A method, device and equipment for building an automated model

Abstract

The present invention discloses an automated model building method, device and equipment, which obtains training data and extracts feature information of the training data; searches for at least one model training process information corresponding to the feature information from a model data set, wherein the model building data set includes multiple feature correspondences, and each feature correspondence includes a feature information and at least one model training process information; determines an optimization scheme for the model training process information based on a model training strategy; trains the training data according to the optimization scheme to obtain at least one second model; so that after the user inputs the training data, the complete process of machine learning can be automatically completed without the user's participation, which can reduce the time and energy consumed by engineers in the machine learning modeling process and lower the user's threshold for using artificial intelligence.

Description

Technical Field

[0001] The present application relates to the technical field of data modeling, and in particular to an automated model building method, device and equipment. Background Art

[0002] In recent years, with the rapid development of artificial intelligence (AI) technology, more and more companies or individuals are applying AI technology to various aspects of production and life to solve practical problems faced in actual production and life. However, when using AI technology to solve practical problems, AI engineers are required to conduct preliminary analysis of data and business scenarios, and manually set some or all steps of the machine learning modeling process. This is highly dependent on engineers' understanding of the business background of problem solving and their experience in the AI ​​field, and requires continuous trial and error and exhaustive enumeration, which greatly consumes engineers' time and energy. Summary of the invention

[0003] The embodiments of the present invention provide an automated model building method, device and equipment to solve the problem in the prior art that when using AI technology to solve practical problems, artificial intelligence engineers are required to perform preliminary analysis of data and business scenarios, and manually set some or all steps of the machine learning modeling process, which greatly consumes the time and energy of engineers.

[0004] In order to solve the above problems, in the first aspect, an embodiment of the present invention provides an automated model building method, including: obtaining training data and extracting feature information of the training data; searching for at least one model training process information corresponding to the feature information from a model data set, wherein the model building data set includes multiple feature correspondences, and each feature correspondence includes a feature information and at least one model training process information; determining an optimization scheme for the model training process information based on a model training strategy; and training the training data according to the optimization scheme to obtain at least one second model.

[0005] Optionally, the model training process information includes the first model, alternative algorithms, sub-processes and corresponding hyper-parameter value ranges.

[0006] Optionally, an optimization scheme for model training process information is determined based on the model training strategy, including: selecting at least one algorithm, at least one sub-process and at least one corresponding set of hyper-parameter values ​​from alternative algorithms, sub-processes and corresponding hyper-parameter value ranges according to the model training strategy and the first model; and determining an optimization scheme for model training process information based on at least one algorithm, at least one sub-process and at least one corresponding set of hyper-parameter values.

[0007] Optionally, the training data includes training samples and verification samples; the training data is trained according to the optimization scheme to obtain at least one second model, including: training the training samples according to at least one algorithm, at least one sub-process and at least one corresponding set of hyperparameter values ​​to obtain at least one second model.

[0008] Optionally, after obtaining at least one second model, the automated model building method further includes: inputting a verification sample into at least one second model to obtain a corresponding first output result; determining a second model with the best performance from at least one second model based on the first output result; and training the training samples and verification samples according to the algorithm, sub-process and corresponding hyper-parameter values ​​of the second model with the best performance to obtain a third model.

[0009] Optionally, after obtaining the third model, the automated model building method further includes: acquiring test data; inputting the test data into the third model to obtain a corresponding second output result; and performing performance evaluation on the third model according to the second output result.

[0010] Optionally, after obtaining the training data and before determining an optimization scheme for the model training process information based on the model training strategy, the automated model building method further includes: obtaining at least part of the model training process information input by the user; and merging at least part of the model training process information input by the user with at least one model training process information.

[0011] Optionally, after obtaining at least one second model, the automated model building method further includes: recording feature information and model training process information corresponding to the at least one second model.

[0012] In the second aspect, an embodiment of the present invention provides an automated model building device, comprising: an acquisition unit, used to acquire training data and extract feature information of the training data; a search unit, used to search for at least one model training process information corresponding to the feature information from a model building data set, the model building data set including multiple feature correspondences, each feature correspondence including at least feature information and model training process information; a determination unit, used to determine an optimization scheme for the model training process information based on a model training strategy; and a training unit, used to train the training data according to the optimization scheme to obtain at least one second model.

[0013] In a third aspect, an embodiment of the present invention provides an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor executes the automated model building method as in the first aspect or any embodiment of the first aspect.

[0014] The automated model building method, apparatus and device provided by the embodiments of the present invention obtain feature information of training data by acquiring training data; search for at least one model training process information corresponding to the feature information from the model data set, wherein the model building data set includes multiple feature correspondences, and each feature correspondence includes a feature information and at least one model training process information; determine an optimization scheme for the model training process information based on the model training strategy; train the training data according to the optimization scheme to obtain at least one second model; so that when the user inputs the training data, the model training process information corresponding to the training data can be matched through the feature information of the training data, and then the training data can be trained based on the optimization scheme of the model training process information to obtain at least one second model, so that the complete process of machine learning can be automatically completed without the participation of the user, and at least one second model corresponding to the training data can be obtained, which can reduce the time and energy consumed by engineers in the machine learning modeling process and lower the user's threshold for using artificial intelligence.

[0015] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic diagram of a flow chart of an automated model building method in an embodiment of the present invention;

[0017] Figure 2 A schematic diagram of a flow chart of an optimization scheme for determining model training process information based on a model training strategy in an embodiment of the present invention;

[0018] Figure 3 A schematic diagram of another automatic model building method according to an embodiment of the present invention;

[0019] Figure 4 A schematic diagram of another automatic model building method according to an embodiment of the present invention;

[0020] Figure 5 A schematic diagram of another automatic model building method according to an embodiment of the present invention;

[0021] Figure 6 A schematic diagram of another automatic model building method according to an embodiment of the present invention;

[0022] Figure 7 A schematic diagram of another automatic model building method according to an embodiment of the present invention;

[0023] Figure 8It is a structural schematic diagram of an automatic model building device in an embodiment of the present invention;

[0024] Fig. 9 The figure is a schematic diagram of the hardware structure of an electronic device in an embodiment of the present invention. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0026] The embodiment of the present invention provides an automated model building method, which can be applied to any application scenario that uses artificial intelligence technology to analyze data, for example, an application scenario that uses artificial intelligence technology to analyze petrochemical industry data, an application scenario that uses artificial intelligence technology to analyze financial data, an application scenario that uses artificial intelligence technology to analyze image data, etc. Figure 1 As shown, the automated model building method includes:

[0027] S101. Obtain training data and extract feature information of the training data; specifically, users can select training data according to the needs of life and production for training models. For example, when users need to analyze financial data, the training data can be selected as specific financial data, and then feature information of the training data can be extracted. Feature information is used to characterize the features of the training data, and the feature information may include indicators such as the number of rows, the number of features, the number of discrete data types, the number of continuous data types, and the skewness and kurtosis of labels.

[0028] S102. Search at least one model training process information corresponding to the feature information from the model data set, the model construction data set includes multiple feature correspondences, each feature correspondence includes a feature information and at least one model training process information; specifically, the historical model training process information can be recorded, and the data features corresponding to the model training process information can be recorded, so as to form multiple feature correspondences and obtain the model data set. When the feature information of the training data is extracted, at least one model training process information corresponding to the feature information can be found from the model data set. The model training process information is the information recorded during the model training process, which may include the model structure optimization process, some representative excellent model structures, and the model structure finally adopted.

[0029] S103. Determine an optimization scheme for the model training process information based on the model training strategy; specifically, the model training strategy includes Bayesian optimization. The model training strategy may be included in the model training process information or may be preset. Based on the selected model training strategy, at least one set of data consisting of sub-processes, algorithms, and hyper-parameters required in the model training process may be selected from the model training process information, thereby obtaining an optimization scheme for the model training process information.

[0030] S104. Train the training data according to the optimization scheme to obtain at least one second model. Specifically, the training data can be trained based on at least one set of data consisting of sub-processes, algorithms and hyperparameters required in the model training process to obtain at least one second model.

[0031] The automated model building method provided by the embodiment of the present invention obtains training data and extracts feature information of the training data; searches for at least one model training process information corresponding to the feature information from the model data set, wherein the model building data set includes multiple feature correspondences, and each feature correspondence includes a feature information and at least one model training process information; determines an optimization scheme for the model training process information based on the model training strategy; trains the training data according to the optimization scheme to obtain at least one second model; so that when the user inputs the training data, the model training process information corresponding to the training data can be matched through the feature information of the training data, and then the training data can be trained based on the optimization scheme of the model training process information to obtain at least one second model, so that the complete process of machine learning can be automatically completed without the participation of the user, and at least one second model corresponding to the training data can be obtained, which can reduce the time and energy consumed by engineers in the machine learning modeling process and lower the user's threshold for using artificial intelligence.

[0032] In an optional embodiment, the model training process information includes a first model, an alternative algorithm, a sub-process and a corresponding hyper-parameter value range.

[0033] Specifically, the first model is the model structure finally adopted by the historical model training. The sub-process is each process in the model training process. The alternative algorithm is at least one algorithm corresponding to the sub-process. The hyperparameter value range is the value range of the parameter of the algorithm corresponding to the sub-process.

[0034] In an embodiment of the present invention, by recording the first model, alternative algorithms, sub-processes and corresponding hyper-parameter value ranges in the model training process information, after matching the model training process information according to the feature information of the training data, the training plan of the training data can be obtained according to the first model, alternative algorithms, sub-processes and corresponding hyper-parameter value ranges, thereby realizing the automated machine learning process of the training data.

[0035] In an optional embodiment, step S103, an optimization scheme for model training process information is determined based on the model training strategy, such as Figure 2 As shown, it specifically includes: S1031. Select at least one algorithm, at least one sub-process and at least one corresponding set of hyper-parameter values ​​from the alternative algorithms, sub-processes and corresponding hyper-parameter value ranges according to the model training strategy and the first model; S1032. Determine the optimization plan for the model training process information according to at least one algorithm, at least one sub-process and at least one corresponding set of hyper-parameter values.

[0036] For example, when the model training strategy is Bayesian optimization, at least one sub-process can be selected from the sub-processes, and then at least one algorithm can be selected for each sub-process from the alternative algorithms corresponding to the sub-processes, and then at least one set of hyper-parameter values ​​can be selected from the hyper-parameter value range. Based on the selected at least one algorithm, at least one sub-process and the corresponding at least one set of hyper-parameter values, an optimization solution for the model training process information can be obtained. When the model training strategy is Bayesian optimization based on the initial model structure, at least one sub-process and at least one algorithm that are the same as the sub-process and algorithm of the first model structure can be selected from the alternative algorithms and sub-processes, and at least one set of hyper-parameter values ​​can be selected from the hyper-parameter value range. Based on the one algorithm, at least one sub-process and the selected at least one set of hyper-parameter values, an optimization solution for the model training process information can be obtained.

[0037] In an embodiment of the present invention, by determining at least one algorithm, at least one sub-process and at least one corresponding set of hyper-parameter values ​​in the model training process, that is, an optimization scheme for the model training process information, the training data can be trained according to the at least one algorithm, at least one sub-process and at least one corresponding set of hyper-parameter values, and at least one second model corresponding to the training data can be automatically obtained.

[0038] In an optional embodiment, the training data includes training samples and verification samples; step S104, training the training data according to the optimization scheme to obtain at least one second model, such as Figure 3 As shown, it includes: S1041. Training samples are trained according to at least one algorithm, at least one sub-process and at least one corresponding set of hyper-parameter values ​​to obtain at least one second model.

[0039] Specifically, when the user inputs training data, the training data can be grouped, and a first preset percentage of the training data can be used as training samples to train at least one second model, and then the remaining training data can be used as verification samples to test the at least one second model, so that the performance of each second model can be tested, and a model structure with the best performance can be selected from the second models, and each sub-process and the strategy for analyzing data for each sub-process can be determined based on the model structure with the best performance.

[0040] In an optional embodiment, in step S1041, after obtaining at least one second model, the automated model building method is as follows: Figure 4 As shown, it also includes: S105. Inputting the verification sample into at least one second model to obtain a corresponding first output result; determining a second model with the best performance from at least one second model according to the first output result; training the training samples and the verification samples according to the algorithm, sub-process and corresponding hyper-parameter values ​​of the second model with the best performance to obtain a third model.

[0041] Specifically, when testing each second model, the verification sample can be input into each second model respectively, so that the corresponding first output result can be obtained. According to the first output result, the performance of each second model can be determined, so that the second model with the best performance can be determined from each second model. Since the second model is trained with a preset percentage of training samples, and the training data input by the user also includes verification samples, in order to optimize the second model structure with the best performance and obtain the model structure that the user ultimately wants, the training samples and verification samples can be retrained according to the algorithm, sub-process and corresponding hyperparameter values ​​of the second model with the best performance to obtain the third model.

[0042] In an optional embodiment, in step S105, after obtaining the third model, the automated model building method is as follows: Figure 5 As shown, the method further includes: S106, obtaining test data; inputting the test data into the third model to obtain a corresponding second output result; and performing performance evaluation on the third model according to the second output result.

[0043] After training based on the training data to obtain the third model, the performance of the third model can be tested using the test data to evaluate the performance of the third model. When the performance of the third model is tested, the test data can be input into the third model, and the performance of the third model can be evaluated based on the second output result of the third model. For example, if the second output result shows that the test accuracy of the test data is above 80%, then the performance of the third model is good and can be used for data analysis.

[0044] By performing performance evaluation on the third model, the performance of the third model can be tested, thereby determining whether the third model can be used for data analysis in actual production and life.

[0045] In an optional embodiment, after obtaining training data in step S101, and before determining an optimization scheme for model training process information based on a model training strategy in step S103, an automated model building method is as follows: Figure 6As shown, it also includes: S107. Obtaining at least part of the model training process information input by the user; merging at least part of the model training process information input by the user with at least one model training process information.

[0046] Specifically, the user can control or adjust the model building process according to the actual situation. For example, manually setting at least part of the model training process information can be manually setting the model training strategy, setting the initial point of model building, alternative algorithms, sub-processes or hyper-parameter value ranges, etc. When the user manually sets part of the model training process information, after determining at least one model training process information based on the training data, the part of the model training process information manually set by the user can be merged with at least one model training process information. When merging, for the different parts, the part of the model training process information manually set by the user is used as the main one.

[0047] In an optional embodiment, after obtaining at least one second model, the automated model building method further includes: recording feature information and model training process information corresponding to the at least one second model.

[0048] Specifically, recording the feature information and model training process information corresponding to the at least one second model can provide model training process information for other data corresponding to the feature information in the future, so that other data can be trained according to the model training process information, thereby realizing an automated machine learning process for other data.

[0049] In order to further illustrate the automated model building method of the present invention, the automated model building in the petrochemical industry will be taken as an example in the following description.

[0050] In the petrochemical industry, it is hoped that the yield of key products under different conditions can be predicted by analyzing the historical data of the catalytic cracking unit. The embodiment of the present invention provides a method for constructing a machine learning model for predicting the yield of key products of the catalytic cracking unit based on the historical data of the catalytic cracking unit and the process data and material property data of the catalytic cracking unit, and establishing a fitting relationship between the process indicators of the unit and the yield of key products. The automated model construction process is as follows: Figure 7 As shown, including:

[0051] S201: inputting training data such as device process data, material property data and product yield in the catalytic cracking system, and setting at least part of the model training process information;

[0052] The user needs to input training data (training samples and validation samples, where training samples must be input and validation samples are optional), where product yield is the label. Since the product yield is a continuous value, the model is a model for solving regression problems. The input training data (training samples, validation samples) and the output result data are both tabular data, and the data format can be csv.

[0053] Users can also control the model building process according to actual conditions. Here, it is assumed that the user settings are as follows: the model structure building time is 30 minutes, the model training strategy is Bayesian optimization, the optimization space uses the preset medium-sized optimization space, and the evaluation indicator uses the mean square error (MSE). Other parameters use the default values.

[0054] The optimization space includes the following:

[0055] 1. Sub-process: feature engineering, regression;

[0056] 2. Alternative algorithms for feature engineering, or combinations of feature engineering algorithms (Note: In order to ensure that the regression algorithm can be used for the data after feature engineering, in actual operation, some feature engineering algorithms may be packaged)

[0057] 3. Alternative regression algorithms, such as linear regression, support vector machine (SVM), etc.

[0058] 4. The range of hyperparameter values ​​corresponding to different algorithms (including feature engineering and alternative regression algorithms). For example, the possible hyperparameters of linear regression are: whether there is an intercept. Here, a certain algorithm may correspond to multiple hyperparameters, and each hyperparameter value may be an integer, real number, discrete value, Boolean value (yes or no), and other data types.

[0059] S202: extracting characteristic information of training data such as device process data, material property data and product yield in the catalytic cracking system, referencing the model construction data set, and determining an optimization scheme for the model training process information;

[0060] The characteristic information of the training data such as the process data of the device, material property data and product yield in the catalytic cracking system includes: the number of rows of data, the number of features, the number of discrete type data, the number of continuous types, the skewness and kurtosis of the label, etc. After extracting these characteristic information, these characteristic information can be used as a basis to find the data in the model construction data set that is most similar to the characteristic information. Finding the most similar data means that the current training data is similar to the most similar data, and a similar model construction process can be used. The model construction data set will save the model training process information corresponding to the most similar data. Such as: model construction time, model training strategy, model structure (first model) finally adopted by model training, alternative algorithms, sub-processes, hyper-parameter value ranges, etc.

[0061] In the embodiment of the present invention, since the model training strategy, sub-process, alternative algorithm, and hyper-parameter value range have been set in advance by the user, the model training strategy, sub-process, alternative algorithm, and hyper-parameter value range are selected as set by the user. And the first model is selected as the initial point of Bayesian optimization. Based on Bayesian optimization, at least one sub-process, at least one algorithm, and at least one set of hyper-parameter values ​​are selected from the alternative algorithms, sub-processes, and hyper-parameter value ranges to obtain an optimization scheme for the model training process information.

[0062] S203: Using machine learning technology to construct a model structure to obtain at least one second model;

[0063] Based on the at least one sub-process, the at least one algorithm and the at least one set of hyperparameter values, the training samples are trained to obtain at least one second model.

[0064] S204: Determine a second model with the best performance;

[0065] In order to judge the quality of the second model structure, the verification samples are input into each second model structure respectively to evaluate the quality of each second model structure. Finally, the second model with the best performance is determined. The second model with the best performance includes feature engineering algorithm, regression algorithm and corresponding hyperparameter values.

[0066] S205: Based on the second model with the best performance obtained in the fourth step, training is performed using the complete training data to obtain a third model;

[0067] Based on the feature engineering algorithm, regression algorithm and corresponding hyperparameter values ​​corresponding to the second model with the best performance obtained in the third step, the third model is trained with complete training data (training samples + verification samples). For the trained third model, the test data can be used to evaluate the model. The interpretability module of the third model can also be run to try to analyze the trained third model. In an embodiment of the present invention, if the feature information extracted from the training data is different from all records in the database, the record is saved. If the feature information extracted from the training data is the same as a record in the database, a record with a better final result and a better evaluation index is saved.

[0068] The third model can be used to: 1) analyze the impact of different device indicators on the yield of key products; 2) predict the device yield; 3) guide the optimization of the device process.

[0069] The embodiment of the present invention also provides an automated model building device, such as Figure 8As shown, it includes: an acquisition unit 21, which is used to acquire training data and extract feature information of the training data; the specific implementation method is described in step S101 of the above method embodiment, which will not be repeated here. A search unit 22, which is used to find at least one model training process information corresponding to the feature information from the model construction data set, and the model construction data set includes multiple feature correspondences, and each feature correspondence includes at least feature information and model training process information; the specific implementation method is described in step S102 of the above method embodiment, which will not be repeated here. A determination unit 23, which is used to determine an optimization scheme for the model training process information based on the model training strategy; the specific implementation method is described in step S103 of the above method embodiment, which will not be repeated here. A training unit 24, which is used to train the training data according to the optimization scheme to obtain at least one second model. The specific implementation method is described in step S104 of the above method embodiment, which will not be repeated here.

[0070] The automated model building device provided by the embodiment of the present invention obtains the feature information of the training data by acquiring the training data; searches for at least one model training process information corresponding to the feature information from the model data set, wherein the model building data set includes multiple feature correspondences, and each feature correspondence includes a feature information and at least one model training process information; determines an optimization scheme for the model training process information based on the model training strategy; trains the training data according to the optimization scheme to obtain at least one second model; so that when the user inputs the training data, the model training process information corresponding to the training data can be matched through the feature information of the training data, and then the training data can be trained based on the optimization scheme of the model training process information to obtain at least one second model, so that the complete process of machine learning can be automatically completed without the participation of the user, and at least one second model corresponding to the training data can be obtained, which can reduce the time and energy consumed by engineers in the machine learning modeling process and lower the user's threshold for using artificial intelligence.

[0071] Based on the same inventive concept as the automatic model building method in the aforementioned embodiment, the present invention also provides an electronic device, such as Fig. 9 As shown, it includes: a processor 31 and a memory 32, wherein the processor 31 and the memory 32 can be connected via a bus or other means. Fig. 9 The bus connection is used as an example for explanation.

[0072] The processor 31 may be a central processing unit (CPU). The processor 31 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or a combination of the above chips.

[0073] The memory 32 is a non-transitory computer-readable storage medium that can be used to store non-transitory software programs, non-transitory computer executable programs and modules, such as program instructions / modules corresponding to the automated model building method in the embodiment of the present invention. The processor 31 executes various functional applications and data processing of the processor by running the non-transitory software programs, instructions and modules stored in the memory 32, that is, the automated model building method in the above method embodiment is implemented.

[0074] The memory 32 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and applications required for at least one function; the data storage area may store data created by the processor 31, etc. In addition, the memory 32 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage devices. In some embodiments, the memory 32 may optionally include a memory remotely arranged relative to the processor 31, and these remote memories may be connected to the processor 31 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0075] The above one or more modules are stored in the memory 32, and when executed by the processor 31, the following is performed: Figure 1 The automated model building method in the illustrated embodiment.

[0076] For details of the above electronic equipment, please refer to Figure 1 The corresponding related descriptions and effects in the illustrated embodiments can be understood and will not be repeated here.

[0077] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the storage medium can be a disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), a flash memory, a hard disk drive (HDD) or a solid-state drive (SSD), etc.; the storage medium can also include a combination of the above-mentioned types of memory.

[0078] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0079] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable information processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable information processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0080] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable information processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0081] These computer program instructions can also be loaded onto a computer or other programmable information processing device so that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0082] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. An automated model building method, applied to an application scenario where artificial intelligence technology is used to analyze image data, the method comprising: Acquire training data, and extract feature information of the training data; The training data includes image data; the feature information includes: the number of rows of data, the number of features, the number of discrete type data, the number of continuous type data, the skewness and kurtosis of the label; Searching for at least one model training process information corresponding to the feature information from a model building data set, wherein the model building data set includes a plurality of feature correspondences, each feature correspondence includes a feature information and at least one model training process information; the model training process information includes a first model, an alternative algorithm, a sub-process and a corresponding hyper-parameter value range; the first model is a model structure finally adopted by the historical model training; Determining an optimization scheme for model training process information based on a model training strategy includes: selecting at least one algorithm, at least one sub-process, and at least one corresponding set of hyper-parameter values ​​from candidate algorithms, sub-processes, and corresponding hyper-parameter value ranges according to the model training strategy and the first model; determining an optimization scheme for model training process information according to the at least one algorithm, at least one sub-process, and at least one corresponding set of hyper-parameter values; The training data is trained according to the optimization scheme to obtain at least one second model.

2. According to the automated model building method of claim 1, the training data comprises a training sample and a validation sample; The step of training the training data according to the optimization scheme to obtain at least one second model comprises: The training samples are trained according to the at least one algorithm, the at least one sub-process and the corresponding at least one set of hyperparameter values ​​to obtain at least one second model.

3. The automated model building method according to claim 2, after obtaining at least one second model, further comprising: Inputting the verification sample into the at least one second model to obtain a corresponding first output result; Determine a second model with the best performance from the at least one second model according to the first output result; The training samples and the validation samples are trained according to the algorithm, sub-process and corresponding hyper-parameter values ​​of the second model with the best performance to obtain a third model.

4. The automated model building method according to claim 3, after obtaining the third model, further comprising: Get test data; Inputting the test data into the third model to obtain a corresponding second output result; A performance evaluation is performed on the third model according to the second output result.

5. The automated model building method according to claim 1, after acquiring the training data and before determining an optimization scheme for the model training process information based on the model training strategy, further comprises: Obtain at least part of the model training process information input by the user; At least part of the model training process information input by the user is merged with the at least one model training process information.

6. The automated model building method according to claim 1, after obtaining at least one second model, further comprising: Record feature information and model training process information corresponding to the at least one second model.

7. An automated model building device, applied to an application scenario where artificial intelligence technology is used to analyze image data, the device comprising: An acquisition unit, used to acquire training data and extract feature information of the training data; The training data includes image data; the feature information includes: the number of rows of data, the number of features, the number of discrete type data, the number of continuous type data, the skewness and kurtosis of the label; A search unit is used to search for at least one model training process information corresponding to the feature information from a model construction data set, wherein the model construction data set includes a plurality of feature correspondences, each feature correspondence includes at least feature information and model training process information; the model training process information includes a first model, an alternative algorithm, a sub-process and a corresponding hyper-parameter value range; the first model is a model structure finally adopted by the historical model training; A determination unit, used to determine an optimization scheme for model training process information based on a model training strategy, including: selecting at least one algorithm, at least one sub-process and at least one corresponding set of hyper-parameter values ​​from candidate algorithms, sub-processes and corresponding hyper-parameter value ranges according to the model training strategy and the first model; determining an optimization scheme for model training process information according to the at least one algorithm, at least one sub-process and at least one corresponding set of hyper-parameter values; A training unit is used to train the training data according to the optimization scheme to obtain at least one second model.

8. An electronic device, characterized in that: include: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor executes the automated model building method as described in any one of claims 1-6.

Images