A machine learning model training and verification visualization arrangement method, device, equipment and medium

Abstract

The present application discloses a visual arrangement method, device, equipment and medium for machine learning model training and verification, and relates to the field of computer technology. It includes moving the functional unit node from the node library to the workbench in the visual workflow orchestrator, and initializing the node configuration of the functional unit node; connecting the functional unit nodes on the workbench in series through connecting lines, and creating corresponding data queues according to the node connection order to obtain a visual process; adjusting the node configuration of the functional unit node through the node configuration interface in the visual workflow orchestrator to obtain the target functional node; if the execution instruction is triggered, the original data is input into the visual process, and the target functional node is run to obtain the visual output data. In this way, the whole process management of machine learning model training and verification can be realized, the development threshold can be reduced, and the compatibility of model deployment can be improved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method, device, equipment and medium for visual arrangement of machine learning model training and verification. Background Art

[0002] The training and verification of machine learning models are becoming increasingly popular in many fields, especially in scenarios with high demand for data analysis, such as industry, finance, and medical care. However, existing model training and verification tools generally rely on code implementation and do not have intuitive visualization functions. This increases the difficulty for model developers to understand and operate. Before model training and verification, there are many steps in data processing, and the data preprocessing, missing value processing, feature engineering and other processes in the existing system are complex and difficult to configure modularly. Reasonable data partitioning and sampling strategies are required in model training to improve the representativeness of training data, but the partitioning strategies provided in existing tools are relatively simple and cannot meet the needs of complex application scenarios. Existing tools rely on professional codes for model verification and cannot provide interactive and visual verification and evaluation methods, which limits the intuitive analysis of model effects and is not conducive to quickly understanding the advantages and disadvantages of the model and the direction of improvement. Existing tools mostly rely on specific formats or development environments, lack a universal and standard model storage method, and are difficult to support the seamless migration and deployment of models in different systems and language environments, hindering the application and promotion of models between different platforms.

[0003] From the above, it can be seen that how to achieve full-process management of machine learning model training and verification, reduce development barriers, and improve the compatibility of model deployment is an urgent problem to be solved. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide a method, device, equipment and medium for visual arrangement of machine learning model training and verification, which can realize the whole process management of machine learning model training and verification, reduce the development threshold, and improve the compatibility of model deployment. The specific scheme is as follows:

[0005] In a first aspect, the present application provides a method for visually arranging machine learning model training and verification, which is applied to a visual workflow orchestrator, comprising:

[0006] Move the function unit node from the node library to the workbench in the visual workflow choreographer, and initialize the node configuration of the function unit node;

[0007] Connecting the functional unit nodes on the workbench in series through connecting lines, and creating corresponding data queues according to the node connection sequence to obtain a visual process; the data queue can generate and transmit new data according to the update of the node configuration of the functional unit node;

[0008] Adjusting the node configuration of the functional unit node through the node configuration interface in the visual workflow orchestrator to obtain a target functional node; the target functional node includes a data reading node, a data partitioning node, a data processing node, a preset training node, a preset format writing node and a model prediction node;

[0009] If the execution instruction is triggered, the original data is input into the visualization process, and the target function node is run to obtain visualization output data.

[0010] Optionally, the visual workflow orchestrator performs visual process orchestration through a workflow design interface;

[0011] Among them, the workflow design interface includes the node library, the workbench, the node configuration interface, a monitor for real-time monitoring of the operating status of each functional unit node on the workbench, and a data display area for intuitively displaying data through locally pre-integrated preset visualization components.

[0012] Optionally, adjusting the node configuration of the functional unit node through the node configuration interface to obtain a target functional node includes:

[0013] Determine standardized node configuration rules through a preset interface; the node configuration rules are rules for configuring attribute definitions, data input and output structures, verification rules, and state management methods of functional unit nodes;

[0014] Adjusting the node configuration of the functional unit node based on the node configuration rule through the preset interface to obtain an updated node configuration;

[0015] The updated node configuration is verified, and data in a data queue corresponding to the functional unit node is cleared to obtain a target functional node.

[0016] Optionally, if the execution instruction is triggered, the original data is input into the visualization process, and the target function node is run to obtain visualization output data, including:

[0017] If the execution instruction is triggered, the visualization process is verified based on the node connection sequence, and invalid nodes are removed from the verification result to complete the initialization operation of the visualization process;

[0018] Inputting raw data into the initialized visualization process, and processing the raw data through the target function node to obtain processed data;

[0019] The processed data is cached, and visual output data is generated according to the features of the processed data.

[0020] Optionally, after connecting the functional unit nodes on the workbench in series through connecting lines and creating corresponding data queues according to the node connection sequence to obtain a visualized process, the method further includes:

[0021] The visualization process is saved based on a preset format to obtain a visualization process file, so that a third party can run the visualization process through the visualization process file.

[0022] Optionally, the process of training and validating the machine learning model includes:

[0023] The data reading node obtains the original data set generated by the data partitioning node according to the data partitioning strategy, and the data processing node processes the original data set to obtain the target data set;

[0024] Using the target data set as input data of a preset training node, and configuring the preset training node based on the input data, so as to obtain a target model by running the preset training node;

[0025] Based on the target model, a target model file is written into the node in a preset format so that the model prediction node can verify the effect of the target model through the target model file.

[0026] Optionally, the process of generating the original data set by the data partitioning node according to the data partitioning strategy includes:

[0027] Extract data from the original input data based on the data partition ratio and the data sampling strategy to generate an original data set;

[0028] The data sampling strategy includes any one or more of a simple sampling strategy, a random sampling strategy, a stratified sampling strategy, a linear sampling strategy and a time series sampling strategy.

[0029] In a second aspect, the present application provides a machine learning model training and verification visual orchestration device, which is applied to a visual workflow orchestrator, including:

[0030] A node configuration initialization module, used to move the function unit node from the node library to the workbench in the visual workflow orchestrator, and initialize the node configuration of the function unit node;

[0031] A visualization process creation module, used to connect the functional unit nodes on the workbench in series through connecting lines, and create corresponding data queues according to the node connection sequence to obtain a visualization process; the data queue can generate and transmit new data according to the update of the node configuration of the functional unit node;

[0032] A target function node generation module, used to adjust the node configuration of the function unit node through the node configuration interface in the visual workflow orchestrator to obtain a target function node; the target function node includes a data reading node, a data partitioning node, a data processing node, a preset training node, a preset format writing node and a model prediction node;

[0033] The visualization process execution module is used to input the original data into the visualization process and run the target function node to obtain visualization output data if the execution instruction is triggered.

[0034] In a third aspect, the present application provides an electronic device, including:

[0035] Memory, used to store computer programs;

[0036] A processor is used to execute the computer program to implement the aforementioned machine learning model training and verification visualization orchestration method.

[0037] In a fourth aspect, the present application provides a computer-readable storage medium for storing a computer program, wherein the computer program, when executed by a processor, implements the aforementioned machine learning model training and verification visualization orchestration method.

[0038] The present application provides a method for visual orchestration of machine learning model training and verification, which comprises the following steps: first, moving a functional unit node from a node library to a workbench in the visual workflow orchestrator, and initializing the node configuration of the functional unit node; then, connecting the functional unit nodes on the workbench in series through connecting lines, and creating corresponding data queues in the order of node connection to obtain a visual process; the data queue can generate and transmit new data according to the update of the node configuration of the functional unit node; then, adjusting the node configuration of the functional unit node through the node configuration interface in the visual workflow orchestrator to obtain a target functional node; the target functional node includes a data reading node, a data partitioning node, a data processing node, a preset training node, a preset format writing node, and a model prediction node; finally, if an execution instruction is triggered, the original data is input into the visual process, and the target functional node is run to obtain visual output data.

[0039] As can be seen from the above, this application configures the functional nodes based on the input data, and can integrate custom nodes according to needs to implement specific functions or models, thereby enhancing the flexibility of the platform and meeting complex and diverse analysis and processing needs; completing machine learning model training and verification through several functional nodes, and completing data feature extraction and model verification operations without writing code. This enables the full process management of machine learning model training and verification, reduces development barriers, and improves the compatibility of model deployment. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0041] Figure 1 A flowchart of a method for visual choreography and verification of a machine learning model applied to a visual workflow choreographer disclosed in the present application;

[0042] Figure 2 A schematic diagram of a method for constructing a visual orchestrator for training and verifying a machine learning model disclosed in this application;

[0043] Figure 3 A schematic diagram of a workflow design interface architecture disclosed in this application;

[0044] Figure 4 A schematic diagram of a model training and verification architecture disclosed in this application;

[0045] Figure 5 A schematic diagram of a data partition node architecture disclosed in this application;

[0046] Figure 6 A flowchart of a specific machine learning model training and verification visual orchestration method for a visual workflow orchestrator disclosed in this application;

[0047] Figure 7 A schematic diagram of a machine learning model training and verification visual orchestration device for a visual workflow orchestrator disclosed in the present application;

[0048] Figure 8 This is a structural diagram of an electronic device disclosed in this application. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] At present, the training and verification of machine learning models are gradually becoming popular in multi-field applications, especially in scenarios with high demand for data analysis such as industry, finance, and medical care. However, most traditional model building processes rely on programming, resulting in a high development threshold, and the requirements for code writing, data processing, and model configuration are relatively complex, which limits the participation of non-professionals. In addition, existing model building tools often only cover a single link of training or verification, lack comprehensive management of the entire model life cycle, and make it difficult to efficiently connect the process from data preparation to model verification. Therefore, the present application provides a visual orchestration solution for machine learning model training and verification, which can realize the full process management of machine learning model training and verification, reduce the development threshold, and improve the compatibility of model deployment.

[0051] See also Figure 1 As shown, the embodiment of the present application discloses a machine learning model training and verification visual orchestration method applied to a visual workflow orchestrator, including:

[0052] Step S11: Move the function unit node from the node library to the workbench in the visual workflow orchestrator, and initialize the node configuration of the function unit node.

[0053] In this embodiment, the functional unit node is moved from the node library to the workbench of the workflow design interface in the visual workflow orchestrator, and the node configuration of the functional unit node is initialized. The visual workflow orchestrator decomposes large functions into specific functional nodes, abstracts complex analysis processes into software processes, and evolves a visual workflow orchestrator. Users can use drag and click operations to select nodes, configure parameters through the configuration panel, organize the execution order through connections, view the node execution status through monitoring, view data results through visual output, and analyze the accuracy of the model. Workflows in different scenarios can be constructed based on visualization to solve modeling problems in different scenarios. See Figure 2 As shown in the figure, the visual workflow choreographer integrates functional modules such as data preprocessing, analysis, model training, and model verification, and flexibly combines different functional nodes into task processes through visual task arrangement. The processing results of each node can be fed back in real time, and users can perform operation verification, parameter tuning, etc., thereby realizing the full process automation of data preprocessing, analysis, model training, evaluation, and result display.

[0054] For further information, see Figure 3As shown, the visual workflow choreographer performs visual process choreography through a workflow design interface; the workflow design interface includes the node library, the workbench, the node configuration interface, a monitor for real-time monitoring of the operating status of each functional unit node on the workbench, and a data display area for intuitively displaying data through locally pre-integrated preset visual components. Among them, the node library is used to store nodes as independent functional modules (such as data input, analysis, and output nodes), and package different functions through a plug-in mechanism to make the node library extensible; the node library allows the development of custom nodes, and manages the loading and registration of nodes through APIs, which is convenient for dynamic expansion and updating. The UI layer of the workbench adopts the MVC (Model-View-Controller) architecture, which separates the representation (view), logical processing (controller), and underlying data (model) of the workflow to achieve visual drag-and-drop operations; the workbench responds to the user's drag and connection operations through event-driven, and instantly updates the graphical display of the workflow, and supports the saving, loading and version control of the workflow, which is convenient for the management of complex processes. The UI of the node configuration interface is separated from the specific computing logic through the node model, so that the configuration interface is reusable and easy to maintain; the input component of the node configuration interface is dynamically bound to the node parameters to ensure that the parameter modification takes effect immediately; the node configuration interface automatically performs parameter verification when the user configures the node, provides instant feedback, and ensures the legality of the configuration. The monitor uses the task management module to monitor the running status of the node in real time, including task progress, error status and resource usage, and integrates logging and notification functions to facilitate tracking of node running status and error information; the monitor collects running information in real time through the system interface and displays it on the user interface to achieve real-time monitoring of workflows and nodes. The data display area realizes intuitive display of data by integrating visual components (such as tables, charts, dashboards, etc.). When the node is completed, the data display area quickly loads and displays the data output by the node through the data flow cache management system; the data display area allows users to perform interactive analysis operations (such as filtering, sorting, aggregation) on the data to improve the flexibility of data exploration.

[0055] Step S12: connect the functional unit nodes on the workbench in series through connecting lines, and create corresponding data queues according to the node connection sequence to obtain a visualized process.

[0056] In this embodiment, each functional unit node on the workbench is connected in series through a connecting line, that is, each functional unit node represents a functional module or processing step, and the connecting line is used to clarify the data transmission path. When two functional unit nodes establish a connection, the background will create a last-in-first-out data queue for them to store and transmit data; the data queue can generate new data and transmit it according to the update of the node configuration of the functional unit node. When the node configuration or node connection information of the functional unit node changes, the background will automatically clean up the old data in the data queue, and provide a data queue reset operation to manually clear the data generated by the old configuration in the queue. When the connecting line or functional unit node is deleted, the system will automatically clear the data in the data queue and destroy the data queue to release resources. When the connection between functional unit nodes changes, the background will also record in detail the connection relationship between each node, the specific position of the functional unit node on the canvas, the configuration information of each functional unit node, etc., for complete restoration of the layout when reloading; at the same time, it is convenient for debugging and troubleshooting, and the problem can be quickly located through clear connection relationships.

[0057] Step S13: adjusting the node configuration of the functional unit node through the node configuration interface in the visual workflow orchestrator to obtain a target functional node.

[0058] In this embodiment, the processing logic and behavior of the functional unit node are controlled by configuration, thereby affecting the execution effect of the entire functional unit node. The configuration items are defined by the background attributes and parameters, and the front-end visualization and interactive settings are realized through dynamic rendering, so that the target functional node can flexibly adapt to different scenarios, customize parameters and rules, and meet specific business needs. It is worth mentioning that in the process of machine learning model training and verification, the target functional node includes a data reading node, a data partitioning node, a data processing node, a preset training node, a preset format writing node, and a model prediction node. Specifically, the node configuration of the functional unit node is adjusted through the node configuration interface to obtain the target functional node, which may include: determining a standardized node configuration rule through a preset interface; the node configuration rule is a rule for configuring the attribute definition, data input and output structure, verification rules, and state management method of the functional unit node; adjusting the node configuration of the functional unit node based on the node configuration rule through the preset interface to obtain an updated node configuration; verifying the updated node configuration, and clearing the data in the data queue corresponding to the functional unit node to obtain the target functional node. That is, through a unified interface, standardized attribute definitions, data input and output structures, verification rules and status management methods, flexible dynamic configuration is achieved and the legality of node configuration input is ensured; by verifying the updated node configuration and clearing the data in the data queue corresponding to the functional unit node, parameter self-verification can be achieved, the flexibility and accuracy of node configuration can be improved, and the occurrence of configuration errors can be reduced.

[0059] Step S14: If the execution instruction is triggered, the original data is input into the visualization process, and the target function node is run to obtain visualization output data.

[0060] In this embodiment, after the visualization process is constructed, you can choose to start the entire workflow to execute all nodes, or execute a certain node separately. Specifically, if the execution instruction is triggered, the original data is input into the visualization process, and the target function node is run to obtain the visualization output data, which may include: if the execution instruction is triggered, the visualization process is verified based on the node connection order, and the invalid nodes are removed from the verification result to complete the initialization operation of the visualization process; the original data is input into the initialized visualization process, and the original data is processed through the target function node to obtain the processed data; the processed data is cached, and the visualization output data is generated according to the characteristics of the processed data. Among them, the visualization output data supports multiple forms of result display, including but not limited to data tables, charts and model outputs. Users can view the processed data in a tabular form through the data processing node, and intuitively display the data features through graphic nodes such as box plots, scatter plots, and pie charts. For the machine learning model training and verification process, the system provides model output nodes and verification nodes. The model output nodes support export in PMML format for further analysis; the model verification nodes display indicators such as confusion matrix and accuracy in a table to achieve comprehensive model result visualization.

[0061] In this embodiment, the visual orchestrator supports the creation of groups in the workflow, and then the nodes with the same processing logic are constructed under the group to achieve process reuse. The design of complex workflows is simplified, which facilitates unified management and reuse, improves the readability and maintenance efficiency of the process, and reduces the workload of repeated configuration. Furthermore, in the visual interface, the running status can be intuitively displayed through the node status icon. For example, in a specific implementation, yellow indicates that the target function node is being executed, green indicates successful execution, red indicates failed execution, and an exclamation mark indicates the existence of error information. When an exception occurs in a target function node in the process, the subsequent target function node will automatically stop running so that the user can correct the problem of the target function node in time.

[0062] It is worth mentioning that in order to ensure the integrity and reproducibility of the workflow, the embodiment of the present application can save the visualization process based on a preset format to obtain a visualization process file, so that a third party can run the visualization process through the visualization process file. For example, in a specific implementation, the constructed workflow can be exported as an XML file for saving and sharing. Specifically, after connecting the functional unit nodes on the workbench in series through connecting lines and creating corresponding data queues in the order of node connection to obtain a visualization process, it can also include: saving the visualization process based on the XML file format to obtain a visualization process file, so that a third party can run the visualization process through the visualization process file.

[0063] In this embodiment, see Figure 4 As shown, for the process of training and verifying the machine learning model, the visual workflow orchestrator first performs a data preparation step. Specifically, the original data set generated by the data partitioning node according to the data partitioning strategy is obtained through the data reading node, and the original data set is processed through the data processing node to obtain the target data set, which may include: the visual workflow orchestrator imports the original data into the workflow by adding a data reading node (such as a CSV reading node or an Excel reading node); through the missing value node, the original data is processed according to the data type or the requirements of a specific column; wherein, it is possible to choose to delete the rows containing missing values, fill in fixed values ​​or calculate the average value to replace the missing values; through the column filtering node, the filtering rules are applied to remove the columns that do not meet the analysis requirements, and only the necessary feature columns are retained; other data processing nodes are used to further clean and transform the data to ensure that all data types are consistent, complete, and meet the characteristics required for analysis. Through the data preparation of the above steps, a clean data set that meets the requirements can be generated.

[0064] Furthermore, the visual workflow orchestrator trains the machine learning model. Specifically, the target data set is used as input data of a preset training node, and the preset training node is configured based on the input data so as to obtain the target model by running the preset training node, which may include: selecting a suitable machine learning training node in the node library, such as a decision tree or linear regression node; connecting the last node of the data preparation step to the selected training node as the training node input data, and automatically parsing the column information in the input data through the node configuration item so as to configure the required feature columns in the training node; and running the training node to complete the training of the machine learning model.

[0065] Furthermore, the visual workflow orchestrator verifies the machine learning model. Specifically, the target model file is generated by writing the node in a preset format based on the target model so that the model prediction node verifies the effect of the target model through the target model file, which may include: after the model training is completed, the node is written through PMML (Predictive Model Markup Language, a general specification for data mining), and the trained model is used as input; the PMML writing node is run, the model is converted into a PMML format file through the JPMML library, and the generated PMML file information is returned for downloading. That is, by exporting the trained model as a PMML file, it is easy to share and deploy the model on different platforms and language environments, so that the model can achieve prediction independently of the training environment; accordingly, the visual workflow orchestrator supports loading standard PMML format models through PMML reading nodes, realizing plug-and-play and cross-system deployment of models without rewriting model logic or manual adaptation. After loading the PMML file in the model loading node, prepare the test data in the format required by the file, and use the corresponding prediction node to perform necessary preprocessing and format conversion on the test data, input the processed data into the node of the verification model, and finally analyze and evaluate the model prediction results by checking the confusion matrix, F1 score, accuracy and other evaluation indicators through the model accuracy node. Among them, the prediction nodes include but are not limited to decision tree prediction nodes, regression prediction nodes, and ARIMA (AutoregressiveIntegrated Moving Average, i.e., differential integrated moving average autoregressive model) prediction nodes.

[0066] It is worth mentioning that see Figure 5As shown, the process of generating the original data set by the data partition node according to the data partition strategy may include: extracting data from the original input data based on the data partition ratio and the data sampling strategy to generate the original data set; wherein the data sampling strategy includes any one or more of a simple sampling strategy, a random sampling strategy, a stratified sampling strategy, a linear sampling strategy and a time series sampling strategy. That is, by designing and encapsulating the data partition node, the automatic partitioning function of the data set is realized. The data partition node integrates multiple partitioning strategies and can flexibly divide the data set according to different needs. The simple sampling strategy is one of the most commonly used data extraction strategies, which extracts data in sequence or in reverse order and divides it into two data sets. After the user sets the partition ratio, the data allocation method is determined by selecting sequential or reverse reading. After the configuration is completed, the node will first load the input data into the memory and perform preprocessing according to the set ratio and sequence rules: when extracting in sequence, read the data in sequence by row or ratio starting from the first row of the data; when extracting in reverse order, read the data in sequence by row or ratio starting from the end of the input data. The random sampling strategy combines visual configuration of random seeds and partition ratios to achieve random data extraction for data partitioning. The strategy does not enable random seeds by default and uses java.util.Random (a random number generation tool) as the default random number generator. The data will be redistributed each time it is run. After setting the random seed, the extraction results will remain consistent and easy to reproduce or trace. The stratified sampling strategy divides the data into different subsets by specifying feature columns and partitioning the data by partition ratio. Specifically,

[0067] According to the feature column specified in the configuration, the node divides the input data into several subsets according to the value of the column; according to the configuration, in each subset, samples are randomly selected according to the set sampling ratio (such as 50% for each layer); the data extracted from each subset are merged into the final data set and combined with the total data to verify whether the data is legal under the partition ratio; random seed is optional. If a random seed is set, the sampling results will remain consistent for reproduction; through this stratified sampling method, the final sample set will better reflect the proportional distribution of the specified feature column in the overall data, and improve the representativeness of the sample. The linear sampling strategy can make the data evenly covered, and the overall trend and representativeness of the data can be maintained by extracting the sample size at fixed intervals. The embodiment of the present application calculates the target sample number n by the partition ratio, and calculates the sampling interval k according to the ratio of the total number of data items N and the target sample number n. The time series sampling strategy first selects the time column as the key feature field and sorts the data set according to the time column. Then, the number of extractions is calculated according to the set ratio and the total number of data items N, and the data is extracted in sequence in the sorted order to generate a sub-data set that meets the set ratio. The time series sampling strategy ensures the temporal continuity of the extraction results.

[0068] As can be seen from the above, the embodiment of the present application uses a series of visual arrangement interfaces, and users drag and click to assemble data processing or machine learning processes, and complete data feature extraction and model verification operations through parameter configuration without writing code, which improves decision-making progress, avoids human code errors, improves analysis accuracy, and promotes work and collaboration efficiency; by providing a data overview, it helps users quickly understand the overall characteristics and distribution of the data; identifies outliers and extreme values, and reveals potential data problems; simplifies analysis and decision-making, so that users can intuitively understand the data and support data-driven decision-making; the embodiment of the present application also provides a basis for further data processing and model building, which helps to optimize the analysis process; has good scalability and can realize custom extensions to meet various needs, integrating basic data processing nodes including character processing, time conversion, row and column filtering, data splitting and merging, mathematical formula calculation, script parsing, DB / file reading (writing), etc., and also integrates machine learning nodes such as ARIMA, decision tree, regression, and model accuracy verification; In the view field, traditional pie charts, line charts, bar charts, etc. are integrated, as well as box plots, heat maps, scatter plots, residual plots, etc., which are used to understand model performance and data sub-maps. Users can complete data preprocessing, model training, evaluation and result display on the same platform, avoiding switching between multiple tools; the system's built-in visualization components enable users to view the data processing progress, model training status and prediction results of each node in real time during task execution. This real-time feedback function provides an intuitive display of the analysis process, helping users to quickly identify potential problems and adjust analysis strategies, and enhance users' understanding and control of results; the task flow designed by the user can be saved as a template for easy reuse, and automatically triggered to execute under set conditions or time through the system's automatic scheduling function, which can reduce manual intervention and improve processing efficiency; the system also supports custom extensions, allowing users to integrate custom nodes to implement specific functions or models according to needs, enhancing the flexibility of the platform and meeting complex and diverse analysis and processing needs. This can achieve full process management of machine learning model training and verification, reduce development barriers, and improve the compatibility of model deployment.

[0069] See also Figure 6 As shown, the embodiment of the present application discloses a specific machine learning model training and verification visualization orchestration method applied to a visualization workflow orchestrator, including:

[0070] In this embodiment, after the execution operation is triggered, the background will sort out the process sequence according to the node connection information, remove the invalid nodes without connection, and perform initialization operations on each node in turn, including but not limited to configuration reading, cache object preparation, establishing database connection and creating session, etc. After successful initialization, the node enters the execution state and waits for data input.

[0071] In this embodiment, during the node execution process, the system will determine whether to execute immediately or wait for the arrival of input data based on the connection relationship of the node. If the input data has not arrived, the node will be in a blocked state until the required data is received. After the data arrives, the node executes its core functions and converts the processed data into a standard defined format and continues to pass it to the downstream node, while writing to the cache for fast access by the front end.

[0072] In this embodiment, the system supports pure memory cache and distributed cache services, and the cache mode can be flexibly switched through configuration items. When the node is executed, it determines whether to write data into the corresponding cache service according to the whether to cache data parameter in the configuration item.

[0073] In this embodiment, when extracting features, the configuration item determines whether to enable statistical parameters, and the corresponding method is executed to generate key indicators such as mean, median, and quartiles, so as to intuitively display the characteristics and distribution of the data. These statistical information can quickly summarize the central tendency and dispersion of the data, identify outliers, and provide a basis for subsequent analysis and modeling.

[0074] In this embodiment, the system uses cache to accelerate visualization and supports multiple forms of result display, including but not limited to data tables, charts, and model outputs. Users can view processed data in tabular form through data processing nodes, and intuitively display data features through graphic nodes such as box plots, scatter plots, and pie charts. In terms of models, the system provides model output nodes and verification nodes. Model output supports export in PMML format for further analysis, and model verification displays indicators such as confusion matrix and accuracy through tables to achieve comprehensive model result visualization.

[0075] In this embodiment, the system supports creating groups in the workflow, and then building nodes with the same processing logic under the group to achieve process reuse. This simplifies the design of complex workflows, facilitates unified management and reuse, improves the readability and maintenance efficiency of the process, and reduces the workload of repeated configuration.

[0076] As can be seen from the above, the embodiment of the present application provides a basis for further processing and model building of data, which helps to optimize the analysis process; it has good scalability and can realize custom extension to meet various needs, integrating basic data processing nodes including character processing, time conversion, row and column filtering, data splitting and merging, mathematical formula calculation, script parsing, DB / file reading (writing), etc., and also integrates machine learning nodes such as ARIMA, decision tree, regression, and model accuracy verification; in the view field, traditional pie charts, line charts, bar charts, etc. are integrated, while box plots, heat maps, scatter plots, residual plots, etc. are integrated for understanding model performance and data sub-graphs. Users can complete data preprocessing, model training, evaluation and result display on the same platform, avoiding switching between multiple tools; the task flow designed by the user can be saved as a template for easy reuse, and automatically triggered to execute under set conditions or time through the system's automatic scheduling function, which can reduce manual intervention and improve processing efficiency; the system also supports custom extensions, allowing users to integrate custom nodes according to needs to implement specific functions or models, enhance the flexibility of the platform, and meet complex and diverse analysis and processing needs. This enables full-process management of machine learning model training and verification, reduces development barriers, and improves the compatibility of model deployment.

[0077] Accordingly, see Figure 7 As shown, the embodiment of the present application discloses a machine learning model training and verification visual orchestration device applied to a visual workflow orchestrator, comprising:

[0078] A node configuration initialization module 11 is used to move the function unit node from the node library to the workbench in the visual workflow orchestrator and initialize the node configuration of the function unit node;

[0079] A visualization process creation module 12 is used to connect the functional unit nodes on the workbench in series through connecting lines, and create a corresponding data queue according to the node connection sequence to obtain a visualization process; the data queue can generate and transmit new data according to the update of the node configuration of the functional unit node;

[0080] A target function node generation module 13 is used to adjust the node configuration of the function unit node through the node configuration interface in the visual workflow orchestrator to obtain a target function node; the target function node includes a data reading node, a data partitioning node, a data processing node, a preset training node, a preset format writing node and a model prediction node;

[0081] The visualization process execution module 14 is used to input the original data into the visualization process and run the target function node to obtain visualization output data if the execution instruction is triggered.

[0082] As can be seen from the above, the embodiment of the present application configures the functional nodes based on the input data, and can integrate custom nodes according to needs to implement specific functions or models, thereby enhancing the flexibility of the platform and meeting complex and diverse analysis and processing needs; completing machine learning model training and verification through a number of functional nodes, and completing data feature extraction and model verification operations without writing code. This enables the full process management of machine learning model training and verification, reduces development barriers, and improves the compatibility of model deployment.

[0083] In some specific implementations, the target function node generation module 13 may specifically include:

[0084] A node configuration rule determination unit, used to determine a standardized node configuration rule through a preset interface; the node configuration rule is a rule for configuring the attribute definition, data input and output structure, verification rule and state management method of the functional unit node;

[0085] A node configuration adjustment unit, configured to adjust the node configuration of the functional unit node based on the node configuration rule through the preset interface to obtain an updated node configuration;

[0086] The node configuration verification unit is used to verify the updated node configuration and clear the data in the data queue corresponding to the functional unit node to obtain the target functional node.

[0087] In some specific implementations, the visualization process execution module 14 may specifically include:

[0088] A visualization process initialization unit, for verifying the visualization process based on the node connection sequence if an execution instruction is triggered, and removing invalid nodes based on the verification result to complete the initialization operation of the visualization process;

[0089] A raw data processing unit, used for inputting raw data into the initialized visualization process, and processing the raw data through the target function node to obtain processed data;

[0090] The visualization output data generating unit is used to cache the processed data and generate visualization output data according to the characteristics of the processed data.

[0091] In some specific implementations, the machine learning model training and verification visualization arrangement device may further include:

[0092] A workflow design module, which is used for a visual workflow choreographer to perform visual process choreography through a workflow design interface; the workflow design interface includes the node library, the workbench, the node configuration interface, a monitor for real-time monitoring of the operating status of each functional unit node on the workbench, and a data display area for intuitively displaying data through locally pre-integrated preset visualization components;

[0093] A visualization process file generation module, used to save the visualization process based on a preset format to obtain a visualization process file, so that a third party can run the visualization process through the visualization process file;

[0094] The machine learning model training and verification module is used to perform machine learning model training and verification processes.

[0095] Furthermore, in some specific implementations, the machine learning model training and verification module may specifically include:

[0096] The target data set generation submodule is used to obtain the original data set generated by the data partitioning node according to the data partitioning strategy through the data reading node, and process the original data set through the data processing node to obtain the target data set;

[0097] A target model generating unit, configured to use the target data set as input data of a preset training node, and configure the preset training node based on the input data, so as to obtain a target model by running the preset training node;

[0098] The target model effect verification unit is used to write a node to generate a target model file based on the target model through a preset format, so that the model prediction node verifies the effect of the target model through the target model file.

[0099] Furthermore, in some specific implementations, the target data set generation submodule may specifically include:

[0100] An original data set generating unit, used for extracting data from original input data based on a data partition ratio and a data sampling strategy to generate an original data set;

[0101] The data sampling strategy includes any one or more of a simple sampling strategy, a random sampling strategy, a stratified sampling strategy, a linear sampling strategy and a time series sampling strategy.

[0102] Furthermore, the present application also discloses an electronic device. Figure 8It is a structural diagram of an electronic device 20 shown according to an exemplary embodiment, and the content in the figure cannot be regarded as any limitation on the scope of use of this application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input and output interface 25 and a communication bus 26. Among them, the memory 22 is used to store a computer program, and the computer program is loaded and executed by the processor 21 to implement the relevant steps in the machine learning model training and verification visualization arrangement method disclosed in any of the aforementioned embodiments. In addition, the electronic device 20 in this embodiment can specifically be an electronic computer.

[0103] In this embodiment, the power supply 23 is used to provide working voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and the external device, and the communication protocol it follows is any communication protocol that can be applied to the technical solution of the present application, and is not specifically limited here; the input and output interface 25 is used to obtain external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs and is not specifically limited here.

[0104] In addition, the memory 22, as a carrier for storing resources, can be a read-only memory, a random access memory, a disk or an optical disk, etc. The resources stored thereon can include an operating system 221, a computer program 222, etc., and the storage method can be temporary storage or permanent storage.

[0105] The operating system 221 is used to manage and control the hardware devices and computer program 222 on the electronic device 20, which can be Windows Server, Netware, Unix, Linux, etc. In addition to including a computer program that can be used to complete the machine learning model training and verification visualization arrangement method performed by the electronic device 20 disclosed in any of the aforementioned embodiments, the computer program 222 can further include a computer program that can be used to complete other specific tasks.

[0106] Furthermore, the present application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, the aforementioned disclosed machine learning model training and verification visualization arrangement method is implemented. The specific steps of the method can refer to the corresponding contents disclosed in the aforementioned embodiments, and will not be repeated here.

[0107] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0108] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0109] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0110] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0111] The technical solution provided by the present application is introduced in detail above. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea. At the same time, for general technicians in this field, according to the idea of ​​the present application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A visual arrangement method for machine learning model training and verification, characterized in that: Applications in visual workflow orchestrators, including: Move the function unit node from the node library to the workbench in the visual workflow choreographer, and initialize the node configuration of the function unit node; Connecting the functional unit nodes on the workbench in series through connecting lines, and creating corresponding data queues according to the node connection sequence to obtain a visual process; the data queue can generate and transmit new data according to the update of the node configuration of the functional unit node; Adjusting the node configuration of the functional unit node through the node configuration interface in the visual workflow orchestrator to obtain a target functional node; the target functional node includes a data reading node, a data partitioning node, a data processing node, a preset training node, a preset format writing node and a model prediction node; If the execution instruction is triggered, the original data is input into the visualization process, and the target function node is run to obtain visualization output data.

2. The method for visually arranging machine learning model training and verification according to claim 1, characterized in that: The visual workflow choreographer performs visual process choreography through a workflow design interface; Among them, the workflow design interface includes the node library, the workbench, the node configuration interface, a monitor for real-time monitoring of the operating status of each functional unit node on the workbench, and a data display area for intuitively displaying data through locally pre-integrated preset visualization components.

3. The method for visually arranging machine learning model training and verification according to claim 2, characterized in that: Adjusting the node configuration of the functional unit node through the node configuration interface to obtain a target functional node includes: Determine standardized node configuration rules through a preset interface; the node configuration rules are rules for configuring attribute definitions, data input and output structures, verification rules, and state management methods of functional unit nodes; Adjusting the node configuration of the functional unit node based on the node configuration rule through the preset interface to obtain an updated node configuration; The updated node configuration is verified, and data in a data queue corresponding to the functional unit node is cleared to obtain a target functional node.

4. The method for visually arranging machine learning model training and verification according to claim 2, characterized in that: If the execution instruction is triggered, the original data is input into the visualization process, and the target function node is run to obtain visualization output data, including: If the execution instruction is triggered, the visualization process is verified based on the node connection sequence, and invalid nodes are removed from the verification result to complete the initialization operation of the visualization process; Inputting raw data into the initialized visualization process, and processing the raw data through the target function node to obtain processed data; The processed data is cached, and visual output data is generated according to the features of the processed data.

5. The method for visual arrangement of machine learning model training and verification according to claim 2, characterized in that: After connecting the functional unit nodes on the workbench in series through connecting lines and creating corresponding data queues according to the node connection sequence to obtain a visualized process, the method further includes: The visualization process is saved based on a preset format to obtain a visualization process file, so that a third party can run the visualization process through the visualization process file.

6. The method for visual arrangement of machine learning model training and verification according to any one of claims 1 to 5, characterized in that: The process of training and validating the machine learning model includes: The data reading node obtains the original data set generated by the data partitioning node according to the data partitioning strategy, and the data processing node processes the original data set to obtain the target data set; Using the target data set as input data of a preset training node, and configuring the preset training node based on the input data, so as to obtain a target model by running the preset training node; Based on the target model, a target model file is written into the node in a preset format so that the model prediction node can verify the effect of the target model through the target model file.

7. The method for visual arrangement of machine learning model training and verification according to claim 6, characterized in that: The process of generating the original data set by the data partition node according to the data partition strategy includes: Extract data from the original input data based on the data partition ratio and the data sampling strategy to generate an original data set; The data sampling strategy includes any one or more of a simple sampling strategy, a random sampling strategy, a stratified sampling strategy, a linear sampling strategy and a time series sampling strategy.

8. A visual arrangement device for machine learning model training and verification, characterized in that: Applications in visual workflow orchestrators, including: A node configuration initialization module, used to move the function unit node from the node library to the workbench in the visual workflow orchestrator, and initialize the node configuration of the function unit node; A visualization process creation module, used to connect the functional unit nodes on the workbench in series through connecting lines, and create corresponding data queues according to the node connection sequence to obtain a visualization process; the data queue can generate and transmit new data according to the update of the node configuration of the functional unit node; A target function node generation module, used to adjust the node configuration of the function unit node through the node configuration interface in the visual workflow orchestrator to obtain a target function node; the target function node includes a data reading node, a data partitioning node, a data processing node, a preset training node, a preset format writing node and a model prediction node; The visualization process execution module is used to input the original data into the visualization process and run the target function node to obtain visualization output data if the execution instruction is triggered.

9. An electronic device, characterized in that: include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the method for visually arranging machine learning model training and verification as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: Used to store a computer program, wherein the computer program, when executed by a processor, implements the machine learning model training and verification visualization orchestration method as described in any one of claims 1 to 7.

Images