ) When, indicates that P is used j The simulation runtime after a resource is greater than the usage P k Simulation run time after resources. When S is j (x j ,y j )＝S k (x k ,y k ) When, indicates that P is used j The simulation runtime after a resource is equal to the usage P k The simulation run time after the resource. When S is j (x j ,y j )＜S k (x k ,y k ) When, indicates that P is used j The simulation runtime after a resource is less than using P k The simulation run time after the resource. Based on the above, the above sequence learning task is converted into a binary classification task for learning the relative sequence between two sample pairs. Based on this, each sample pair is assigned with a weight value D, which represents the importance of the key sample pair being correctly judged. The distribution of D is defined as:
wherein c represents a positive value constant and satisfies:
therefore, the objective of the ranking learning model is to use the positive feedback sample pair with weight D to learn a final ranking function H, and the probability of the occurrence of the ranking error between the best ranking and the predicted ranking in the learning process is called ranking loss (ranking loss), and its utility function Func and loss function Rloss are respectively defined as:
wherein,r denotes the number of iterations, α r Weight, h, representing the weak classifier obtained at the current iteration r Represents a function having a value of 0-1 when α r H > 0 denotes r The ordering effect of (c) is positive, i.e., the accuracy is more than half. Therefore, only by selecting an appropriate α in each weak learning process r And h r The sorting penalty Rloss of the final sorting result can be minimized. In the present embodiment, α can be set by referring to the existing literature in the field r The weight value, defined as:
α r ＝0.5ln(1+r(h) max )/(1-r(h) max ) (6)
in an embodiment, as shown in fig. 4, regarding step S16, the following steps S161 to S165 may be specifically included:
s161, initializing weight distribution of the sample pairs in the related characteristic data and setting a positive constant of the weight distribution;
s162, in the current iteration process of the sequencing learning model, taking the maximized utility function as a learning target, carrying out weak learning iterative training by using the weight distribution of the current iteration to generate a plurality of weak classifiers and calculating the weight of the weak classifier obtained by the current iteration;
s163, calculating the sorting result of the current iteration according to the weight of the weak classifier obtained by the current iteration and the weak classifier;
s164, updating the weight distribution of the sample pairs according to the loss function of the sequencing learning model and entering next iteration; the processing flow of the next iteration is the same as that of the current iteration;
and S165, when the comprehensive sequence obtained in the continuous N iterations does not change, ending the iterative training of the sequence learning model and outputting the comprehensive sequence as the priority ranking of the use amount of different cloud computing resources. N is a positive integer not less than 2.
Specifically, the algorithm steps corresponding to the sequence learning model can be shown in the following flow of prediction algorithm:
in the prediction algorithm, S (X, Y) represents the input simulation application features and label data, and D represents the initialized sample pair weight distribution.
1) First initialize D 1 D, and the initial D value in each sample pair is made the same by setting the c value in equation (2).
2) In the iterative process, the maximum utility function r (h) is taken as a learning target, and D is used r Performing weak learning iterative training to generate a plurality of weak learners h r And alpha is obtained by calculation r The value is obtained. Finally, the result of the sorting is calculated as
Where R represents the number of iterations and H (x) represents the ranking score, with larger scores indicating higher ranking samples.
3) According to the formula:
and updating the D distribution value, wherein Rloss is the minimum objective function value in the current iteration and is calculated by using the formula (5).
4) And repeating the steps 2) and 3) until the comprehensive ranking sequence obtained in N continuous iterations is not changed, taking the comprehensive ranking sequence as the final predicted ranking and outputting the optimal sequence of the resource use. N may take the value of 2 or 3, and may be specifically selected according to actual needs.
In one embodiment, further, the loss function of the ranked learning model is:
wherein S is j (x j ,y j ) And S k (x k ,y k ) Represents a sample pair, ε (x) j,k ) Representing penalty factors，D(S j (x j ,y j ),S k (x k ,y k ) Denotes the weight distribution of the sample pairs, r denotes the number of iterations, α r Weight, h, representing the weak classifier obtained at the current iteration r Representing a function having a value of 0-1.
It will be appreciated that FIG. 5 is a flow diagram of the above method, wherein the loss function is a Hinge loss function and the concat represents the function. In order to make the order learning model more adaptive to the resource prediction problem of the simulation application, the following steps must be considered: the optimal resource sequence required by the operation of a simulation application is assumed to be a Correct ranking = {4,2,3,1} core. In which there are two prediction sorting results, respectively, a Ranking 1= {2,4,3,1} kernel and a Ranking 2= {4,2,1,3} kernel. It can be seen that both rankings for rankings 1 and 2 are incorrect. But in simulation practice, there is a greater concern about the amount of resource usage, i.e., the first ranked amount of resources, that will minimize the runtime of this simulation application. The cost of a false Ranking in Ranking 1 is therefore significantly higher than the cost of a false Ranking in Ranking 2, since the shortest simulation application run time can likewise be guaranteed on the basis of the Ranking results in Ranking 2. That is, a greater penalty should be added on the basis of Ranking 1. However, the conventional Rankboost Ranking learning method does not take this into account, and assumes that the costs of the wrong rankings in rankings 1 and 2 are equal. Based on this, the present embodiment adds a penalty factor ε (x) j,k ) To ensure that more penalties are added when the ranking of the best resource is incorrect. Therefore, the loss function in the above equation (5) is converted into the final loss function equation (8).
By adopting the loss function shown in the formula (8), the sorting cost in the resource sorting process is optimized and simulated, and compared with traditional machine learning methods such as linear regression, multilayer perceptron, regression tree and random forest, the accuracy of sorting can be obviously improved, and the error rate is reduced.
In an embodiment, the calculation process of the penalty factor may specifically include the steps of:
establishing a correct ranking according to the requirements of all cloud computing resources in the simulation application group, using MRR as a judgment standard of ranking performance and calculating a score MRR of the correct ranking;
for each sample pair in the simulation application set, exchanging the positions of the module pairs and calculating a new ranking score, MRR;
and calculating the average reduction ratio of the sequencing scores MRR of all the sample pairs after exchange relative to the correctly ranked scores MRR, and taking the average reduction ratio as a penalty factor.
It will be appreciated that a key issue with simulation computing resource prediction based on rank learning is the definition of penalty factors. The present embodiment uses a heuristic approach to calculate ε (x) j,k ) The specific implementation steps can be shown as the following algorithm flow:
first, the algorithm creates a correct ranking according to all the computing resource requirements in the simulation application group, and uses MRR as the criterion for ranking performance and calculates the optimal score MRR best (see equation 9).
Second, for each sample pair in the simulation application set, Q = { S = { (S) } j (x j ,y j ),S k (x k ,y k ) Exchange the positions of its module pairs and calculate a new ranking score MRR j,k 。
Finally, the average reduction ratio of the MRR values of all sample pairs after exchange to the correctly ranked MRR values is calculated and used as a penalty factor epsilon (x) j,k ) The value of (c).
In some embodiments, the present specification also provides evaluation indexes tested by the prediction method as described above:
to test the accuracy of the rank learning prediction method, two indices were considered: mean regenerative Rank (MRR, mean Reciprocal Rank) and Normalized discrete temporal Gain (NDCG, normalized mapped Cumulative Gain). Formal definition of the metric is as follows:
(1) The MRR focuses on the position where the first related element appears in the sorting result, and if the position where the first related element appears is more advanced, the MRR is larger, and the specific calculation is as follows:
where K denotes the number of samples, rank i Indicating the location in the ordered list of predicted computational resources where the best resource occurs for the ith sample.
(2) The NDCG is considered to be more valuable by introducing a position influence factor, and the influence on an evaluation result is larger when the samples are more advanced in the sorting, and the specific calculation is as follows:
wherein, Z n The value range of NDCG is limited to [0,1 ] for generalization factor]And c (i) represents the correlation level of the sample arranged at the i-th position, 2 c(i) -1 is the estimated gain value of the sample. Log (1 + i) represents the discount weight of the ranking position of the document, i.e. smaller i represents the ranking position is more advanced. Finally, NDCG @ n adds up all the evaluation gain values of the first n documents in the ranking as a performance evaluation value of the ranking model.
It should be understood that although the steps in the flowcharts of fig. 1-4 are shown in order as indicated by the arrows, the steps are not necessarily performed in order as indicated by the arrows. The steps are not performed in the exact order shown and described, and may be performed in other orders, unless explicitly stated otherwise. Moreover, at least some of the steps of fig. 1-4 may include multiple sub-steps or multiple stages that are not necessarily performed at the same time, but may be performed at different times, and the order of performance of the sub-steps or stages is not necessarily sequential, but may be performed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.
Referring to fig. 6, in an embodiment, a cloud simulation computing resource prediction apparatus 100 based on rank learning is provided, which includes a data acquisition module 11, a feature extraction module 13, a ranking prediction module 15, and a result output module 17. The data acquisition module 11 is used for acquiring a feature data set of a simulation application on a cloud computing node; the data in the feature dataset includes pre-runtime parameters and runtime parameters of the simulation application. The feature extraction module 13 is configured to extract relevant feature data that significantly affects the runtime of the simulation application from the feature data set based on the SHAP feature extraction method. The ranking prediction module 15 is used for inputting the relevant feature data into the constructed sequencing learning model and predicting the priority ranking of the usage amount of different cloud computing resources. The result output module 17 is configured to use the highest-ranking cloud computing resource usage amount output by the ranking learning model as an optimal cloud simulation computing resource prediction result.
According to the cloud simulation computing resource prediction device 100 based on sequencing learning, after the pre-operation parameters and the operation parameters of the simulation applications on the cloud computing nodes are obtained through cooperation of all modules, relevant feature data which obviously affect the operation time of the simulation applications are extracted based on a SHAP feature extraction method, then the priority ranking of the usage amount of different cloud computing resources is predicted by inputting the relevant feature data into a constructed sequencing learning model, and finally the highest ranked cloud computing resource usage amount is used as an optimal cloud simulation computing resource prediction result.
Therefore, the importance of the influence factors of the simulation running time of the complex system is quantitatively evaluated by adopting a feature extraction method based on SHAP interpretability, namely, a factor set influencing the running efficiency of the simulation application is analyzed firstly, and static features before the running of the simulation application and dynamic features in the running of the simulation application are extracted to serve as training sample feature data. And then, calculating the important contribution degree of each feature based on the SHAP value, extracting relevant feature data which obviously influences the operation time of the simulation application as model input, and predicting the use amount of the highest-ranked cloud computing resources by using the constructed sequencing learning model, thereby greatly improving the cloud computing resource prediction performance.
In an embodiment, the cloud simulated computing resource prediction apparatus 100 based on rank learning may be further configured to implement additional steps or sub-steps in other embodiments of the cloud simulated computing resource prediction method based on rank learning.
For specific limitations of the cloud-emulated computing resource prediction apparatus 100 based on rank learning, reference may be made to the corresponding limitations of the cloud-emulated computing resource prediction method based on rank learning, and details are not repeated here. The modules in the above-described order learning-based cloud simulation computing resource prediction apparatus 100 may be implemented in whole or in part by software, hardware, and a combination thereof. The modules may be embedded in a hardware form or a device independent of a specific data processing function, or may be stored in a memory of the device in a software form, so that a processor can call and execute operations corresponding to the modules, where the device may be, but is not limited to, various types of data processing devices existing in the art.
In one embodiment, there is also provided a computer device comprising a memory and a processor, the memory storing a computer program, the processor implementing the following processing steps when executing the computer program: acquiring a feature data set of simulation application on a cloud computing node; the data in the feature data set comprises pre-run parameters and run-time parameters of the simulation application; extracting relevant feature data which obviously influences the operation time of the simulation application from the feature data set based on a SHAP feature extraction method; inputting the relevant feature data into the constructed sequencing learning model, and predicting the priority ranking of the usage amount of different cloud computing resources; and taking the highest-ranking cloud computing resource usage amount output by the sequencing learning model as an optimal cloud simulation computing resource prediction result.
It is understood that the computer device includes, in addition to the memory and the processor, other software and hardware components not listed in this specification, which may be determined according to the model of the specific data processing device in different application scenarios, and detailed descriptions are not listed in this specification.
In one embodiment, the processor, when executing the computer program, may further implement the additional steps or sub-steps in the foregoing method for predicting cloud simulation computing resources based on rank learning.
In one embodiment, there is also provided a computer readable storage medium having stored thereon a computer program which when executed by a processor performs the process steps of: acquiring a feature data set of simulation application on a cloud computing node; the data in the feature data set comprises pre-run parameters and run-time parameters of the simulation application; extracting relevant feature data which obviously influences the operation time of the simulation application from the feature data set based on an SHAP feature extraction method; inputting the relevant feature data into the constructed sequencing learning model, and predicting the priority ranking of the usage amount of different cloud computing resources; and taking the cloud computing resource usage with the highest ranking output by the sequencing learning model as an optimal cloud simulation computing resource prediction result.
In one embodiment, when being executed by a processor, the computer program may further implement the additional steps or sub-steps in the foregoing cloud simulation computing resource prediction method based on rank learning.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above may be implemented by hardware that is instructed by a computer program, and the computer program may be stored in a non-volatile computer-readable storage medium, and when executed, may include the processes of the embodiments of the methods described above. Any reference to memory, storage, database, or other medium used in the embodiments provided herein may include non-volatile and/or volatile memory, among others. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically Programmable ROM (EPROM), electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double Data Rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous Link DRAM (SLDRAM), rambus DRAM (RDRAM), and interface DRAM (DRDRAM).
The technical features of the above embodiments can be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the above embodiments are not described, but should be considered as the scope of the present specification as long as there is no contradiction between the combinations of the technical features. The above examples only express several embodiments of the present application, and the description thereof is more specific and detailed, but not construed as limiting the scope of the invention. It should be noted that, for those skilled in the art, various changes and modifications can be made without departing from the spirit of the present application, and all of them fall within the scope of the present application. Therefore, the protection scope of the present patent should be subject to the appended claims.
Claims (10)
1. A cloud simulation computing resource prediction method based on sequencing learning is characterized by comprising the following steps:
acquiring a feature data set of simulation application on a cloud computing node; the data in the feature dataset comprises pre-runtime parameters and runtime parameters of the simulation application;
extracting relevant feature data which significantly influences the operation time of the simulation application from the feature data set based on a SHAP feature extraction method;
inputting the relevant feature data into a constructed sequencing learning model, and predicting the priority ranking of the usage amount of different cloud computing resources;
and taking the cloud computing resource usage with the highest ranking output by the sequencing learning model as an optimal cloud simulation computing resource prediction result.
2. The sequencing learning based cloud simulation computing resource prediction method of claim 1, wherein the step of extracting relevant feature data from the feature dataset that significantly affects simulation application runtime based on a SHAP feature extraction method comprises:
analyzing the influence degree of each feature dimension in the feature data set on the simulation application running time based on a SHAP feature extraction method, and determining the correlation of each feature dimension on the simulation application running time;
removing feature data in the feature data set that is not runtime-related to the simulation application;
selecting the relevant characteristic data which obviously influences the running time of the simulation application; the correlated feature data includes strongly correlated feature data and weakly correlated and non-redundant feature data.
3. The method for predicting cloud simulation computing resources based on rank learning according to claim 1 or 2, wherein the process for constructing the rank learning model comprises:
determining a feedback function of the sequencing learning model according to a sequencing learning objective of the cloud computing resource usage;
based on the feedback function, respectively endowing each sample pair in the characteristic data set with a sample weight and determining the weight distribution of the sample pairs;
and establishing a utility function and a loss function of the sequencing learning model based on the feedback function and the weight distribution, and completing the construction of the sequencing learning model.
4. The method for predicting cloud simulation computing resources based on rank learning according to claim 3, wherein the loss function of the rank learning model is as follows:
wherein S is j (x j ，y j ) And S k (x k ，y k ) Represents a sample pair, ε (x) j，k ) Denotes a penalty factor, D (S) j (x j ，y j )，S k (x k ，y k ) Represents the weight distribution of the sample pairs, r represents the number of iterations, α r Weight, h, representing the weak classifier obtained at the current iteration r Representing a function having a value of 0-1.
5. The method for predicting cloud simulation computing resources based on rank learning according to claim 4, wherein the calculation process of the penalty factor comprises:
establishing a correct ranking according to the requirements of all cloud computing resources in the simulation application group, using MRR as a judgment standard of ranking performance, and calculating a score MRR of the correct ranking;
for each sample pair in the simulation application set, exchanging the positions of the module pairs and calculating a new ranking score, MRR;
and calculating the average reduction ratio of the sorting scores MRR of all the sample pairs after exchange relative to the correctly ranked scores MRR, and taking the average reduction ratio as the penalty factor.
6. The sequencing learning-based cloud simulation computing resource prediction method according to claim 3, wherein the step of inputting the relevant feature data into the constructed sequencing learning model to predict the priority ranking of the usage amount of different cloud computing resources comprises:
initializing weight distribution of sample pairs in the related characteristic data and setting a positive constant of the weight distribution;
in the current iteration process of the sequencing learning model, taking a maximized utility function as a learning target, performing weak learning iteration training by using the weight distribution of the current iteration to generate a plurality of weak classifiers and calculating the weight of the weak classifier obtained by the current iteration;
calculating the sequencing result of the current iteration according to the weight of the weak classifier obtained by the current iteration and the weak classifier;
updating the weight distribution of the sample pairs according to the loss function of the sequencing learning model and entering next iteration; the processing flow of the next iteration is the same as that of the current iteration;
when the comprehensive sequence obtained in the continuous N iterations does not change, ending the iterative training of the sequence learning model and outputting the comprehensive sequence as the priority ranking of the different cloud computing resource usage amounts; n is a positive integer not less than 2.
7. The method of claim 1, wherein the method further comprises:
preprocessing the data in the feature data set; the preprocessing comprises data cleaning, error screening and data standardization.
8. A cloud simulation computing resource prediction device based on rank learning is characterized by comprising:
the data acquisition module is used for acquiring a characteristic data set of the simulation application on the cloud computing node; the data in the feature dataset comprises pre-runtime parameters and runtime parameters of the simulation application;
the characteristic extraction module is used for extracting relevant characteristic data which obviously influences the running time of the simulation application from the characteristic data set based on a SHAP characteristic extraction method;
the ranking prediction module is used for inputting the related characteristic data into the constructed sequencing learning model and predicting the priority ranking of the use amount of different cloud computing resources;
and the result output module is used for taking the highest-ranking cloud computing resource usage amount output by the sequencing learning model as an optimal cloud simulation computing resource prediction result.
9. A computer device comprising a memory and a processor, the memory storing a computer program, wherein the processor when executing the computer program implements the steps of the method for cloud emulated computing resource prediction based on rank learning of any one of claims 1 to 7.
10. A computer-readable storage medium having stored thereon a computer program, wherein the computer program, when executed by a processor, implements the steps of the method for ordering learning based cloud simulation computing resource prediction of any of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210997240.8A CN115344386A (en) | 2022-08-19 | 2022-08-19 | Method, device and equipment for predicting cloud simulation computing resources based on sequencing learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210997240.8A CN115344386A (en) | 2022-08-19 | 2022-08-19 | Method, device and equipment for predicting cloud simulation computing resources based on sequencing learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115344386A true CN115344386A (en) | 2022-11-15 |
Family
ID=83954136
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210997240.8A Pending CN115344386A (en) | 2022-08-19 | 2022-08-19 | Method, device and equipment for predicting cloud simulation computing resources based on sequencing learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115344386A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116562054A (en) * | 2023-07-06 | 2023-08-08 | 西安羚控电子科技有限公司 | Construction method and device of multi-entity collaborative real-time simulation system |
-
2022
- 2022-08-19 CN CN202210997240.8A patent/CN115344386A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116562054A (en) * | 2023-07-06 | 2023-08-08 | 西安羚控电子科技有限公司 | Construction method and device of multi-entity collaborative real-time simulation system |
| CN116562054B (en) * | 2023-07-06 | 2023-10-13 | 西安羚控电子科技有限公司 | Construction method and device of multi-entity collaborative real-time simulation system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110163261B (en) | Unbalanced data classification model training method, device, equipment and storage medium | |
| US20220076150A1 (en) | Method, apparatus and system for estimating causality among observed variables | |
| CN111258767B (en) | Cloud computing resource intelligent distribution method and device for complex system simulation application | |
| US20200372342A1 (en) | Systems and methods for predictive early stopping in neural network training | |
| Andrews et al. | Learning high-dimensional directed acyclic graphs with mixed data-types | |
| US20170330078A1 (en) | Method and system for automated model building | |
| CN104978612A (en) | Distributed big data system risk predicating method based on AHP-RBF | |
| CN114861788A (en) | Load abnormity detection method and system based on DBSCAN clustering | |
| CN113516019B (en) | Hyperspectral image unmixing method and device and electronic equipment | |
| Sarkar et al. | An algorithm for DNA read alignment on quantum accelerators | |
| Zhang et al. | Enabling in-situ data analysis for large protein-folding trajectory datasets | |
| CN115344386A (en) | Method, device and equipment for predicting cloud simulation computing resources based on sequencing learning | |
| Koutsoumpakis | Spark-based application for abnormal log detection | |
| CN116861373A (en) | Query selectivity estimation method, system, terminal equipment and storage medium | |
| US20190180180A1 (en) | Information processing system, information processing method, and recording medium | |
| Ding et al. | Efficient model-based collaborative filtering with fast adaptive PCA | |
| CN111984418B (en) | Automatic adjusting and optimizing method and device for granularity parameters of sparse matrix vector multiplication parallel tasks | |
| Zha et al. | Recognizing plans by learning embeddings from observed action distributions | |
| CN112667591A (en) | Data center task interference prediction method based on mass logs | |
| CN113111948A (en) | Time sequence data classification method and system based on feature recalibration mechanism | |
| Graner | Scalable Algorithms in NonparametricComputational Statistics | |
| WO2019209571A1 (en) | Proactive data modeling | |
| CN111026661A (en) | Method and system for comprehensively testing usability of software | |
| CN111612082A (en) | Method and device for detecting abnormal subsequence in time sequence | |
| CN112650770B (en) | MySQL parameter recommendation method based on query work load analysis |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |