TWI765266B - Training method, device for machine learning model and electronic device - Google Patents

Abstract

The present disclosure relates to a training method, a device for machine learning model, and an electronic device. The methods includes: receiving a problem type selected by user, determining a machine learning model corresponding to the problem type, receiving sample data entered by the user to train the machine learning model, analyzing training results of the machine learning model obtained after the machine learning model is trained, displaying the training results that meet a preset condition, and providing the machine learning model corresponding to the training results that meet the preset condition to the user. This present disclosure enables users to quickly determine a corresponding machine learning model based on the problem type, simplifies a process of selecting machine learning model for training, and improves the training efficiency of the machine learning model.

Description

Machine learning model training method, device and electronic device

The invention relates to the field of machine learning, in particular to a machine learning model training method, device and electronic equipment.

At present, because there are many types of machine learning, such as random forest, SVM, naive Bayes, knn, gbdt, xgboost, LR, etc., it is difficult for those in the field to choose a suitable machine learning model when training sample data using a machine learning model. The training result is poor and does not meet the requirements, and the model needs to be re-selected for training. However, it takes too much time and cost to repeatedly train the machine learning model, which seriously affects the training progress of the machine learning model.

In view of the above content, it is necessary to propose a machine learning model training method, device and electronic device to simplify the process of selecting a machine learning model for training and improve the training efficiency of the machine learning model.

A first aspect of the present application provides a method for training a machine learning model, the method includes: receiving a question type to be processed selected by a user; determining a machine learning model corresponding to the question type; receiving sample data pairs input by a user The determined machine learning model is trained; Analyzing the training results of the machine learning model obtained after the training is completed, displaying the training results that meet the preset conditions; and providing the machine learning model corresponding to the training results that meet the preset conditions to the user.

Preferably, the determining the machine learning model corresponding to the problem type includes: searching a type and model relationship table by using the determined problem type to determine the machine learning model corresponding to the problem type, wherein the type is related to the model The table defines the correspondence between multiple question types and multiple machine learning models.

Preferably, the sample data categories include at least a first sample category and a second sample category, and the training of the determined machine learning model by receiving the sample data input by the user includes: obtaining sample data of positive samples and negative samples and label the sample data of the positive sample with the sample category, so that the sample data of the positive sample carries the sample category label; the sample data of the positive sample and the sample data of the negative sample are randomly divided into a first preset ratio The training set and the verification set of the second preset ratio, use the training set to train the machine learning model, and use the verification set to verify the accuracy of the trained machine learning model; and if the accuracy When the accuracy rate is greater than or equal to the preset accuracy rate, the training is ended; if the accuracy rate is less than the preset accuracy rate, the number of positive samples and the number of negative samples are increased to retrain the machine learning model until the accuracy rate is greater than or equal to Preset accuracy.

Preferably, the training result is an accuracy rate, a precision rate or a network size, the preset condition corresponding to the accuracy rate is that the accuracy rate is greater than a first preset value, and the preset condition corresponding to the accuracy rate For the accuracy rate to be greater than the second preset value, the preset condition corresponding to the network size is that the number of network layers of the machine learning model is smaller than the third preset value.

Preferably, the machine learning model includes multiple convolutional layers, multiple max-pooling sampling layers, and fully-connected layers, wherein the multiple convolutional layers and the multiple max-pooling sampling layers are alternately connected, and the multiple max-pooling sampling layers are alternately connected. The max-pool sampling layer is used to perform feature extraction on the sample data to obtain feature vectors, the multiple fully-connected layers are connected to each other, and the last max-pool sampling layer in the multiple max-pool sampling layers is associated with the multiple max-pool sampling layers. The first fully-connected layer of the fully-connected layers is connected to input the feature vector obtained by feature extraction to the first fully-connected layer, and the last fully-connected layer in the multiple fully-connected layers is a classification The classifier is used for classifying the feature vector to obtain a detection result.

Preferably, receiving the sample data input by the user to train the determined machine learning model includes: establishing a training set by using the sample data and the class labels corresponding to the sample data; using the convolution layer of the machine learning model Perform convolution and sampling operations on the sample data in the training set; connect the last convolutional layer of the machine learning model to one or more fully-connected layers, wherein the fully-connected layers are configured to perform convolution and sampling Synthesize the features of the sample data extracted after the operation and output the training parameters and feature models; determine whether the machine learning model satisfies the convergence conditions, wherein, when the feature model is consistent with the preset standard feature model or the category label of the sample data It is determined that the machine learning model satisfies the convergence condition, otherwise it is determined that the machine learning model does not meet the convergence condition; when the feature model is consistent with the preset standard feature model, the feature model is output; and when the feature model does not meet the convergence condition, the back-propagation adjustment The weight matrix of the machine learning model.

A second aspect of the present application provides an apparatus for training a machine learning model, the apparatus comprising: a question type receiving module for receiving a question type to be processed selected by a user; a model determining module for determining a problem related to the question The machine learning model corresponding to the type; the training module is used to receive the sample data input by the user to train the determined machine learning model; the display module is used to analyze the training of the machine learning model obtained after the training is completed As a result, the training results that meet the preset conditions are displayed; and the model providing module is used to provide the user with the machine learning model corresponding to the training results that meet the preset conditions.

Preferably, the sample data categories include at least a first sample category and a second sample category, and the training of the determined machine learning model by receiving the sample data input by the user includes: Obtain the sample data of the positive sample and the sample data of the negative sample, and label the sample data of the positive sample with the sample category, so that the sample data of the positive sample can carry the sample category label; The sample data is randomly divided into a training set of a first preset ratio and a verification set of a second preset ratio, the training set is used to train the machine learning model, and the verification set is used to verify the validity of the trained machine learning model. accuracy; and if the accuracy is greater than or equal to the preset accuracy, end the training; if the accuracy is less than the preset accuracy, increase the number of positive samples and the number of negative samples to retrain the machine learning Model until the accuracy rate is greater than or equal to the preset accuracy rate.

Preferably, the machine learning model includes multiple convolutional layers, multiple max-pooling sampling layers, and fully-connected layers, wherein the multiple convolutional layers and the multiple max-pooling sampling layers are alternately connected, and the multiple max-pooling sampling layers are alternately connected. The max-pool sampling layer is used to perform feature extraction on the sample data to obtain feature vectors, the multiple fully-connected layers are connected to each other, and the last max-pool sampling layer in the multiple max-pool sampling layers is connected to a plurality of the fully-connected layers. The first fully-connected layer of the connection layer is connected to input the feature vector obtained by feature extraction to the first fully-connected layer, and the last fully-connected layer in the multiple fully-connected layers is a classifier. The classifier is used to classify the feature vector to obtain the detection result.

The third party aspect of the present application provides an electronic device, the electronic device includes a processor and a memory, and the processor is configured to implement the machine learning model training method when executing a computer program stored in the memory.

In this case, the type of problem to be processed selected by the user is received, the machine learning model corresponding to the problem type is determined, the sample data input by the user is received to train the determined machine learning model, and the training results that meet the preset conditions are corresponding to the training results. The machine learning model is provided to the user, so that the user can quickly determine the corresponding machine learning model according to the type of problem, which simplifies the process of selecting a machine learning model for training and improves the training efficiency of the machine learning model.

50: Question type selection interface

51: First Type Options

52: Second type option

53: Third type option

60: Type and model relationship table

40: Machine learning model training device

401: Problem type receiving module

402: Model determination module

403: Training Module

404: Display module

405: Model provides mods

406: set mod

6: Electronic equipment

61: Memory

62: Processor

63: Computer Programs

S11~S15: Steps

FIG. 1 is a flowchart of a method for training a machine learning model in an embodiment of the present invention.

FIG. 2 is a schematic diagram of a question type selection interface in an embodiment of the present invention.

FIG. 3 is a schematic diagram of a type-model relationship table in an embodiment of the present invention.

FIG. 4 is a structural diagram of an apparatus for training a machine learning model in an embodiment of the present invention.

FIG. 5 is a schematic diagram of an electronic device in an embodiment of the present invention.

In order to more clearly understand the above objects, features and advantages of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments may be combined with each other in the case of no conflict.

In the following description, many specific details are set forth in order to facilitate a full understanding of the present invention, and the described embodiments are only some, but not all, embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. The terms used herein in the description of the present invention are for the purpose of describing specific embodiments only, and are not intended to limit the present invention.

Preferably, the machine learning model training method of the present invention is applied in one or more electronic devices. The electronic device is a device that can automatically perform numerical calculations and/or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application specific integrated circuits (ASICs) , Programmable Gate Array (Field-Programmable Gate Array, FPGA), Digital Signal Processor (Digital Signal Processor, DSP), embedded devices, etc.

The electronic device may be a computing device such as a desktop computer, a notebook computer, a tablet computer, and a cloud server. The device can interact with the user by means of a keyboard, a mouse, a remote control, a touch pad or a voice control device.

Example 1

FIG. 1 is a flowchart of a method for training a machine learning model in an embodiment of the present invention. According to different requirements, the order of the steps in the flowchart can be changed, and some steps can be omitted.

Referring to Figure 1, the machine learning model training method specifically includes the following steps:

Step S11, receiving the question type to be handled selected by the user.

In this embodiment, the problem type is the type corresponding to the problem that can be solved by the machine learning model. Specifically, the problem types include object classification in images, segmentation of images, and object detection in images.

In this embodiment, the receiving a user-selected question type to be handled includes: receiving a user-selected question type to be handled through the question type selection interface 50 . Please refer to FIG. 2 , which is a schematic diagram of a question type selection interface 50 in an embodiment of the present invention. The question type selection interface 50 includes a first type option 51 , a second type option 52 and a third type option 53 . Among them, the first type option 51 is used for the user to select the object classification in the image, the second type option 52 is used for the user to select the segmentation of the image, and the third type option 53 is used for the user to select the image in the image. object detection. That is, when the user selects the first type option 51, the question type selection interface 50 receives the object classification in the image selected by the user as the question type; when the user selects the second type option 52, the question type The type selection interface 50 receives the segmentation of the image selected by the user as the question type; when the user selects the third type option 53 , the question type selection interface 50 receives the object detection in the image selected by the user as the question type.

Step S12, determining a machine learning model corresponding to the problem type.

In this embodiment, the determining the machine learning model corresponding to the problem type includes: searching the type-model relationship table 60 for the determined problem type to determine the machine learning model corresponding to the problem type, wherein the type The relationship with the model table 60 defines the corresponding relationship between a plurality of question types and a plurality of machine learning models. Please refer to FIG. 3 , which is a schematic diagram of a type-model relationship table 60 in an embodiment of the present invention. The type and model relationship table 60 defines the problem type as the correspondence between the object classification in the image and the machine learning model as the first model, wherein the first model can be the image Object classification model; define the problem type as the correspondence between image segmentation and machine learning model as the second model, where the second model can be an image segmentation model; define the problem type as object detection in images and machine learning The model is the corresponding relationship of the third model, wherein the third model may be an object detection model in the image. In this embodiment, the first model, the second model, and the third model are at least one model among random forest, SVM, Naive Bayes, knn, gbdt, xgboost, and LR.

In another embodiment, the determining the machine learning model corresponding to the question type includes: receiving a machine learning model input by a user, and determining the received machine learning model as the machine learning model corresponding to the question type . In this embodiment, the machine learning model input by the user is determined as the machine learning model corresponding to the problem type, so that professional R&D personnel can add more machine learning models for sample training according to their own plans.

Step S13, receiving the sample data input by the user to train the determined machine learning model.

In this embodiment, the sample data category includes at least two different categories: a first sample category and a second sample category. The training of the determined machine learning model by receiving the sample data input by the user includes:

1) Obtain the sample data of the positive sample and the sample data of the negative sample, and label the sample data of the positive sample with the sample category, so that the sample data of the positive sample can carry the sample category label.

For example, to select 500 sample data corresponding to the first sample category and the second sample category, and label each sample data with a category, "1" can be used as the sample data label of the first sample category, and "2" can be used as the sample data label of the first sample category. As the sample data label of the second sample category.

2) randomly dividing the sample data of the positive sample and the sample data of the negative sample into a training set of a first preset ratio and a verification set of a second preset ratio, and using the training set to train the machine learning model, And use the verification set to verify the accuracy of the trained machine learning model.

First distribute the training samples in the training set of different sample categories to different folders. For example, the training samples of the first sample category are distributed to the first folder, and the training samples of the second sample category are distributed to the second folder. Then extract the first preset ratio from different folders (for example For example, 70%) of the training samples are used as the total training samples to train the machine learning model, and the remaining second preset proportion (for example, 30%) of the training samples are taken from different folders as the total test samples for training. The completed machine learning model is verified for accuracy.

3) If the accuracy rate is greater than or equal to the preset accuracy rate, end the training; if the accuracy rate is less than the preset accuracy rate, increase the number of positive samples and the number of negative samples to retrain the machine learning model until The accuracy rate is greater than or equal to a preset accuracy rate.

In this embodiment, the machine learning model includes multiple convolutional layers, max-pooling sampling layers, and fully-connected layers. Among them, the convolution layer and the maximum pool sampling layer are alternately connected, and are used for feature extraction of the sample data to obtain feature vectors. The plurality of fully connected layers are connected to each other. The last max-pool sampling layer in the multiple max-pool sampling layers is connected to the first fully-connected layer of the multiple fully-connected layers, and is used for inputting the feature vector obtained through feature extraction to the first fully-connected layer , the last fully connected layer of the multiple fully connected layers is a classifier, and the classifier is used to classify the feature vector to obtain a detection result.

In one embodiment, receiving the sample data input by the user to train the determined machine learning model includes: establishing a training set using sample data (such as image data) and class labels corresponding to the sample data; Perform convolution and sampling operations on the sample data in the training set through the convolutional layer of the machine learning model; connect the last convolutional layer to one or more fully connected layers, where the fully connected layer is configured to perform convolution and sampling operations The features of the sample data extracted after the operation are synthesized and the training parameters and the feature model are output, wherein the feature model is an abstract feature expression of the sample data; judging whether the machine learning model satisfies the convergence condition, that is, judging whether the feature model is consistent with the preset Whether the standard feature model of the sample data or the class label of the sample data are consistent, wherein, when the feature model is consistent with the preset standard feature model or the class label of the sample data, it is determined that the machine learning model satisfies the convergence condition, otherwise it is determined that the machine learning model The convergence condition is not satisfied; when the characteristic model is consistent with the preset standard characteristic model, the characteristic model is output; and when the characteristic model does not satisfy the convergence condition, the weight matrix of the machine learning model is adjusted by backpropagation.

In this embodiment, in the training process of the machine learning model, if there is an error between the output feature model and the standard feature model, the error information is transmitted back along the original path by backpropagation, so as to correct each layer (for example, The training parameters of the convolution layer and the sampling layer), for example, the training parameters can include weighting values and biases, and then use the revised convolution layer and the sampling layer to perform convolution and sampling operations on the training data again until the feature model satisfies the end condition. . In this embodiment, when performing the convolution operation, multiple feature maps can be applied to the sample data to obtain different features of the sample data, wherein each feature map extracts one feature of the sample data. In this embodiment, methods such as mean value combination, maximum value combination, and random combination can be used to process the data in the training set when performing the sampling operation.

Step S14, analyzing the training results of the machine learning model obtained after the training is completed, and displaying the training results that meet the preset conditions.

In this embodiment, the training result is an accuracy rate, a precision rate, or a network size. The preset conditions corresponding to the accuracy rate, precision rate or network size are: the accuracy rate is greater than the first preset value (eg 90%), the accuracy rate is greater than the second preset value (eg 85%), the machine learning model The number of network layers is less than the third preset value (eg 100). In this embodiment, when the accuracy of the machine learning model obtained after training is greater than the first preset value, the accuracy of the machine learning model is displayed, or when the accuracy of the machine learning model obtained after training is greater than When the second preset value is used, the accuracy rate of the machine learning model is displayed, or when the number of network layers of the machine learning model obtained after training is less than the third preset value, the network layer of the machine learning model is displayed. number is displayed.

In this embodiment, the accuracy of the machine learning model is calculated according to the formula A=(TP+TN)/(TP+FP+FN+TN), where A is the accuracy, and TP is predicted by the machine learning model as The number of positive positive samples, FP is the number of negative samples predicted as positive by the machine learning model, FN is the number of positive samples predicted as negative by the machine learning model, and TN is the number of negative samples predicted as negative by the machine learning model. In this embodiment, the accuracy of the machine learning model is calculated according to the formula P=TP/(TP+FP).

In step S15, the machine learning model corresponding to the training result that meets the preset condition is provided to the user.

In this embodiment, the method further includes: when the training result of the machine learning model does not meet the preset conditions, receiving an operation of modifying the parameters of the machine learning model by the user, and retraining the machine learning model according to the modified parameters. described machine learning model. In a specific embodiment, a parameter setting interface (not shown in the figure) is provided, and the user's operation of modifying the parameters of the machine learning model is received through the parameter setting interface, and the modified parameters are retrained. described machine learning model.

In this embodiment, the method further includes: receiving a user's modification operation on the correspondence between the question type and the machine learning model in the type-model relationship table 60, and setting the question type and the machine learning model according to the modification operation Correspondence to the model.

In this embodiment, the problem type to be processed selected by the user is received, the machine learning model corresponding to the problem type is determined, the sample data input by the user is received to train the determined machine learning model, and the determined machine learning model will meet the preset conditions. The machine learning model corresponding to the training result is provided to the user, so that the user can quickly determine the corresponding machine learning model according to the type of problem, which simplifies the process of selecting a machine learning model for training and improves the training efficiency of the machine learning model. .

Example 2

FIG. 4 is a structural diagram of a machine learning model training apparatus 40 in an embodiment of the present invention.

In some embodiments, the machine learning model training apparatus 40 runs in an electronic device. The machine learning model training apparatus 40 may include a plurality of functional modules composed of program code segments. The code of each program segment in the machine learning model training device 40 can be stored in the memory and executed by at least one processor to perform the function of machine learning model training.

In this embodiment, the machine learning model training apparatus 40 can be divided into a plurality of functional modules according to the functions performed by the machine learning model training apparatus 40 . Referring to FIG. 4 , the machine learning model training apparatus 40 may include a question type receiving module 401 , a model determining module 402 , a training module 403 , a display module 404 , a model providing module 405 and a setting module 406 . The module referred to in the present invention refers to a module that can be used by at least one A series of computer program segments executed by a processor and capable of performing fixed functions, which are stored in memory. In some embodiments, the functions of each module will be described in detail in subsequent embodiments.

The question type receiving module 401 receives the question type to be processed selected by the user.

In this embodiment, the problem type is the type corresponding to the problem that can be solved by the machine learning model. Specifically, the problem types include object classification in images, segmentation of images, and object detection in images.

In this embodiment, the question type receiving module 401 receiving the question type selected by the user to be handled includes: receiving the question type to be handled selected by the user through the question type selection interface 50 . The question type selection interface 50 includes a first type option 51 , a second type option 52 and a third type option 53 . Among them, the first type option 51 is used for the user to select the object classification in the image, the second type option 52 is used for the user to select the segmentation of the image, and the third type option 53 is used for the user to select the image in the image. object detection. That is, when the user selects the first type option 51, the question type selection interface 50 receives the object classification in the image selected by the user as the question type; when the user selects the second type option 52, the question type The type selection interface 50 receives the segmentation of the image selected by the user as the question type; when the user selects the third type option 53 , the question type selection interface 50 receives the object detection in the image selected by the user as the question type.

The model determination module 402 determines a machine learning model corresponding to the problem type.

In this embodiment, the model determining module 402 determines the machine learning model corresponding to the problem type includes: searching the type-model relationship table 60 by using the determined problem type to determine the machine learning model corresponding to the problem type , wherein the type-model relationship table 60 defines the corresponding relationship between multiple problem types and multiple machine learning models. The type and model relationship table 60 defines the problem type as the correspondence between the object classification in the image and the machine learning model as the first model, wherein the first model can be an image object classification model; define the problem type as an image The corresponding relationship between the segmentation and the machine learning model is the second model, wherein the second model can be the segmentation model of the image; define the question The question type is the correspondence between the object detection in the image and the machine learning model being the third model, where the third model may be the object detection model in the image. In this embodiment, the first model, the second model, and the third model are at least one model among random forest, SVM, Naive Bayes, knn, gbdt, xgboost, and LR.

In another embodiment, the model determining module 402 determining the machine learning model corresponding to the question type includes: receiving the machine learning model input by the user, and determining the received machine learning model as corresponding to the question type The corresponding machine learning model. In this embodiment, the machine learning model input by the user is determined as the machine learning model corresponding to the problem type, so that professional R&D personnel can add more machine learning models for sample training according to their own plans.

The training module 403 receives the sample data input by the user to train the determined machine learning model.

In this embodiment, the sample data category includes at least two different categories: a first sample category and a second sample category. The training module 403 receives the sample data input by the user to train the determined machine learning model, including:

1) Obtain the sample data of the positive sample and the sample data of the negative sample, and label the sample data of the positive sample with the sample category, so that the sample data of the positive sample can carry the sample category label.

For example, to select 500 sample data corresponding to the first sample category and the second sample category, and label each sample data with a category, "1" can be used as the sample data label of the first sample category, and "2" can be used as the sample data label of the first sample category. As the sample data label of the second sample category.

2) randomly dividing the sample data of the positive sample and the sample data of the negative sample into a training set of a first preset ratio and a verification set of a second preset ratio, and using the training set to train the machine learning model, And use the verification set to verify the accuracy of the trained machine learning model.

First distribute the training samples in the training set of different sample categories to different folders. For example, the training samples of the first sample category are distributed to the first folder, and the training samples of the second sample category are distributed to the second folder. Then, the training samples of the first preset proportion (for example, 70%) are respectively extracted from different folders as the total training samples to train the machine learning model. The remaining second preset proportion (for example, 30%) of the training samples in the data folder are respectively taken as the total test samples to verify the accuracy of the trained machine learning model.

3) If the accuracy rate is greater than or equal to the preset accuracy rate, end the training; if the accuracy rate is less than the preset accuracy rate, increase the number of positive samples and the number of negative samples to retrain the machine learning model until The accuracy rate is greater than or equal to a preset accuracy rate.

In this embodiment, the machine learning model includes multiple convolutional layers, max-pooling sampling layers, and fully-connected layers. Among them, the convolution layer and the maximum pool sampling layer are alternately connected, and are used for feature extraction of the sample data to obtain feature vectors. The plurality of fully connected layers are connected to each other. The last max-pool sampling layer in the multiple max-pool sampling layers is connected to the first fully-connected layer of the multiple fully-connected layers, and is used for inputting the feature vector obtained through feature extraction to the first fully-connected layer , the last fully connected layer of the multiple fully connected layers is a classifier, and the classifier is used to classify the feature vector to obtain a detection result.

In one embodiment, the training module 403 uses sample data (such as image data) and class labels corresponding to the sample data to create a training set; the sample data in the training set is rolled through the convolution layer of the machine learning model. Product operation and sampling operation; connect the last convolutional layer to one or more fully connected layers, where the fully connected layer is configured to synthesize the features of the sample data extracted after the convolution and sampling operations and output the training parameters and feature model, wherein the feature model is an abstract feature expression of the sample data; judging whether the machine learning model satisfies the convergence condition, that is, judging whether the feature model is consistent with the preset standard feature model or the category label of the sample data, wherein , when the feature model is consistent with the preset standard feature model or the category label of the sample data, it is determined that the machine learning model satisfies the convergence condition, otherwise it is determined that the machine learning model does not meet the convergence condition; when the feature model is consistent with the preset standard feature model When they are consistent, the feature model is output; and when the feature model does not meet the convergence condition, the weight matrix of the machine learning model is adjusted by backpropagation.

In this embodiment, in the training process of the machine learning model, if there is an error between the output feature model and the standard feature model, the training module 403 backpropagates the error information along the original path by backpropagation, Thus, the training parameters of each layer (for example, the convolutional layer and the sampling layer) are modified, and the training The training parameters may include, for example, weighted values and biases, and then the training data are re-convolutional and sampled by using the revised convolutional layer and sampling layer until the feature model satisfies the end condition. In this embodiment, when performing the convolution operation, multiple feature maps can be applied to the sample data to obtain different features of the sample data, wherein each feature map extracts one feature of the sample data. In this embodiment, methods such as mean value combination, maximum value combination, and random combination can be used to process the data in the training set when performing the sampling operation.

The display module 404 analyzes the training results of the machine learning model obtained after the training is completed, and displays the training results that meet the preset conditions.

In this embodiment, the training result is an accuracy rate, a precision rate, or a network size. The preset conditions corresponding to the accuracy rate, precision rate or network size are: the accuracy rate is greater than the first preset value (eg 90%), the accuracy rate is greater than the second preset value (eg 85%), the machine learning model The number of network layers is less than the third preset value (eg 100). In this embodiment, the display module 404 displays the accuracy of the machine learning model when the accuracy of the machine learning model obtained after the training is greater than the first preset value, or when the accuracy of the machine learning model obtained after the training is completed is displayed. When the accuracy rate of the learning model is greater than the second preset value, the accuracy rate of the machine learning model is displayed, or when the number of network layers of the machine learning model obtained after training is less than the third preset value, the machine learning model is displayed. The number of network layers of the learning model is displayed.

In this embodiment, the accuracy of the machine learning model is calculated according to the formula A=(TP+TN)/(TP+FP+FN+TN), where A is the accuracy, and TP is predicted by the machine learning model as The number of positive positive samples, FP is the number of negative samples predicted as positive by the machine learning model, FN is the number of positive samples predicted as negative by the machine learning model, and TN is the number of negative samples predicted as negative by the machine learning model. In this embodiment, the accuracy of the machine learning model is calculated according to the formula P=TP/(TP+FP).

The model providing module 405 provides the user with the machine learning model corresponding to the training result that meets the preset condition.

In this embodiment, the training module 403 is further configured to receive the user to modify the parameters of the machine learning model when the training result of the machine learning model does not meet the preset conditions operation, and retrain the machine learning model according to the modified parameters. In a specific embodiment, a parameter setting interface (not shown in the figure) is provided, and the user's operation of modifying the parameters of the machine learning model is received through the parameter setting interface, and the modified parameters are retrained. described machine learning model.

In this embodiment, the setting module 406 is configured to: receive a user's modification operation on the correspondence between the question type and the machine learning model in the type-model relationship table 60, and set the question type according to the modification operation Correspondence with machine learning models.

In this embodiment, the problem type to be processed selected by the user is received, the machine learning model corresponding to the problem type is determined, the sample data input by the user is received to train the determined machine learning model, and the determined machine learning model will meet the preset conditions. The machine learning model corresponding to the training result is provided to the user, so that the user can quickly determine the corresponding machine learning model according to the type of problem, which simplifies the process of selecting a machine learning model for training and improves the training efficiency of the machine learning model. .

Example 3

FIG. 5 is a schematic diagram of an electronic device 6 in an embodiment of the present invention.

The electronic device 6 includes a memory 61 , a processor 62 and a computer program 63 stored in the memory 61 and executable on the processor 62 . When the processor 62 executes the computer program 63 , the steps in the above-mentioned embodiment of the machine learning model training method are implemented, for example, steps S11 to S15 shown in FIG. 1 . Alternatively, when the processor 62 executes the computer program 63 , the functions of each module/unit in the above-mentioned embodiment of the machine learning model training apparatus, such as modules 401 to 406 in FIG. 4 , are realized.

Exemplarily, the computer program 63 can be divided into one or more modules/units, and the one or more modules/units are stored in the memory 61 and executed by the processor 62 , to complete the present invention. The one or more modules/units may be a series of computer program instruction segments capable of performing specific functions, and the instruction segments are used to describe the execution process of the computer program 63 in the electronic device 6 . For example, the computer program 63 can be divided into the question type receiving module 401, the model determining module 402, the training module 403, the display module 404, the model providing module 405 and the setting module 406 in FIG. For the specific functions of the module, see Embodiment 2.

In this embodiment, the electronic device 6 may also be a computing device such as a desktop computer, a notebook, a palmtop computer, and a cloud terminal device. Those skilled in the art can understand that the schematic diagram is only an example of the electronic device 6, and does not constitute a limitation to the electronic device 6, and may include more or less components than the one shown, or combine some components, or different Components such as the electronic device 6 may also include input and output devices, network access devices, bus bars, and the like.

The processor 62 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs) ), Field-Programmable Gate Array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor 62 can also be any conventional processor, etc. The processor 62 is the control center of the electronic device 6, and uses various interfaces and lines to connect the entire electronic device 6. various parts.

The memory 61 can be used to store the computer programs 63 and/or modules/units, and the processor 62 runs or executes the computer programs and/or modules/units stored in the memory 61, And call the data stored in the memory 61 to realize various functions of the electronic device 6 . The memory 61 may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; storage data The area may store data (such as audio data, phone book, etc.) created according to the use of the electronic device 6, and the like. In addition, the memory 61 may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, Smart Media Card (SMC), Secure Digital (Secure Digital, SD) card, flash memory card (Flash Card), at least one disk memory device, flash memory device, or other volatile solid state memory device.

If the modules/units integrated in the electronic device 6 are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the present invention can implement all or part of the processes in the methods of the above embodiments, and can also be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage In the medium, when the computer program is executed by the processor, the steps of the above-mentioned method embodiments can be implemented. Wherein, the computer program includes computer program code, and the computer program code may be in the form of original code, object code, executable file, or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, removable hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM, Read-only memory) Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content contained in the computer-readable medium may be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction, for example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium Electric carrier signals and telecommunication signals are not included.

In the several embodiments provided by the present invention, it should be understood that the disclosed electronic devices and methods may be implemented in other manners. For example, the above-described electronic device embodiments are only illustrative. For example, the division of the modules is only a logical function division, and other division methods may be used in actual implementation.

In addition, each functional module in each embodiment of the present invention may be integrated in the same processing module, or each module may exist physically alone, or two or more modules may be integrated in the same module. The above-mentioned integrated modules can be implemented in the form of hardware, or can be implemented in the form of hardware plus software function modules.

It will be apparent to those skilled in the art that the present invention is not limited to the details of the above-described exemplary embodiments, but that the present invention may be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments are to be regarded in all respects as illustrative and not restrictive, and the scope of the invention is to be defined by the appended claims rather than the foregoing description, which are therefore intended to fall within the scope of the claims. All changes within the meaning and range of the equivalents of , are included in the present invention. Any reference signs in the claims shall not be construed as limiting the involved claim. Furthermore, it is clear that the word "comprising" does not exclude other modules or steps, and the singular does not exclude the plural. A plurality of modules or electronic devices recited in the electronic device claims can also be realized by one and the same module or electronic device by means of software or hardware. The terms first, second, etc. are used to denote names and do not denote any particular order.

To sum up, the present invention complies with the requirements of an invention patent, and a patent application can be filed in accordance with the law. However, the above descriptions are only the preferred embodiments of the present invention, and for those who are familiar with the techniques of this case, equivalent modifications or changes made in accordance with the creative spirit of this case shall be included in the scope of the following patent application.

S11~S15: Steps

Claims (10)

A machine learning model training method, which is improved in that the method comprises: receiving a question type to be processed selected by a user; determining a machine learning model corresponding to the question type; The machine learning model is trained; the training results of the machine learning model obtained after training are analyzed, and the training results that meet the preset conditions are displayed, wherein the training results include the accuracy rate, and the preset conditions are the The accuracy rate is greater than the first preset value, and the accuracy rate is calculated according to the formula A=(TP+TN)/(TP+FP+FN+TN), where A is the accuracy rate, and TP is predicted by the machine learning model as The number of positive positive samples, FP is the number of negative samples predicted as positive by the machine learning model, FN is the number of positive samples predicted as negative by the machine learning model, TN is the number of negative samples predicted as negative by the machine learning model; and The machine learning model corresponding to the training result that meets the preset conditions is provided to the user. The method for training a machine learning model according to claim 1, wherein the determining the machine learning model corresponding to the problem type comprises: searching a type-model relationship table by using the determined problem type to determine the problem type corresponding to the problem type The machine learning model, wherein the type and model relationship table defines the corresponding relationship between multiple problem types and multiple machine learning models. The method for training a machine learning model according to claim 1, wherein the sample data categories include at least a first sample category and a second sample category, and the sample data received by the user is used to perform a training procedure on the determined machine learning model. The training includes: obtaining the sample data of the positive samples and the sample data of the negative samples, and labeling the sample data of the positive samples with the sample category, so that the sample data of the positive samples carry the sample category labels; The sample data of the negative samples are randomly divided into a training set of a first preset ratio and a verification set of a second preset ratio, use the training set to train the machine learning model, and use the verification set to verify the trained machine the accuracy of the learned model; and If the accuracy rate is greater than or equal to the preset accuracy rate, end the training; if the accuracy rate is less than the preset accuracy rate, increase the number of positive samples and the number of negative samples to retrain the machine learning model until the The accuracy rate is greater than or equal to the preset accuracy rate. The machine learning model training method according to claim 1, wherein the training result is an accuracy rate, a precision rate or a network size, and a preset condition corresponding to the accuracy rate is that the accuracy rate is greater than a first preset The preset condition corresponding to the accuracy rate is that the accuracy rate is greater than the second preset value, and the preset condition corresponding to the network size is that the number of network layers of the machine learning model is less than the third preset value. set value. The machine learning model training method according to claim 1, wherein the machine learning model includes multiple convolutional layers, multiple max-pooling sampling layers, and fully connected layers, wherein the multiple convolutional layers and the multiple The max-pool sampling layers are alternately connected, the multiple max-pool sampling layers are used to perform feature extraction on the sample data to obtain feature vectors, the multiple fully-connected layers are connected to each other, and the multiple max-pool sampling layers are The last max-pool sampling layer is connected to the first fully-connected layers of the multiple fully-connected layers, and is used to input the feature vector obtained through feature extraction to the first fully-connected layer. The last fully connected layer is a classifier, and the classifier is used to classify the feature vector to obtain the detection result. The method for training a machine learning model according to claim 5, wherein the receiving the sample data input by the user to train the determined machine learning model comprises: using the sample data and the class labels corresponding to the sample data to establish training Convolution operation and sampling operation are performed on the sample data in the training set by the convolution layer of the machine learning model; the last convolution layer of the machine learning model is connected to one or more fully connected layers, wherein The fully connected layer is configured to synthesize the features of the sample data extracted after the convolution and sampling operations, and output the training parameters and the feature model; to determine whether the machine learning model satisfies the convergence condition, wherein, when the feature model matches the preset standard When the feature model or the category label of the sample data is consistent, it is determined that the machine learning model satisfies the convergence conditions, otherwise it is determined that the machine learning model does not meet the convergence conditions; when the feature model and the preset When the standard feature model of the machine learning model is consistent, the feature model is output; and when the feature model does not meet the convergence condition, the weight matrix of the machine learning model is adjusted by backpropagation. A machine learning model training device, which is improved in that the device comprises: a question type receiving module for receiving a question type to be processed selected by a user; a model determining module for determining a question type corresponding to the question type The training module is used to receive the sample data input by the user to train the determined machine learning model; the display module is used to analyze the training results of the machine learning model obtained after the training is completed, and The training results that meet the preset conditions are displayed, wherein the training results include the accuracy rate, the preset condition is that the accuracy rate is greater than the first preset value, and the accuracy rate is based on the formula A=(TP+TN) /(TP+FP+FN+TN), where A is the accuracy rate, TP is the number of positive samples predicted as positive by the machine learning model, FP is the number of negative samples predicted as positive by the machine learning model, and FN is The number of positive samples predicted to be negative by the machine learning model, TN is the number of negative samples predicted to be negative by the machine learning model; and the model providing module is used to provide the machine learning model corresponding to the training results that meet the preset conditions to the user. The machine learning model training device according to claim 7, wherein the sample data categories include at least a first sample category and a second sample category, and the sample data inputted by the user is used to perform training on the determined machine learning model. The training includes: obtaining the sample data of the positive samples and the sample data of the negative samples, and labeling the sample data of the positive samples with the sample category, so that the sample data of the positive samples carry the sample category labels; The sample data of the negative samples are randomly divided into a training set of a first preset ratio and a verification set of a second preset ratio, use the training set to train the machine learning model, and use the verification set to verify the trained machine the accuracy of the learned model; and If the accuracy rate is greater than or equal to the preset accuracy rate, end the training; if the accuracy rate is less than the preset accuracy rate, increase the number of positive samples and the number of negative samples to retrain the machine learning model until the The accuracy rate is greater than or equal to the preset accuracy rate. The apparatus for training a machine learning model according to claim 7, wherein the machine learning model comprises multiple convolutional layers, multiple max-pooling sampling layers, and fully connected layers, wherein the multiple convolutional layers and the multiple The max-pool sampling layers are alternately connected, the multiple max-pool sampling layers are used to perform feature extraction on the sample data to obtain feature vectors, the multiple fully-connected layers are connected to each other, and the multiple max-pool sampling layers are The last max-pool sampling layer is connected to the first fully-connected layers of the multiple fully-connected layers, and is used to input the feature vector obtained through feature extraction to the first fully-connected layer. The last fully connected layer is a classifier, and the classifier is used to classify the feature vector to obtain the detection result. An electronic device, which is improved in that the electronic device includes a processor and a memory, and the processor is used to implement the machine according to any one of claims 1 to 6 when executing a computer program stored in the memory Learn about model training methods.

Images