Autonomous learning training visualization method, computer equipment and storage medium

Abstract

The invention discloses an autonomous learning training visualization method, computer equipment and a storage medium, and the method comprises the steps: building an autonomous learning training visualization system, calling a network model editing module of the autonomous learning training visualization system, creating a target network model in a network model editing interface main window, and calling a data processing module. Selecting a target data set for preprocessing and splitting to obtain training data, test data and verification data, calling a hyper-parameter setting module, and setting hyper-parameters for training a target network model and a training data storage path; training data, test data, verification data and set hyper-parameters are adopted to train, test and verify a target network model, a log module and a data statistics broken line graph module are called to store and monitor training information and hyper-parameters of the target network model in the training, test and verification processes, and a statistical graph is output. According to the method, the training process of the target network model is visualized, and the training effect of the model can be quickly obtained.

Description

Technical Field

[0001] The present invention relates to the field of intelligent perception algorithm visualization, and in particular to an autonomous learning training visualization method, a computer device and a storage medium. Background Art

[0002] In today's era of rapid development of artificial intelligence, deep learning technology has become the key to technological innovation in many fields. However, the structure of intelligent perception algorithms is complex, the training process is complicated and consumes hardware resources, and the model is not interpretable, which poses a challenge to technology beginners and professional researchers and limits the development of deep learning technology. Existing intelligent perception algorithm visualization methods include visualization of the training process of intelligent perception algorithms and visualization of the structure of intelligent perception algorithms. The intelligent perception algorithm based on the Pytorch framework can use the TensorBored visualization tool to visualize the structure of the intelligent perception algorithm and monitor the accuracy, loss function, model convergence degree, etc. of the intelligent perception algorithm during the training process. However, this method requires users to write visualization code and view the visualization results through the TensorBored web link, which is complicated to use.

[0003] Based on the above challenges, it is particularly important to provide a visual operation interface and intelligent perception algorithm training optimization so that users can complete the training, testing and visualization of the intelligent perception algorithm through simple operations. Summary of the invention

[0004] The present invention proposes a neural network visualization configuration system and method based on mindspore, builds a network model based on the Huawei mindspore framework, and builds a system visualization interface based on a cross-platform application development framework, which can provide a variety of visualization methods for intelligent perception algorithms, allowing users to complete the training, testing and visualization of intelligent perception algorithms through efficient and simple operations.

[0005] On one hand, the present invention provides an autonomous learning training visualization method, comprising the following steps: S1. Build an autonomous learning and training visualization system, which includes a network model editing module, a data processing module, a hyperparameter setting module, a log module, and a data statistics line chart module; S2. Open the autonomous learning training visualization system, call the network model editing module, and use the preset network model creation method to create the target network model in the main window of the network model editing interface; S3, calling the data processing module, selecting a data source in the data loading window of the data processing module, selecting a target data set from the data source, preprocessing the target data set according to a preset data set preprocessing method, setting a data set splitting ratio, splitting the preprocessed data set into training data, test data, and verification data, and displaying them; S4, calling the hyperparameter setting module, and setting the hyperparameters for training the target network model and the training data saving path in the hyperparameter setting window of the hyperparameter setting module; S5. Use the training data, test data, verification data and set hyperparameters to train, test and verify the target network model, call the log module to form a training log and save the training information of the target network model during the training, testing and verification process, call the data statistics line chart module to monitor the hyperparameters of the target network model during the training, testing and verification process, and output the statistical chart after the training, testing and verification are completed.

[0006] Preferably, the autonomous learning training visualization system also includes a network model encapsulation library, a network operator encapsulation library, and a network model storage module. The preset network model creation method used in S2 to create the target network model specifically includes: directly calling the pre-packaged classic deep neural network model from the network model encapsulation library as the target network model, selecting several network operators from the network operator encapsulation library and creating the target network model, and selecting the historically saved network model json file from the network model storage module and creating the target network model.

[0007] Preferably, the autonomous learning training visualization system further includes a network model analysis module, a network model parsing module and a mindspore framework calling module, which selects several network operators from the network operator encapsulation library and creates a target network model. The specific process is as follows: S21. Click the operator library button in the main window of the network model editing interface to call the network operator encapsulation library, select several network operators from the network layer operator library sub-window and drag them to the designer sub-window of the main window of the network model editing interface; S22, switch the mouse mode to the connection mode, connect the selected network operators in the required order, and after the connection is completed, click the mouse mode button to switch the connection mode back to the mouse mode; S23, hover the mouse over the target network operator, click the target network operator and pop up the property bar of the target network operator, view and set the name and hyperparameters of the target network operator in the property bar, and obtain the initial network model; S24, click the analysis button in the main window of the network model editing interface to call the network model analysis module, which analyzes the initial network model to confirm whether there is any abnormality in the initial network model; S25. If the initial network model does not have any abnormality, click the Save button in the network model editing interface window to call the network model storage module to save the initial network model without any abnormality into a json file; S26. The network model parsing module parses the json file, and the mindspore framework calling module builds the target network model according to the parsing result of the network model parsing module.

[0008] Preferably, the network operator encapsulation library in S21 specifically includes 8 classification operators: convolution and fully connected layer, activation layer, input layer, data preprocessing layer, sequence layer, normalization layer, output layer, and pooling layer.

[0009] Preferably, S3 specifically includes the following steps: S31. Click the data button in the main window of the network model editing interface to call the data processing module, select load data in the data subwindow, select the data source in the data set selection subwindow in the load data window, and select the target data set according to the data source; S32, setting a data preprocessing option for the target data set in the data set selection subwindow according to the training requirements of the target network model, and preprocessing the target data set according to a preset data preprocessing method to obtain a preprocessed data set; S33, setting a split ratio in a data set split ratio setting subwindow in a loading data window according to the training requirements of a target network model, and splitting the preprocessed data set into training data, test data, and verification data according to the split ratio; S34, click the Submit button in the Load Dataset window, and the data processing module processes the target data set according to the selected data preprocessing method and split ratio to obtain the split information of the target data set; S35, the data statistics line chart module counts the split information of the target data set and draws a bar chart; S36. After clicking the Submit button, the loading data window is closed, and the splitting information and bar chart of the target data set are displayed in the data window.

[0010] Preferably, the data preprocessing methods preset in S32 include: random flipping, random rotation, random scaling, random horizontal translation, and random vertical translation.

[0011] Preferably, S4 specifically includes: S41. Click the training button in the main window of the network model editing interface to pop up a training sub-window. Click the hyperparameter button in the training sub-window to call the hyperparameter setting module. Set the hyperparameters for training the target network model and the path and name for saving the target network model after training in the hyperparameter setting window. S42, confirming the saving path and file name of the target model after training in the pop-up window for setting the saving path and name of the target model after training; S43. Click the Submit button in the hyperparameter setting window. If a successful submission pop-up window appears, it means that the hyperparameters and the saving path and file name settings of the trained target model are effective.

[0012] Preferably, S5 specifically includes: S51. Click the training button in the main window of the network model editing interface, a training sub-window pops up, and click the start training button in the training sub-window, and the target network model enters the training state; S52, training, testing and verifying the target network model using the training data, test data, verification data and set hyperparameters; S53, the log module forms a training log and saves the training information of the target network model during the training, testing and verification process; S54, the data statistics line graph module monitors the hyperparameters of the target network model during the training, testing and verification process, and records the data and draws a statistical graph according to a preset number of printing intervals; S55. After the training is completed, click the training display button to display the statistical chart in the training sub-window.

[0013] Another aspect of the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned autonomous learning training visualization method when executing the computer program.

[0014] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned autonomous learning training visualization method are implemented.

[0015] The above-mentioned autonomous learning training visualization method, computer equipment and storage medium are used to build an autonomous learning training visualization system, which includes a network model editing module, a data processing module, a hyperparameter setting module, a log module and a data statistics line chart module. The autonomous learning training visualization system is opened, the network model editing module is called, and a target network model is created using a preset network model creation method in the main window of the network model editing interface. The data processing module is called, and a data source is selected in the data loading window of the data processing module. A target data set is selected from the data source, and the target data set is preprocessed according to a preset data set preprocessing method, and a data set splitting ratio is set. For example, the preprocessed data set is split into training data, test data, and validation data and displayed. The hyperparameter setting module is called. The hyperparameter setting window of the hyperparameter setting module sets the hyperparameters for training the target network model and the training data saving path. The target network model is trained, tested, and validated using the training data, test data, validation data, and the set hyperparameters. The log module is called to form a training log and save the training information of the target network model during the training, testing, and validation process. The data statistics line chart module is called to monitor the hyperparameters of the target network model during the training, testing, and validation process, and a statistical chart is output after the training, testing, and validation are completed. This method visualizes the training process of the target network model and can quickly learn the training effect of the target network model. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a flowchart of a visualization method for self-learning training; Figure 2 It is a line chart showing the changes in the training loss value during the 1-100 rounds of training, with the caltech101 dataset as the target dataset and resnet18 as the target model; Figure 3 This is a line chart showing the changes in test loss, training loss, test accuracy, and training accuracy during the training process, with the caltech101 dataset as the target dataset and resnet18 as the target model; Figure 4 A schematic block diagram of a computer device. DETAILED DESCRIPTION

[0017] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings.

[0018] In one embodiment, see Figure 1 , a self-learning training visualization method, comprising the following steps: S1. Build an autonomous learning and training visualization system, which includes a network model editing module, a data processing module, a hyperparameter setting module, a log module, and a data statistics line chart module.

[0019] Specifically, the above-mentioned autonomous learning training visualization system is based on Huawei's mindspore framework to build a network model, and based on a cross-platform application development framework to build a system visualization interface. The autonomous learning training visualization system provides a variety of visualization methods for intelligent perception algorithms, and users can complete the training, testing and visualization of intelligent perception algorithms through efficient and simple operations. The autonomous learning training visualization system includes a network model editing module, a data processing module, a hyperparameter setting module, a log module and a data statistics line chart module, wherein the network model editing module is used to create a target network model according to demand, the data processing module is used to create a data set to train the target network model, the hyperparameter setting module is used to set the hyperparameters of the target network model during the training process, the log module is used to form a training log for the training information of the target network model during the training process, and the data statistics line chart module is used to monitor the hyperparameters of the target network model during the training process, and is also used to display the monitoring information of the data set and hyperparameters in the form of a chart.

[0020] The above functional modules are set in parallel and can be selected according to needs. For example, the data processing module will be used when setting the data set, and the hyperparameter setting module will be used when setting the hyperparameter.

[0021] S2. Open the autonomous learning training visualization system, call the network model editing module, and use the preset network model creation method to create the target network model in the main window of the network model editing interface.

[0022] Furthermore, the autonomous learning training visualization system also includes a network model encapsulation library, a network operator encapsulation library, and a network model storage module. The preset network model creation method used in S2 to create the target network model specifically includes: 1) directly calling the pre-packaged classic deep neural network model from the network model encapsulation library as the target network model; 2) selecting several network operators from the network operator encapsulation library and creating the target network model; 3) selecting the historically saved network model json file from the network model storage module and creating the target network model.

[0023] 1) Directly call the pre-packaged classic deep neural network model from the network model encapsulation library as the target network model. The specific process is as follows: The network model encapsulation library pre-packages multiple classic deep neural network models. Select one from multiple deep neural network models as the target network model. Multiple deep neural network models include: ResNet18, ResNet34, ResNet50, ResNet101, ResNet152, MobilenetV2, EfficientNet, AlexNet and YoloV5.

[0024] Click the "Import Model" button in the main window of the network model editing interface, and the "Select Imported Network Model" window will appear. This window displays the classic deep neural network model preset by the network model package library. Click to select a suitable neural network as the target network model. The neural network turns green to indicate that it is selected. To cancel the selection, just click the selected neural network again. After completion, click Confirm to cancel the selected neural network.

[0025] 2) The autonomous learning training visualization system also includes a network model analysis module, a network model parsing module, and a mindspore framework calling module. It selects several network operators from the network operator encapsulation library and creates a target network model. The specific process is as follows: S21. Click the "Operator Library" button in the main window of the network model editing interface to call the network operator encapsulation library, select several network operators from the network layer operator library sub-window and drag them to the designer sub-window of the main window of the network model editing interface.

[0026] Table 1 Network operator classification and name list ; Furthermore, the network operator encapsulation library specifically includes 8 categories of operators: convolution and fully connected layer, activation layer, input layer, data preprocessing layer, sequence layer, normalization layer, output layer, and pooling layer. There are a total of 43 operators in these 8 categories, see Table 1 for details.

[0027] Select several network operators from the 43 operators in 8 categories in Table 1 according to your needs and drag them to the designer subwindow of the main window of the network model editing interface.

[0028] S22, switch the mouse mode to the connection mode, connect the selected network operators in the required order, and after the connection is completed, click the "Mouse Mode" button to switch the connection mode back to the mouse mode; S23, hover the mouse over the target network operator, click the target network operator and pop up the property bar of the target network operator, view and set the name and hyperparameters of the target network operator in the property bar, and obtain the initial network model; S24. Click the "Analyze" button in the main window of the network model editing interface to call the network model analysis module, which analyzes the initial network model to confirm whether there is any abnormality in the initial network model; S25. If there is no abnormality in the initial network model, click the "Save" button in the network model editing interface window to call the network model storage module to save the initial network model without abnormality into a json file; S26. The network model parsing module parses the json file, and the mindspore framework calling module builds the target network model according to the parsing result of the network model parsing module.

[0029] Specifically, first click the "Operator Library" button from the main window of the network model editing interface to call the network operator encapsulation library, select several network operators from the network layer operator library subwindow, drag them to the designer subwindow of the main window of the network model editing interface, click the "Connection Mode" button in the network model editing interface window, switch the mouse mode to the connection mode, connect the selected network operators in a connection manner as needed, and click "Mouse Mode" in the network model editing interface window after completion to switch the connection mode back to the mouse mode. For each of the several network operators after the connection is completed, place the mouse on each network operator and right-click the mouse. A property bar appears on the right side of the network model editing interface window. The name of the current network operator can be viewed through the property bar and the operator hyperparameters can be set. After all selected network operators have completed property editing, the initial network model is obtained. Then click the "Analysis" button in the network model editing interface window to call the network model analysis module. The network model analysis module analyzes the initial network model to confirm whether there are any abnormalities. When it is confirmed that the initial network model has no abnormalities, click the "Save" button in the network model editing interface window to call the network model storage module. The initial network model without abnormalities is saved as a json file through the storage function in the network model storage module. At this time, the parameter settings of the network operators in the model, the parent-child relationship formed by the connection, etc. are all recorded in the json file; in addition, click "Mouse Mode" in the network model editing interface window. After switching to mouse mode, hover the mouse over a certain network operator or the parameters of the network operator to view the description information of the network operator. Click the "Copy", "Cut", and "Paste" buttons in the network model editing interface window to perform corresponding operations on the network operator. Select one of several network operators with the mouse, and press the DEL key on the computer keyboard to delete the selected network operator. Press and hold the right mouse button in the blank area of ​​the network model editing interface window to drag the network model editing interface window, and zoom in or out the editing interface by operating the scroll wheel. The network layer operator library window can be dragged and closed. After closing, click the operator library button on the toolbar to display it again. After obtaining the json file, the network model parsing module parses the json file, and the mindspore framework calling module builds the target network model according to the parsing results of the network model parsing module.

[0030] 3) Select the historically saved network model json file from the network model storage module and create the target network model. The specific process is as follows: Specifically, click the "Open Model" button in the main window of the network model editing interface, and a neural network model window will pop up. The window displays the historically saved network model json files stored in the network model storage module. Select a historically saved network model json file from the network model storage module. After selecting, click the "open" button in the upper right corner of the neural network model window to call the network model parsing module to parse the json file. The mindspore framework calling module builds the target network model according to the parsing results of the network model parsing module.

[0031] S3. Call the data processing module, select the data source in the data loading window of the data processing module, select the target data set from the data source, preprocess the target data set according to the preset data set preprocessing method, set the data set splitting ratio, split the preprocessed data set into training data, test data, and verification data, and display them.

[0032] Specifically, after the target network model is built, a data source is selected in the loading data window according to the training requirements, a target data set is selected from the data source, and a specific preprocessing method is selected to preprocess the target data set.

[0033] Furthermore, S3 specifically includes the following steps: S31. Click the "Data" button in the main window of the network model editing interface to call the data processing module, select "Load Data" in the data sub-window, select the data source in the data set selection sub-window in the load data window, and select the target data set based on the data source.

[0034] The data sources in S31 include: several classic neural network data sets encapsulated in the data set encapsulation library and user-defined data sets. Among them, the several classic neural network data sets encapsulated in the data set encapsulation library include: cifar10 data set, cifar100 data set, caltech101 data set, caltech256 data set, and mnist data set.

[0035] If you select "ImageDatastore in Workspace" in the drop-down box, the data source is the classic neural network dataset encapsulated in the dataset encapsulation library. In this case, select one of the several classic neural network datasets encapsulated in the dataset encapsulation library as the target dataset; If you select "Folder" in the drop-down box, the data source is a user-defined dataset. In this case, you need to enter a dataset folder path that meets the requirements (there are only subfolders in the folder, each subfolder contains a category of pictures, and the picture format can only be jpg, bmp, png, JPG). According to the input dataset folder path, the data processing module calls the corresponding dataset folder as the target dataset.

[0036] S32. According to the training requirements of the target network model, a data preprocessing option of the target data set is set in the data set selection subwindow, and the target data set is preprocessed according to a preset data preprocessing method to obtain a preprocessed data set.

[0037] Furthermore, the data preprocessing methods preset in S32 include: random flipping, random rotation, random scaling, random horizontal translation, and random vertical translation. One or more data preprocessing methods can be selected to preprocess the target data set.

[0038] S33. According to the training requirements of the target network model, a split ratio is set in the data set split ratio setting subwindow in the loading data window, and the preprocessed data set is split into training data, test data, and verification data according to the split ratio.

[0039] It should be noted that the split ratio divides the target data set into data categories, and splits each category of data into training data, test data, and validation data in proportion. The percentage of training data, the percentage of test data, and the percentage of validation data cannot be zero, otherwise the system will display an error pop-up window.

[0040] The target data set is split into training data, test data, and validation data in proportion to the data category. Specifically, stratified sampling can be used to obtain the data according to the category. For example, in the Caltech256 data set, the split ratio is 70% for training data, 20% for test data, and 10% for validation data. For the 98 images in the "ak47" category, 70% of the data, that is, 69 images, are randomly selected as training data. The specific calculation method is as follows: When the number of extracted images is not an integer, it is rounded up. When the total number of images after rounding up is greater than the total number of images of this category, it is rounded up first according to the size of the digit after the decimal point. For example, in this sampling, the number of training images is: , rounded up to 69, number of test images: , rounded up to 20, the number of verification images: , rounded up to 10, but the total number of images after rounding up is: , which is greater than the total number of images of this category, 98, then the decimal place is rounded up first. The decimal place 8 in 9.8 is greater than the decimal place 6 in 68.6 and 19.6, so the number of verification images is rounded up to 10 first. When the decimal places are the same, the training data>test data>verification data priority is used to round up. In this sampling, the number of training images, 68.6, and the number of test images, 19.6, have the same decimal place. According to the priority of training data>test data>verification data, the training images are rounded up first. Therefore, the final result of this sampling is: 69 training data, 19 test data, and 10 verification data.

[0041] It should also be noted that when the target dataset is the cifar10 dataset, the cifar100 dataset, or the mnist dataset, the dataset split ratio setting subwindow will not appear in the loading data window because the official data has been divided.

[0042] S34. Click the "Submit" button in the data set loading window. The data processing module processes the target data set according to the selected data preprocessing method and split ratio to obtain the split information of the target data set.

[0043] S35, the data statistics line chart module counts the split information of the target data set and draws a bar chart.

[0044] Specifically, the data statistics line chart module counts the split information of the target data set, including the target data set path, the number of image categories, the total number of images, the number of training images, the number of verification images, and the number of test images, and draws a bar chart after the statistics are completed.

[0045] S36. After clicking the "Submit" button, the loading data window is closed, and the splitting information of the target data set and the corresponding bar chart, that is, the data set diagnosis result, are displayed in the data window.

[0046] S4. Call the hyperparameter setting module, and set the hyperparameters for training the target network model and the training data saving path in the hyperparameter setting window of the hyperparameter setting module.

[0047] Specifically, after selecting the target data set and performing preprocessing and splitting, the user can set the relevant hyperparameters for the target network model training and the model training data set save path in the hyperparameter setting window.

[0048] Furthermore, S4 specifically includes: S41. Click the "Train" button in the main window of the network model editing interface to pop up a training sub-window. Click the "Hyperparameter" button in the training sub-window to call the hyperparameter setting module. Set the hyperparameters for training the target network model and the path and name for saving the target network model after training in the hyperparameter setting window.

[0049] In the hyperparameter setting window, you can set the target network model training related hyperparameters and the save path and name of the target model after training. The hyperparameters include: optimizer, loss function, initial learning rate, data set packing size, indicator set for model evaluation, automatic mixed precision, mode training level, number of training rounds, and number of printing intervals.

[0050] S42. Confirm the save path and file name of the target model after training in the pop-up window for setting the save path and name of the target model after training. The save path and file name of the target model after training is set in the pop-up window, in which the save path and file name of the target network model after training can be modified.

[0051] S43. Click the "Submit" button in the hyperparameter setting window. If a pop-up window indicating successful submission appears, it means that the hyperparameters and the saving path and file name settings of the trained target model are effective.

[0052] Specifically, after completing the hyperparameter setting, click the "Submit" button in the hyperparameter setting window. If a pop-up window with successful submission appears, it means that the hyperparameter setting is effective, and subsequent training of the target network model will be performed according to the set hyperparameters.

[0053] S5. Use the training data, test data, verification data and set hyperparameters to train, test and verify the target network model, call the log module to form and save the training log of the target network model during the training, testing and verification process, call the data statistics line chart module to monitor the hyperparameters of the target network model during the training, testing and verification process, and output the statistical chart after the training, testing and verification are completed.

[0054] Furthermore, S5 specifically includes: S51. Click the "Train" button in the main window of the network model editing interface, a training sub-window pops up, and click the "Start Training" button in the training sub-window, and the target network model enters the training state.

[0055] S52, using the training data, test data, verification data and set hyperparameters to train, test and verify the target network model.

[0056] S53, the log module forms a training log and saves the training information of the target network model during the training, testing and verification process.

[0057] S54, the data statistics line graph module monitors the hyperparameters of the target network model during the training, testing and verification processes, and records the data and draws statistical graphs according to the preset printing interval number.

[0058] S55. After the training is completed, click the "Training Display" button to display the statistical chart in the training sub-window.

[0059] Specifically, click the "Train" button in the main window of the network model editing interface to display the training interface. Click the "Start Training" button in the training sub-window of the training interface to put the target network model into the training state. At this time, both the "Start Training" button and the "Training Display" button will be in an unclickable state. During the training of the target network model (including training, testing, and verification), the log module will redirect the output of the training information printed by the target network model during the training period. That is, the log module will create a log file to enter the training information printed during the training of the target model to form a training log and save it. The data statistics line chart module monitors the loss value, test accuracy, training loss value, and training accuracy of the target network model during the training period, and records data and draws statistical charts according to the preset printing intervals. After the training is completed, click the "Training Display" button to display the statistical chart in the training sub-window. See Figure 2 and Figure 3 , Figure 2 This is a line chart showing the changes in the training loss value during the 1-100 rounds of training, with the caltech101 dataset as the target dataset and resnet18 as the target model. Figure 3 It is a line chart of the changes in test loss, training loss, test accuracy, and training accuracy during the training process, with the caltech101 dataset as the target dataset and resnet18 as the target model.

[0060] The above-mentioned autonomous learning training visualization system includes nine functional modules and three encapsulation libraries. The nine functional modules are network model editing module, network model analysis module, network model storage module, network model parsing module, log module, data processing module, hyperparameter setting module, data statistics line chart module and mindspore framework calling module. The three encapsulation libraries are network model encapsulation library, data set encapsulation library and network operator encapsulation library. Users can select the corresponding functional module or encapsulation library according to their needs to complete the corresponding operations.

[0061] In addition, it should be noted that before the training of the target network model begins, if the target network model is not created in the main interface of the network model editing, an error pop-up window will appear, displaying the error message "No graphics exist in the model editing interface, and the network model acquisition failed"; if the target dataset is not set in the loading data window, an error pop-up window will appear in the system, displaying the error message "No loading data related information is set, and the training is terminated"; if the custom dataset is selected as the data source and the custom dataset path is not entered, an error pop-up window will appear in the system, displaying "The training data folder path is empty!"; if the user-defined dataset does not meet the requirements, an error pop-up window will appear in the system, displaying "Please make sure that only image classification folders exist in the training data folder. There is currently a file type propertyinit.cpp that is not an image classification folder. , the file is abnormally terminated!"; If the data preprocessing option parameter settings are unreasonable (for example, the maximum value of the random vertical translation cannot be less than the minimum value), an error pop-up window will appear on the system, displaying "The maximum value of the random vertical translation cannot be less than the minimum value, please modify!"; If the target data set split ratio does not meet the requirements (for example, the sum of the training data percentage, test data percentage, and validation data percentage is not 100%), an error pop-up window will appear on the system, displaying "The sum of the training data percentage, test data percentage, and validation data percentage must be 100, please confirm!"; If the target data set split ratio does not meet the requirements (for example, one or more of the training data percentage, test data percentage, and validation data percentage is 0), an error pop-up window will appear on the system, displaying "The training data percentage, test data percentage, and validation data percentage must not be 0, please confirm!"; If the target network model training hyperparameters are not set in the hyperparameter setting window, an error pop-up window will appear on the system, displaying the error message "No hyperparameter related information is set, training terminated".

[0062] In one embodiment, a computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the above-mentioned autonomous learning training visualization method are implemented. The computer device can be a mobile phone, a tablet, a mobile computer, or any other computer device that can implement an autonomous learning training visualization.

[0063] In one embodiment, a computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the above-mentioned autonomous learning training visualization method.

[0064] Specifically, see Figure 4 , Figure 4 The figure is a schematic block diagram of a computer device in one embodiment of the present invention.

[0065] The computer device 3 may be a smart phone, tablet computer, laptop computer, desktop computer, etc. that executes a program. The computer device 3 of this embodiment includes at least but is not limited to: a memory 31, a processor 32, and a computer program 33 stored in the memory 31 and executable on the processor 32, which can be connected to each other through a system bus, such as a program corresponding to the above-mentioned autonomous learning training visualization method. It can be understood by those skilled in the art that the schematic diagram is only an example of the computer device 3 and does not constitute a limitation on the computer device 3.

[0066] Among them, the memory 31 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, mobile hard disk, multimedia card, card-type memory (for example: SD or DX memory, etc.), magnetic memory, disk, optical disk, etc. The memory 31 can be an internal storage unit of the electronic device 1, such as a mobile hard disk of the computer device 3, and the memory 31 can also be an external storage device of the computer device 3, such as a plug-in mobile hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the computer device 3. Further, the memory 31 can also include both an internal storage unit and an external storage device of the computer device 3. The memory 31 can not only be used to store application software and various types of data installed in the computer device 3, such as the program code corresponding to the above-mentioned autonomous learning training visualization method, but also can be used to temporarily store data that has been output or is to be output.

[0067] The processor 32 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chips. The processor 32 executes the operating system of the computer device 3 and various installed applications. In this embodiment, the processor 32 executes the computer program 33 stored in the memory 31 to implement the autonomous learning training visualization method disclosed in the embodiment of the present invention.

[0068] Since the aforementioned computer device and computer-readable storage medium both include the aforementioned autonomous learning training visualization method, they also have the beneficial effects of the visualization method, which will not be elaborated here.

[0069] The above-mentioned autonomous learning training visualization method, computer equipment and storage medium are used to build an autonomous learning training visualization system, which includes a network model editing module, a data processing module, a hyperparameter setting module, a log module and a data statistics line chart module. The autonomous learning training visualization system is opened, the network model editing module is called, and a target network model is created using a preset network model creation method in the main window of the network model editing interface. The data processing module is called, and a data source is selected in the data loading window of the data processing module. A target data set is selected from the data source, and the target data set is preprocessed according to a preset data set preprocessing method, and a data set splitting ratio is set. For example, the preprocessed data set is split into training data, test data, and validation data and displayed. The hyperparameter setting module is called, and the hyperparameter setting window of the hyperparameter setting module is used to set the hyperparameters for training the target network model and the training data saving path. The target network model is trained, tested, and validated using the training data, test data, validation data, and the set hyperparameters. The log module is called to form a training log and save the training information of the target network model during the training, testing, and validation process. The data statistics line chart module is called to monitor the hyperparameters of the target network model during the training, testing, and validation process, and the statistical chart is output after the training, testing, and validation are completed. This method integrates the functions of pre-trained model loading, data set loading, training hyperparameter setting, and network training result visualization, realizing a one-stop intelligent perception algorithm visualization solution, which can quickly obtain the training effect of the target network model.

[0070] The above is a detailed introduction to an autonomous learning training visualization method, a computer device and a storage medium provided by the present invention. This article uses specific examples to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the core idea of ​​the present invention. It should be pointed out that for ordinary technicians in this technical field, without departing from the principles of the present invention, the present invention can also be improved and modified, and these improvements and modifications also fall within the scope of protection of the claims of the present invention.

Claims

1. A method for visualizing autonomous learning training, characterized in that: The method comprises: S1. Build an autonomous learning and training visualization system, which includes a network model editing module, a data processing module, a hyperparameter setting module, a log module, and a data statistics line chart module; S2. Open the autonomous learning training visualization system, call the network model editing module, and use the preset network model creation method to create the target network model in the main window of the network model editing interface; S3, calling the data processing module, selecting a data source in the data loading window of the data processing module, selecting a target data set from the data source, preprocessing the target data set according to a preset data set preprocessing method, setting a data set splitting ratio, splitting the preprocessed data set into training data, test data, and verification data, and displaying them; S4, calling the hyperparameter setting module, and setting the hyperparameters for training the target network model and the training data saving path in the hyperparameter setting window of the hyperparameter setting module; S5. Use the training data, test data, verification data and set hyperparameters to train, test and verify the target network model, call the log module to form and save the training log of the target network model during the training, testing and verification process, call the data statistics line chart module to monitor the hyperparameters of the target network model during the training, testing and verification process, and output the statistical chart after the training, testing and verification are completed.

2. The autonomous learning training visualization method according to claim 1, characterized in that: The autonomous learning training visualization system also includes a network model encapsulation library, a network operator encapsulation library, and a network model storage module. S2 uses a preset network model creation method to create a target network model, specifically including: directly calling a pre-packaged classic deep neural network model from the network model encapsulation library as the target network model, selecting several network operators from the network operator encapsulation library and creating a target network model, and selecting a historically saved network model json file from the network model storage module and creating a target network model.

3. The autonomous learning training visualization method according to claim 2, characterized in that: The autonomous learning training visualization system also includes a network model analysis module, a network model parsing module, and a mindspore framework calling module. It selects several network operators from the network operator encapsulation library and creates a target network model. The specific process is as follows: S21. Click the operator library button in the main window of the network model editing interface to call the network operator encapsulation library, select several network operators from the network layer operator library sub-window and drag them to the designer sub-window of the main window of the network model editing interface; S22, switch the mouse mode to the connection mode, connect the selected network operators in the required order, and after the connection is completed, click the mouse mode button to switch the connection mode back to the mouse mode; S23, hover the mouse over the target network operator, click the target network operator and pop up the property bar of the target network operator, view and set the name and hyperparameters of the target network operator in the property bar, and obtain the initial network model; S24, click the analysis button in the main window of the network model editing interface to call the network model analysis module, which analyzes the initial network model to confirm whether there is any abnormality in the initial network model; S25. If the initial network model does not have any abnormality, click the Save button in the network model editing interface window to call the network model storage module to save the initial network model without any abnormality into a json file; S26. The network model parsing module parses the json file, and the mindspore framework calling module builds the target network model according to the parsing result of the network model parsing module.

4. The autonomous learning training visualization method according to claim 3, characterized in that: The network operator encapsulation library in S21 specifically includes 8 classification operators: convolution and fully connected layer, activation layer, input layer, data preprocessing layer, sequence layer, normalization layer, output layer, and pooling layer.

5. The autonomous learning training visualization method according to claim 4, characterized in that: S3 specifically includes the following steps: S31. Click the data button in the main window of the network model editing interface to call the data processing module, select load data in the data subwindow, select the data source in the data set selection subwindow in the load data window, and select the target data set according to the data source; S32, setting a data preprocessing option for the target data set in the data set selection subwindow according to the training requirements of the target network model, and preprocessing the target data set according to a preset data preprocessing method to obtain a preprocessed data set; S33, setting a split ratio in a data set split ratio setting subwindow in a loading data window according to the training requirements of a target network model, and splitting the preprocessed data set into training data, test data, and verification data according to the split ratio; S34, click the Submit button in the Load Dataset window, and the data processing module processes the target data set according to the selected data preprocessing method and split ratio to obtain the split information of the target data set; S35, the data statistics line chart module counts the split information of the target data set and draws a bar chart; S36. After clicking the Submit button, the loading data window is closed, and the splitting information and bar chart of the target data set are displayed in the data window.

6. The autonomous learning training visualization method according to claim 5, characterized in that: The data preprocessing methods preset in S32 include: random flipping, random rotation, random scaling, random horizontal translation, and random vertical translation.

7. The autonomous learning training visualization method according to claim 6, characterized in that: S4 specifically includes: S41. Click the training button in the main window of the network model editing interface to pop up a training sub-window. Click the hyperparameter button in the training sub-window to call the hyperparameter setting module. Set the hyperparameters for training the target network model and the path and name for saving the target network model after training in the hyperparameter setting window. S42, confirming the saving path and file name of the target model after training in the pop-up window for setting the saving path and name of the target model after training; S43. Click the Submit button in the hyperparameter setting window. If a successful submission pop-up window appears, it means that the hyperparameters and the saving path and file name settings of the trained target model are effective.

8. The autonomous learning training visualization method according to claim 7, characterized in that: S5 specifically includes: S51. Click the training button in the main window of the network model editing interface, a training sub-window pops up, and click the start training button in the training sub-window, and the target network model enters the training state; S52, training, testing and verifying the target network model using the training data, test data, verification data and set hyperparameters; S53, the log module forms a training log and saves the training information of the target network model during the training, testing and verification process; S54, the data statistics line graph module monitors the hyperparameters of the target network model during the training, testing and verification process, and records the data and draws a statistical graph according to a preset number of printing intervals; S55. After the training is completed, click the training display button to display the statistical chart in the training sub-window.

9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the autonomous learning training visualization method described in any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the autonomous learning training visualization method described in any one of claims 1 to 8 are implemented.

Images