Hyper-parameter space boundary analysis method and device, electronic equipment and storage medium

Abstract

The invention provides a hyper-parameter space boundary analysis method and device, electronic equipment and a storage medium. The method comprises the steps that first hyper-parameter data of the precision system is obtained, the first hyper-parameter data has a plurality of dimensions, each dimension represents a hyper-parameter, and the hyper-parameters are parameters influencing the operation effect of the precision system; performing dimension reduction processing on the first hyper-parameter data to obtain second hyper-parameter data, wherein the dimension number of the second hyper-parameter data is smaller than that of the first hyper-parameter data; performing continuous function modeling for every m different dimensions in the second hyper-parameter data to obtain a plurality of continuous function relationships; performing spatial optimization for each continuous function relationship to obtain a feature value of the second hyper-parameter data, wherein the feature value indicates the position of a point with the maximum gradient descent step length in the second hyper-parameter space; and carrying out boundary solving and dimension raising processing by utilizing the characteristic value of the second hyper-parameter data to obtain first hyper-parameter space boundary information. According to the invention, the efficiency and accuracy of hyper-parameter space boundary analysis can be improved.

Description

Technical Field

[0001] The present disclosure relates to the field of data processing, and in particular to a hyperparameter space boundary analysis method, device, electronic device and storage medium. Background Art

[0002] A precision system refers to a system that integrates advanced technology and high-precision components, and is designed to achieve specific, high-precision functions and tasks. It is increasingly used in various fields of real life. Usually for a precision system, there are many factors that affect its operating performance. These factors can be characterized by parameters and parameter values. Given the large number of types and wide range of values, they are often figuratively referred to as "super parameters" or "hyperparameters." How to select and set the range of parameters largely determines the operating effect of the precision system. Given the complexity of the operating environment, system conditions, operating conditions, etc. of the precision system, there are many types of parameters that affect the precision system. How to select parameters and determine the range of values, that is, determine the boundaries of the hyperparameter space, has become an urgent problem to be solved.

[0003] Currently, the determination of hyper-parameter space is mostly based on human experience. However, this method is inefficient on the one hand, and heavily relies on expert experience on the other hand. It is costly and difficult to control accuracy, which seriously affects the operating performance of the precision system. Summary of the invention

[0004] In view of this, the embodiments of the present disclosure are committed to providing a hyper-parameter space boundary analysis method, device, electronic device and storage medium, which can efficiently and accurately obtain the boundary information of the hyper-parameter space.

[0005] One aspect of the present disclosure provides a hyperparameter space boundary analysis method, comprising:

[0006] Acquire first hyperparameter data of a precision system, where the first hyperparameter data has multiple dimensions, each dimension represents a hyperparameter, and the hyperparameter is a parameter that affects an operation effect of the precision system;

[0007] Performing dimensionality reduction processing on the first hyperparameter data to obtain second hyperparameter data, wherein the number of dimensions of the second hyperparameter data is smaller than the number of dimensions of the first hyperparameter data;

[0008] Performing continuous function modeling for each m different dimensions in the second hyperparameter data to obtain multiple continuous function relationships, where m is an integer greater than 1, and the value of m is less than the total number of dimensions of the second hyperparameter data;

[0009] Performing spatial optimization for each of the continuous function relationships to obtain an eigenvalue of the second hyperparameter data, the eigenvalue indicating the position of the point with the largest gradient descent step in the second hyperparameter space, the second hyperparameter space corresponding to the second hyperparameter data;

[0010] The eigenvalues ​​of the second hyperparameter data are used to perform boundary solving and dimensionality upgrading to obtain boundary information of the first hyperparameter space. The boundary information of the first hyperparameter space is used to set or optimize the precision system. The first hyperparameter space corresponds to the first hyperparameter data.

[0011] In some implementations of the first aspect of the present disclosure, the method further includes: projecting the second hyperparameter data into an orthogonal space to complete numerical reduction processing of the second hyperparameter data, thereby reducing the data volume of the second hyperparameter data.

[0012] In some embodiments of the first aspect of the present disclosure, the first hyperparameter data includes at least one of the following hyperparameters: environmental parameters, component parameters of the precision system, motion parameters and performance parameters of the precision system.

[0013] In some embodiments of the first aspect of the present disclosure, continuous function modeling is performed on the second hyperparameter data to obtain multiple continuous function relationships, including: performing data analysis on the first dimensional data and the second dimensional data in the second hyperparameter data to construct a first function model, the first function model conforming to the data distribution of the first dimensional data and the second dimensional data in the second hyperparameter data; performing continuous function fitting on the first dimensional data and the second dimensional data in the second hyperparameter data according to the first function model to obtain parameter values ​​of the first function model; substituting the parameter values ​​of the first function model into the first function model to obtain a first continuous function relationship, the first continuous function relationship characterizing the correlation between the first dimension and the second dimension in the second hyperparameter data; wherein the first dimension and the second dimension are any two different dimensions in the second hyperparameter data, respectively.

[0014] In some embodiments of the first aspect of the present disclosure, the spatial optimization is performed for each of the continuous function relationships to obtain the eigenvalues ​​of the second hyperparameter data, including: performing gradient descent processing on each of the continuous function relationships to obtain the eigenvalues ​​corresponding to each of the continuous function relationships, the eigenvalue of a continuous function relationship being used to indicate the position of the point with the largest gradient change on the plane or subspace formed by the dimension corresponding to the continuous function relationship in the second hyperparameter space.

[0015] In some embodiments of the first aspect of the present disclosure, the use of the eigenvalues ​​of the second hyperparameter data to perform boundary solution and dimensionality upgrading to obtain boundary information of the first hyperparameter space includes: using the eigenvalues ​​of the second hyperparameter data to perform boundary solution to obtain second hyperparameter space boundary information, the second hyperparameter space boundary information including the numerical range of each dimension in the second hyperparameter data; and, performing dimensionality upgrading on the boundary information of the second hyperparameter space to obtain first hyperparameter space boundary information, the first hyperparameter space boundary information including the numerical range of each dimension in the first hyperparameter data.

[0016] In some embodiments of the first aspect of the present disclosure, the use of the eigenvalues ​​of the second hyperparameter data to perform boundary solution to obtain boundary information of the second hyperparameter space includes: using the eigenvalues ​​of the second hyperparameter data to perform a grid search in the second hyperparameter space to determine the numerical range of each dimension in the second hyperparameter data.

[0017] A second aspect of the present disclosure provides a hyperparameter space boundary analysis device, comprising:

[0018] An acquisition unit, used to acquire first hyperparameter data of a precision system, wherein the first hyperparameter data has multiple dimensions, each dimension represents a hyperparameter, and the hyperparameter is a parameter that affects an operation effect of the precision system;

[0019] A dimensionality reduction unit, configured to perform dimensionality reduction processing on the first hyperparameter data to obtain second hyperparameter data, wherein the number of dimensions of the second hyperparameter data is smaller than the number of dimensions of the first hyperparameter data;

[0020] a continuous function modeling unit, configured to perform continuous function modeling for each m different dimensions in the second hyperparameter data to obtain a plurality of continuous function relationships, where m is an integer greater than 1, and the value of m is less than the total number of dimensions of the second hyperparameter data;

[0021] an eigenvalue determining unit, configured to perform spatial optimization for each of the continuous function relationships to obtain an eigenvalue of the second hyperparameter data, wherein the eigenvalue indicates the position of the point with the largest gradient descent step in the second hyperparameter space, and the second hyperparameter space corresponds to the second hyperparameter data;

[0022] A boundary information acquisition unit is used to use the eigenvalues ​​of the second hyperparameter data to perform boundary solution and dimensionality upgrading processing to obtain boundary information of the first hyperparameter space, wherein the boundary information of the first hyperparameter space is used to set or optimize the precision system, and the first hyperparameter space corresponds to the first hyperparameter data.

[0023] The third aspect of the present disclosure provides an electronic device, comprising: a processor and a memory; wherein the memory is connected to the processor, and the memory is used to store a computer program; the processor is used to implement the above method by running the computer program stored in the memory.

[0024] A fourth aspect of the present disclosure provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the above method is implemented.

[0025] According to the embodiment of the present application, the hyperparameter data is first reduced in dimension, and then continuous function modeling and spatial optimization are performed for the hyperparameter data obtained by the dimension reduction to obtain the eigenvalues ​​of the hyperparameter data after the dimension reduction, and finally the eigenvalues ​​of the hyperparameter data after the dimension reduction are used to obtain the hyperparameter space boundary of the corresponding original hyperparameter data through boundary solution and dimension increase processing. Among them, continuous function modeling can capture the potential patterns and trends in the data by fitting continuous functions, so as to realize the analysis of the relationship between different dimensions in the hyperparameter data after dimension reduction. This hyperparameter analysis method can better fit the continuous function model to the data, and can better solve the global optimal point on the continuous function model. The boundary required does not only exist between the existing data, but also can predict the data that does not exist, which greatly improves the analysis efficiency and accuracy of the hyperparameter space. Based on this, the precision system can be set or optimized without relying on human experience, which can improve efficiency and accuracy and reduce labor costs. In addition, by reducing the dimension first and then performing continuous function modeling and spatial optimization, the complexity can be greatly reduced, and the cost of multi-dimensional hyperparameter space analysis can be reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the embodiments of the present disclosure or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0027] Figure 1 It is a flow chart of the hyper-parameter space boundary analysis method provided in an embodiment of the present application.

[0028] Figure 2 It is a flow chart of continuous function modeling involved in the embodiments of the present application.

[0029] Figure 3 It is a schematic diagram of the process of obtaining the first hyper-parameter space boundary information involved in an embodiment of the present application.

[0030] Figure 4It is a structural schematic diagram of the hyper-parameter space boundary analysis device provided in an embodiment of the present application.

[0031] Figure 5 It is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0032] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not 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.

[0033] Terminology explanation:

[0034] Hyperparameter space: hyperparameter space, but the "hyperparameter" involved in the embodiments of the present application refers to the parameters that affect the operating effect of the precision system, which are vividly called "hyperparameters" or "super parameters" because of their many types and wide range of values. Accordingly, the hyperparameter space, i.e., the hyperparameter space (Hyperparameter Space), refers to the set of all possible hyperparameter values. In the hyperparameter space, each dimension represents a hyperparameter, and different dimensions have different degrees of correlation, and each point represents a specific set of hyperparameter settings. For example, if there are four hyperparameters: wind direction, wind speed, air pressure, and temperature, then the hyperparameter space is a four-dimensional space. The size and complexity of the hyperparameter space depends on the number of hyperparameters and the range of their possible values. In view of the large number of hyperparameter types that affect the operating effect of the precision system, the hyperparameter space is usually a high-dimensional space, and a high-dimensional space or a hyperparameter with a large number of possible values ​​makes it more difficult to search for the optimal hyperparameter configuration, which is the so-called "dimensionality disaster". Therefore, choosing a suitable hyperparameter space is crucial for precision systems.

[0035] Hyperparameter space boundary: the boundary of the hyperparameter space. The boundary of the hyperparameter space refers to a complete set of possible value ranges of the hyperparameters. The boundary of the hyperparameter space has an important impact on the setting, testing, and optimization of precision systems. In the disclosed embodiment, the hyperparameter space boundary refers to the fastest changing position in the hyperparameter space. The fastest changing position corresponds to a range rather than a value. The range is the boundary or sampling interval.

[0036] Continuous function: It is a basic concept in mathematical analysis, which means that within a certain interval, there is no discontinuity between the value of the function and the independent variable.

[0037] Exemplary Methods

[0038] Figure 1FIG. 1 is a flow chart of a hyperparameter space boundary analysis method provided by an embodiment of the present disclosure. Figure 1 As shown, the hyperparameter space boundary analysis method may include the following steps:

[0039] Step 101, obtaining first hyperparameter data of a precision system, where the first hyperparameter data has multiple dimensions, each dimension represents a hyperparameter, and the hyperparameter is a parameter that affects the operation effect of the precision system;

[0040] Step 102, performing dimensionality reduction processing on the first hyperparameter data to obtain second hyperparameter data, wherein the number of dimensions of the second hyperparameter data is smaller than the number of dimensions of the first hyperparameter data;

[0041] Step 103, performing continuous function modeling for each m different dimensions in the second hyper-parameter data to obtain multiple continuous function relationships, where m is an integer greater than 1, and the value of m is less than the total number of dimensions of the second hyper-parameter data;

[0042] Step 104, performing spatial optimization for each continuous function relationship to obtain an eigenvalue of each continuous function relationship, the eigenvalue is used to indicate the position of the point with the largest gradient descent step length in the second hyperparameter space, and the second hyperparameter space corresponds to the second hyperparameter data;

[0043] Step 105, using the eigenvalues ​​of each continuous function relationship to perform boundary solution and dimensionality increase processing to obtain first hyperparameter space boundary information, the first hyperparameter space boundary information is used to set or optimize the precision system, and the first hyperparameter space corresponds to the first hyperparameter data.

[0044] In some implementations, before step 101, the hyperparameter space boundary analysis method may further include: performing data preprocessing on the first hyperparameter data, wherein the data preprocessing includes one or more of the following: data cleaning, inconsistency processing, normalization processing, and missing value processing. Data preprocessing can effectively improve the data quality and analysis efficiency of the first hyperparameter data, and can also reduce risks, provide a reliable foundation for subsequent steps, and help to further improve the effectiveness and efficiency of multidimensional data and hyperparameter space analysis.

[0045] In some examples, data cleaning may include checking, correcting, and screening the first hyperparameter data to ensure the accuracy and completeness of the first hyperparameter data.

[0046] In some examples, inconsistency processing may include: unifying the units and scales of data of different dimensions in the first hyperparameter data. By normalizing the first hyperparameter data, the dimensional differences between different features may be eliminated.

[0047] In some examples, missing value processing may include: detecting and processing missing values ​​in the first hyperparameter data, filling missing values ​​by means of means, deleting records containing missing values, or processing by interpolation, etc. By processing missing values, the integrity and accuracy of the first hyperparameter data can be ensured.

[0048] In the disclosed embodiment, the first hyperparameter data is a data set of hyperparameters of a precision system, which may include all parameters that affect the operation of the precision system. The precision system may be any type of precision system. For example, aircraft, guided weapons, precision positioning systems, etc. in the aerospace field. Accordingly, the hyperparameters that affect the precision system may include, but are not limited to, environmental parameters, component parameters of the precision system, motion parameters of the precision system, performance parameters, etc.

[0049] Environmental parameters may include parameters of the production or manufacturing environment of precision systems, parameters of the transportation environment, parameters of the operating environment, etc. For example, environmental parameters for manufacturing aircraft include: manufacturing temperature, humidity, manufacturing equipment accuracy, etc. Aircraft operating environment parameters include wind direction, wind speed, air pressure, temperature, space altitude particles, space magnetic field, noise, vibration, etc.

[0050] The number of component parameters of precision systems is even greater. Taking aircraft as an example, they can include frame parameters, power system parameters, control system parameters, etc. Frame parameters can specifically include weight, wheelbase, materials, etc. Power system parameters can include motor parameters (such as nominal no-load KV value, maximum peak current, maximum peak power, etc.), battery parameters (such as capacity, voltage, discharge rate, etc.), propeller parameters (such as pitch, model, chord length, number of blades, safe speed, etc.). Control system parameters include receiver parameters (such as frequency, modulation mode, number of channels, remote control distance), autopilot parameters (such as GPS, IMU, barometer, etc.).

[0051] The motion parameters of a precision system may include speed, attitude, acceleration, etc. Taking an aircraft as an example, they may include take-off speed, cruising speed, maximum speed, glide speed, climb rate, turning radius, pitch angle, yaw angle, roll angle, airflow angle, angular velocity, acceleration, etc.

[0052] Taking an aircraft as an example, performance parameters may include: ceiling, range, endurance, activity radius, maneuverability parameters, take-off performance parameters, landing performance parameters, stability parameters, target hitting accuracy, etc.

[0053] As can be seen from the above description, there are usually a large number of types of hyperparameters that affect the operation of precision systems.

[0054] In some embodiments, if there are N hyperparameters of the precision system, where N is an integer greater than 1, the first hyperparameter data of the model is N-dimensional data, and the first hyperparameter space corresponding to the first hyperparameter data is N-dimensional space. If the model has thousands or tens of thousands of hyperparameters, its first hyperparameter data is high-dimensional data of thousands or tens of thousands of dimensions, and the corresponding first hyperparameter space is thousands of dimensions or tens of thousands of dimensions.

[0055] The dimension of the second hyperparameter data obtained in step 102 is smaller than the first hyperparameter data, and the dimension of the second hyperparameter space corresponding to the second hyperparameter data is also smaller than the first hyperparameter space.

[0056] The dimensionality reduction process in step 102 can be implemented in a variety of applicable ways. In some implementations, the dimensionality reduction process of the first hyperparameter data in step 102 can be implemented in one of the following ways: principal component analysis (PCA), linear discriminant analysis (LDA), singular value decomposition (SVD).

[0057] In step 102, the dimensionality reduction processing can be used to achieve dimensionality reduction of the first hyperparameter data, reduce the number of features in the data set of the first hyperparameter data to reduce the data dimension, and create a data set with smaller dimension from the data set of the second hyperparameter data without losing important information of the original data set, thereby reducing the complexity of the hyperparameter data while retaining the important information in the original hyperparameter data and saving the cost of data processing.

[0058] In some implementations, before step 103, the hyperparameter space boundary analysis method may further include: projecting the second hyperparameter data to the orthogonal space to complete the numerical reduction processing of the second hyperparameter data, thereby reducing the data volume of the second hyperparameter data. The numerical reduction of the second hyperparameter data can be achieved by projecting the second hyperparameter data to the orthogonal space, and the data volume can be reduced by reducing the number of records in the data set of the second hyperparameter data, thereby further reducing the complexity of the second hyperparameter data without affecting the important information in the second hyperparameter data, saving the data processing cost of subsequent steps.

[0059] In step 103, each continuous function relationship represents the correlation of m different dimensions in the second hyperparameter data, where m is an integer greater than 1 and the value of m is less than the total number of dimensions of the second hyperparameter data. Assuming that the second hyperparameter data has 4 dimensions, namely A dimension, B dimension, C dimension, and D dimension, m can be 2 or 3. If m is set to 2 in step 103, that is, continuous function modeling is selected for every two different dimensions, then after the continuous function modeling of step 103, 6 continuous function relationships will be obtained, and these 6 continuous function relationships are: a first continuous function relationship representing the correlation between A dimension and B dimension, a second continuous function relationship representing the correlation between A dimension and C dimension, a third continuous function relationship representing the correlation between A dimension and D dimension, a fourth continuous function relationship representing the correlation between B dimension and C dimension, a fifth continuous function relationship representing the correlation between B dimension and D dimension, and a sixth continuous function relationship representing the correlation between C dimension and D dimension.

[0060] Figure 2 FIG. 1 shows an exemplary flow chart of continuous function modeling in step 103. Figure 2 In step 103, the process of performing continuous function modeling for any two different dimensions in the second hyperparameter data may include:

[0061] Step 201, performing data analysis on the first dimension data and the second dimension data in the second hyperparameter data to construct a first function model, where the first function model conforms to the data distribution of the first dimension data and the second dimension data in the second hyperparameter data;

[0062] Assume that the second hyperparameter data after dimensionality reduction has 5 dimensions, randomly select one dimension as the first dimension x, and the other four dimensions as the second dimension y. Use the data on the first dimension x and the data on the second dimension y to fit a function, which is the first function model. Since the second hyperparameter data conforms to the standard normal distribution, the first function model conforms to the data distribution of the second hyperparameter data.

[0063] By first performing continuous function modeling on the second hyperparameter data through function modeling, a continuous second hyperparameter space is simulated, which can ensure that the first hyperparameter space after subsequent restoration can also be fitted with a continuous model, thereby more accurately and efficiently determining the hyperparameter space boundary of the first hyperparameter space.

[0064] Step 202, performing continuous function fitting on the first dimension data and the second dimension data in the second hyperparameter data according to the first function model to obtain parameter values ​​of the first function model;

[0065] Step 203: Substitute the parameter values ​​of the first function model into the first function model to obtain a first continuous function relationship that characterizes the correlation between the first dimension and the second dimension in the second hyperparameter data.

[0066] Among them, the first dimension and the second dimension are any two different dimensions in the second hyper-parameter data, the first function model can be any type of continuous function model, and by performing steps 201 to 203 for all dimensions in the second hyper-parameter data, all continuous function relationships of the second hyper-parameter data can be obtained, and each continuous function relationship can characterize the correlation between any two different dimensions in the second hyper-parameter data.

[0067] The function model used when modeling the continuous function in step 103 may be, but is not limited to, a linear function, a polynomial function, an exponential function, a logarithmic function, a power function, a sine function, a cosine function, a sine function including an offset, a cosine function including an offset, etc., and the embodiments of the present disclosure do not impose any restrictions on this.

[0068] In step 103, by fitting a continuous function to the data scatter distribution on each m different dimensions in the second hyperparameter data, a mathematical function representation (i.e., a continuous function relationship) between different dimensional features is obtained, thereby realizing continuous function modeling. Since the method based on function fitting has higher flexibility, the continuous function model can model the data more carefully. They allow the output to vary within a range rather than being restricted to discrete values. Therefore, this method can better capture the changes and trends of the data, more comprehensively and flexibly represent the complex relationship between different dimensions in the second hyperparameter data, and provide a more detailed description, which provides reliable support for the subsequent eigenvalue search, thereby effectively improving the accuracy of the hyperparameter space analysis.

[0069] It should be noted that, although the embodiment of the present disclosure selects to perform continuous function modeling for every two different dimensions in the second hyperparameter data, it is understandable that continuous function modeling can also be performed for every three different dimensions, every four different dimensions, etc. in the second hyperparameter data. That is, in step 103, the number of dimensions corresponding to each continuous function relationship can be flexibly set as needed, and the corresponding continuous function model can also be flexibly selected as needed.

[0070] In some implementations, step 104 may be implemented by a gradient descent method.

[0071] Specifically, performing spatial optimization for each continuous function relationship in step 104 may include: performing gradient descent processing for each continuous function relationship to obtain the eigenvalues ​​of each continuous function relationship, and the eigenvalue of a continuous function relationship may be used to indicate the position of the point with the largest gradient change on the plane or subspace formed by the dimension corresponding to the continuous function relationship in the second hyperparameter space. Thus, the position of the point with the largest gradient value change on each subspace or plane in the second hyperparameter space can be obtained. The point with the largest gradient value change is the position with the fastest gradient descent, the point with the fastest change in the second hyperparameter space, that is, the point with the most drastic change in the continuous function relationship, and the position of this point is also the position with the fastest change in other dimensions (the dimension corresponding to the independent variable in the continuous function relationship) under the specified dimension (the dimension corresponding to the function value in the continuous function relationship) in the second hyperparameter space.

[0072] Gradient descent is a search-based optimization algorithm used to minimize a function. In some implementations, in step 104, the gradient descent process of a single continuous function relationship may include: iteratively adjusting the parameters of the continuous function relationship in the opposite direction of the gradient of the continuous function relationship until the local minimum of the continuous function relationship is reached or the continuous function relationship converges to the global optimal solution. Specifically, the derivative and gradient of the continuous function relationship are first calculated to obtain the function value of the continuous function relationship with the most drastic change (i.e., the gradient value changes the most), and the function value is substituted into the continuous function relationship to obtain the value of the independent variable of the continuous function relationship with the most drastic change (i.e., the gradient value changes the most), and the function value and the independent variable value are the aforementioned eigenvalues.

[0073] Still taking the second hyperparameter data of the four dimensions mentioned above as an example, a gradient descent process can be performed on each of the six continuous function relationships to obtain the eigenvalues ​​of the six continuous function relationships, which can indicate the position of the point where the gradient descent is fastest on the hyperparameter plane formed by the two dimensions corresponding to the corresponding continuous function relationship in the second hyperparameter space. For example, assuming that the first continuous function relationship y(x) takes the A dimension as the function value y and the B dimension as the independent variable x, then the eigenvalue obtained by performing the gradient descent process on the first function relationship is a numerical pair, which can be expressed as {x B ,y A}, x B is a value in the B dimension in the second hyperparameter space, y A is a value in the A dimension of the second hyperparameter space, x B The value of is the value of the independent variable with the largest gradient change of the first continuous function relationship y(x), y A The value of is equal to the function value of the first continuous function relationship y(x) with the largest gradient change. B ,y A} represents the point where the gradient decreases fastest on the hyperparameter plane AB formed by the A dimension and the B dimension in the second hyperparameter space.

[0074] As can be seen from the above, by performing spatial optimization on the continuous function relationship, the positions of the points where the gradient descent is fastest in the second hyperparameter space can be found. These points are points on the boundary of the second hyperparameter space.

[0075] There can be multiple specific implementation processes in step 105. In one implementation, the boundary solution can be first performed using the eigenvalues ​​of the second hyperparameter data, and then the result of the boundary solution is subjected to dimensionality increase processing, thereby obtaining the boundary information of the first hyperparameter space. In another implementation, the eigenvalues ​​of the second hyperparameter data can be subjected to dimensionality increase processing, and then the result of the dimensionality increase processing is used to make a boundary request. In these two implementations, the specific implementation methods of boundary solution and dimensionality increase processing are similar, and the difference is the execution order of the two.

[0076] Figure 3 FIG. 1 is a schematic diagram showing an exemplary process of obtaining the boundary information of the first hyperparameter space in step 105. Figure 3 In some implementations, step 105 may include:

[0077] Step 301, using the eigenvalues ​​of the second hyperparameter data to perform boundary solving to obtain second hyperparameter space boundary information, where the second hyperparameter space boundary information includes the value range of each dimension in the second hyperparameter data;

[0078] Step 302: Perform dimensionality upscaling processing on the boundary information of the second hyperparameter space to obtain the boundary information of the first hyperparameter space. The boundary information of the first hyperparameter space includes the numerical range of each dimension in the first hyperparameter data.

[0079] When the eigenvalue, i.e., the point where the gradient in the second hyperparameter space drops fastest, is found, the points on the boundary of the second hyperparameter space are found, and the boundary information of the second hyperparameter space can be determined using these points. Specifically, the boundary solution in step 301 may include: using the eigenvalue of the second hyperparameter data to perform a grid search in the second hyperparameter space to determine the numerical range of each dimension in the second hyperparameter data.

[0080] The grid search algorithm is a method for finding the boundaries of hyperparameters. It determines the optimal boundary combination by traversing pre-specified parameter combinations. Its basic idea is to find the best parameter combination through exhaustive search. Each parameter takes a value within a pre-defined range, and then evaluates all possible parameter combinations and selects the best combination as the final boundary range.

[0081] The grid search algorithm defines a list of possible values ​​of a hyperparameter, and then forms a grid (or Cartesian product) with the combination of these values, thereby constructing the boundary of the entire hyperparameter space. Specifically, the following processing is performed for each eigenvalue of the second hyperparameter data: the point corresponding to the eigenvalue in the second hyperparameter space is moved in steps. In practical applications, the moving step can be determined according to 10% of the value of the point. For example, if the value of the point found is 100, the moving step can be 10; then, according to the moving step, the numerical change of a dimension y in the second hyperparameter space is detected at the same time, and the movement is cyclically moved. When the gradient of the value of the dimension y changes from positive to negative or from negative to positive, the movement stops. Since the gradient can be in the positive direction of the x-axis or in the negative direction of the x-axis, there are two points where the gradient changes from positive to negative or from negative to positive, and these two points are the boundaries of the dimension y. The dimension of the boundary information of the second hyperparameter space obtained in step 301 is the same as the dimension of the second hyperparameter data. In view of this, in some embodiments, the dimensionality increase processing in step 302 can be the inverse process of the dimensionality reduction processing in step 102, and the dimensionality increase processing is used to increase the dimension of the second hyper-parameter space boundary information of the lower dimension to the first hyper-parameter space boundary information of the higher dimension, and the dimension of the first hyper-parameter space boundary information is the same as the dimension of the first hyper-parameter data.

[0082] For example, if an autoencoder is used for dimensionality reduction processing in step 102, then in step 302, the second hyperparameter space boundary information can be reversely substituted into the autoencoder for dimensionality increase processing to obtain the first hyperparameter space boundary information.

[0083] The hyperparameter space boundary analysis method of the disclosed embodiment first reduces the dimension of the hyperparameter data, then performs continuous function modeling and spatial optimization on the hyperparameter data obtained by the dimension reduction to obtain the eigenvalues ​​of the hyperparameter data after the dimension reduction, and finally uses the eigenvalues ​​of the hyperparameter data after the dimension reduction to obtain the hyperparameter space boundary corresponding to the original hyperparameter data through boundary solution and dimension increase processing. Among them, continuous function modeling can capture the potential patterns and trends in the data by fitting continuous functions, thereby realizing the analysis of the relationship between different dimensions in the hyperparameter data after dimension reduction. This hyperparameter analysis method can better fit the continuous function model to the data, and can better solve the global optimal point on the continuous function model. The boundary required does not only exist between the existing data, but also can predict the data that does not exist, which greatly improves the analysis efficiency and accuracy of the hyperparameter space, thereby improving the speed, efficiency and accuracy of multi-dimensional hyperparameter space analysis, improving the objectivity of analysis, reducing the risk of errors, and realizing flexible automation of hyperparameter space analysis. In addition, by reducing the dimension first and then performing continuous function modeling and spatial optimization, the complexity can be greatly reduced and the cost of multi-dimensional hyperparameter space analysis can be reduced.

[0084] After determining the boundary of the hyperparameter space, the precision system can be set or optimized according to the data range of each dimension in the determined second hyperparameter data. For example, after the above process, it is determined that the wind speed between 13 and 20 (m / s) has the greatest impact on the accuracy of the aircraft capturing the target, which means that the accuracy of the aircraft capturing the target is most sensitive to the wind speed in this interval. The aircraft can be further set or optimized accordingly. For example, the weight of the wind speed parameter within this interval can be increased.

[0085] Exemplary Devices

[0086] Figure 4 FIG. 1 shows a schematic diagram of the structure of a hyperparameter space boundary analysis device provided in an embodiment of the present application. Figure 4 , the hyper-parameter space boundary analysis device 400 of the embodiment of the present disclosure may include:

[0087] An acquisition unit 401 is used to acquire first hyperparameter data of a precision system, where the first hyperparameter data has multiple dimensions, each dimension represents a hyperparameter, and the hyperparameter is a parameter that affects an operation effect of the precision system;

[0088] A dimensionality reduction unit 402 is used to perform dimensionality reduction processing on the first hyperparameter data to obtain second hyperparameter data, where the number of dimensions of the second hyperparameter data is less than the number of dimensions of the first hyperparameter data;

[0089] A continuous function modeling unit 403 is used to perform continuous function modeling for each m different dimensions in the second hyperparameter data to obtain multiple continuous function relationships, where m is an integer greater than 1, and the value of m is less than the total number of dimensions of the second hyperparameter data;

[0090] An eigenvalue determining unit 404 is used to perform spatial optimization for each continuous function relationship to obtain an eigenvalue of the second hyperparameter data, the eigenvalue indicating the position of the point with the largest gradient descent step in the second hyperparameter space, and the second hyperparameter space corresponds to the second hyperparameter data;

[0091] The boundary information acquisition unit 405 is used to use the eigenvalues ​​of the second hyperparameter data to perform boundary solution and dimensionality increase processing to obtain the boundary information of the first hyperparameter space. The boundary information of the first hyperparameter space is used to set or optimize the precision system. The first hyperparameter space corresponds to the first hyperparameter data.

[0092] In some implementations, the hyperparameter space boundary analysis device 400 of the disclosed embodiment may further include: a numerical reduction unit 406, for projecting the second hyperparameter data into the orthogonal space to complete the numerical reduction processing of the second hyperparameter data, thereby reducing the data volume of the second hyperparameter data.

[0093] In some embodiments, the hyperparameter space boundary analysis device 400 of the embodiment of the present disclosure may also include: a preprocessing unit 407, which is used to perform data preprocessing on the first hyperparameter data, and the data preprocessing includes one or more of the following: data cleaning, inconsistency processing, normalization processing, and missing value processing.

[0094] In some implementations, the first hyperparameter data of the disclosed embodiments include at least one of the following hyperparameters: environmental parameters, component parameters of the precision system, motion parameters and performance parameters of the precision system.

[0095] In some implementations, the continuous function modeling unit 403 may be specifically used for:

[0096] Performing data analysis on the first dimension data and the second dimension data in the second hyperparameter data to construct a first function model, where the first function model conforms to the data distribution of the first dimension data and the second dimension data in the second hyperparameter data;

[0097] Performing continuous function fitting on the first dimension data and the second dimension data in the second hyperparameter data according to the first function model to obtain parameter values ​​of the first function model;

[0098] Substituting the parameter value of the first function model into the first function model to obtain a first continuous function relationship, where the first continuous function relationship represents the correlation between the first dimension and the second dimension in the second hyperparameter data;

[0099] The first dimension and the second dimension are any two different dimensions in the second hyperparameter data.

[0100] In some embodiments, the eigenvalue determination unit 404 can be specifically used to: perform gradient descent processing on each continuous function relationship respectively to obtain the eigenvalue corresponding to each continuous function relationship, and the eigenvalue of a continuous function relationship is used to indicate the position of the point with the largest gradient change on the plane or subspace formed by the dimension corresponding to the continuous function relationship in the second hyperparameter space.

[0101] In some embodiments, the boundary information acquisition unit 405 can be specifically used to: use the eigenvalues ​​of the second hyperparameter data to perform boundary solution to obtain the second hyperparameter space boundary information, the second hyperparameter space boundary information includes the numerical range of each dimension in the second hyperparameter data; and, perform dimensionality upgrading processing on the boundary information of the second hyperparameter space to obtain the first hyperparameter space boundary information, the first hyperparameter space boundary information includes the numerical range of each dimension in the first hyperparameter data.

[0102] Other technical details of the hyper-parameter space boundary analysis device 400 in the embodiment of the present application can be found in the previous method section and will not be repeated here.

[0103] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can refer to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system or system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiment. The system and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without creative work.

[0104] Exemplary Electronic Devices

[0105] Figure 5 FIG. 1 shows an example diagram of the structure of an electronic device provided in an embodiment of the present application. Figure 5 As shown, the electronic device may include: one or more processors 501, and a memory 502 for storing one or more programs, which are executed by the one or more processors 501 to implement the method flow shown in the above embodiments of the present disclosure and / or program units corresponding to each unit in the device.

[0106] The various components are interconnected using different buses and can be mounted on a common motherboard or in other ways as needed. The processor 501 can process instructions executed within the electronic device, including instructions stored in or on the memory to display graphical information of the user interface on an external input / output device (such as a display device coupled to the interface). In other embodiments, if desired, multiple processors and / or multiple buses can be used with multiple memories and multiple memories.

[0107] The processor 501 may include one or more single-core processors or multi-core processors. The processor 501 may include any combination of general-purpose processors or dedicated processors (such as image processors, application processors, baseband processors, etc.).

[0108] The memory 502 is a computer-readable storage medium provided by the present disclosure, which can be used to store non-transient software programs, non-transient computer executable programs and units, such as the following in the embodiments of the present disclosure: Figure 1 The processor 501 executes the non-transient software programs, instructions and units stored in the memory 502, thereby executing the above method embodiments. Figure 1The methods shown correspond to programs, instructions, and units.

[0109] The electronic device may further include: an input device 503 and an output device 504. The processor 501, the memory 502, the input device 503 and the output device 504 may be connected via a bus or other means. Figure 5 The example of connecting through bus is taken in the following.

[0110] The input device 503 can receive input digital or character information, and generate signal input related to user settings and function control of the camera calibration device, such as a touch screen, a keypad, a mouse, a track pad, a touch pad, an indicator rod, one or more mouse buttons, a trackball, a joystick, and other input devices. The output device 504 may include a display device, an auxiliary lighting device (e.g., an LED), and a tactile feedback device (e.g., a vibration motor). The display device may include, but is not limited to, a liquid crystal display (LCD), a light emitting diode (LED) display, and a plasma display. In some embodiments, the display device may be a touch screen.

[0111] The above-mentioned programs (also referred to as software, software applications, or codes) include machine instructions for programmable processors, and these computer programs can be implemented using object-oriented programming languages, assembly or machine languages.

[0112] With the development of time and technology, the meaning of medium is becoming more and more extensive, and the propagation path of computer programs is no longer limited to tangible media, and can also be downloaded directly from the network, etc. Any combination of one or more computer-readable storage media can be used. Computer-readable storage media can be used but not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or devices, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above. In this document, computer-readable storage media can be any tangible medium containing or storing programs, which can be used by or in combination with instruction execution systems, devices or devices.

[0113] An embodiment of the present application further provides a computer-readable storage medium on which a computer program is stored. The program includes instructions. When the instructions are executed by one or more processors of a computing device, the steps of the method described in the aforementioned method embodiment are executed.

[0114] The embodiment of the present application also provides a computer program product. When the computer program product is run on a computer, it enables the computer to execute the steps of the method described in the aforementioned method embodiment.

[0115] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A hyperparameter space boundary analysis method, comprising: Acquire first hyperparameter data of a precision system, where the first hyperparameter data has multiple dimensions, each dimension represents a hyperparameter, and the hyperparameter is a parameter that affects an operation effect of the precision system; Performing dimensionality reduction processing on the first hyperparameter data to obtain second hyperparameter data, wherein the number of dimensions of the second hyperparameter data is smaller than the number of dimensions of the first hyperparameter data; Performing continuous function modeling for each m different dimensions in the second hyperparameter data to obtain multiple continuous function relationships, where m is an integer greater than 1, and the value of m is less than the total number of dimensions of the second hyperparameter data; Performing spatial optimization for each of the continuous function relationships to obtain an eigenvalue of the second hyperparameter data, the eigenvalue indicating the position of the point with the largest gradient descent step in the second hyperparameter space, the second hyperparameter space corresponding to the second hyperparameter data; The eigenvalues ​​of the second hyperparameter data are used to perform boundary solving and dimensionality upgrading to obtain boundary information of the first hyperparameter space. The boundary information of the first hyperparameter space is used to set or optimize the precision system. The first hyperparameter space corresponds to the first hyperparameter data.

2. The method according to claim 1, characterized in that Also includes: The second hyperparameter data is projected into an orthogonal space to complete numerical reduction processing of the second hyperparameter data, thereby reducing the data volume of the second hyperparameter data.

3. The method according to claim 1, characterized in that The first hyperparameter data includes at least one of the following hyperparameters: environmental parameters, component parameters of the precision system, motion parameters and performance parameters of the precision system.

4. The method according to claim 1, characterized in that: Performing continuous function modeling on the second hyperparameter data to obtain multiple continuous function relationships includes: Performing data analysis on the first dimensional data and the second dimensional data in the second hyperparameter data to construct a first function model, where the first function model conforms to the data distribution of the first dimensional data and the second dimensional data in the second hyperparameter data; Performing continuous function fitting on the first dimension data and the second dimension data in the second hyperparameter data according to the first function model to obtain parameter values ​​of the first function model; Substituting the parameter value of the first function model into the first function model to obtain a first continuous function relationship, wherein the first continuous function relationship represents the correlation between the first dimension and the second dimension in the second hyperparameter data; The first dimension and the second dimension are respectively any two different dimensions in the second hyper-parameter data.

5. The method according to claim 1, characterized in that The performing spatial optimization for each of the continuous function relationships to obtain the eigenvalue of the second hyperparameter data includes: Gradient descent processing is performed on each of the continuous function relationships respectively to obtain the eigenvalue corresponding to each of the continuous function relationships. The eigenvalue of a continuous function relationship is used to indicate the position of the point with the largest gradient change on the plane or subspace formed by the dimension corresponding to the continuous function relationship in the second hyper-parameter space.

6. The method according to claim 1, characterized in that The method of using the eigenvalues ​​of the second hyperparameter data to perform boundary solving and dimensionality upgrading to obtain boundary information of the first hyperparameter space includes: Using the eigenvalues ​​of the second hyperparameter data to perform boundary solving to obtain second hyperparameter space boundary information, where the second hyperparameter space boundary information includes the numerical range of each dimension in the second hyperparameter data; The boundary information of the second hyperparameter space is subjected to dimensionality upgrading processing to obtain the boundary information of the first hyperparameter space, where the boundary information of the first hyperparameter space includes the numerical range of each dimension in the first hyperparameter data.

7. The method according to claim 6, characterized in that The using the eigenvalue of the second hyper-parameter data to perform boundary solution to obtain second hyper-parameter space boundary information includes: A grid search is performed in the second hyperparameter space using the eigenvalues ​​of the second hyperparameter data to determine the numerical range of each dimension in the second hyperparameter data.

8. A hyperparameter space boundary analysis device, characterized in that: include: An acquisition unit, used to acquire first hyperparameter data of a precision system, wherein the first hyperparameter data has multiple dimensions, each dimension represents a hyperparameter, and the hyperparameter is a parameter that affects an operation effect of the precision system; A dimensionality reduction unit, configured to perform dimensionality reduction processing on the first hyperparameter data to obtain second hyperparameter data, wherein the number of dimensions of the second hyperparameter data is smaller than the number of dimensions of the first hyperparameter data; a continuous function modeling unit, configured to perform continuous function modeling for each m different dimensions in the second hyperparameter data to obtain a plurality of continuous function relationships, where m is an integer greater than 1, and the value of m is less than the total number of dimensions of the second hyperparameter data; an eigenvalue determining unit, configured to perform spatial optimization for each of the continuous function relationships to obtain an eigenvalue of the second hyperparameter data, wherein the eigenvalue indicates the position of the point with the largest gradient descent step in the second hyperparameter space, and the second hyperparameter space corresponds to the second hyperparameter data; A boundary information acquisition unit is used to use the eigenvalues ​​of the second hyperparameter data to perform boundary solution and dimensionality upgrading to obtain the boundary information of the first hyperparameter space, wherein the boundary information of the first hyperparameter space is used to set or optimize the precision system, and the first hyperparameter space corresponds to the first hyperparameter data.

9. An electronic device, characterized in that: include: Processor and memory; Wherein, the memory is connected to the processor, and the memory is used to store a computer program; The processor is configured to implement the method according to any one of claims 1 to 7 by running the computer program stored in the memory.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images