CN115952208A - A method, device and electronic equipment for calculating the degree of association between features and tags - Google Patents

Abstract

Embodiments of the present application provide a method, device, and electronic device for calculating the degree of association between features and tags, which are applied in the field of information technology. By applying the method of the present application, the degree of correlation between the target feature and the target label can be detected on the trained model to be detected, and the detected degree of correlation is related and accurate to the model to be detected. By inputting the verification data set into the model to be detected, the target feature value in the verification data set is perturbed multiple times, and the value of the target label is predicted, which realizes the automatic analysis of the correlation between the target feature and the target label in the model to be detected, avoiding manual Do input calculations. When detecting the correlation degree between the target label and the target feature, what is obtained is a correlation degree that can directly represent the influence of the target feature on the target label, without predicting the influence of all features on the target label. Therefore, applying the method of this application, also Can greatly reduce the waste of computing power.

Description

A method, device and electronic equipment for calculating the degree of association between features and tags

technical field

The present application relates to the field of information technology, in particular to a method, device and electronic equipment for calculating the degree of association between features and tags.

Background technique

With the increasing application of artificial intelligence, it is traditionally believed that the deep neural network does not need feature screening, and can load all the features into the deep model, so that the model can learn automatically. For important features, its weight is higher; for Unimportant features have lower weights. However, the calculation of full features requires a lot of computing power. For an actual model, full training is usually carried out based on the latest training data every day. Therefore, not doing any feature screening will cause a lot of waste of computing power.

In order to reduce the waste of computing power, in related technologies, people often calculate the Pearson correlation coefficient before the feature and the label to get the correlation between the feature and the label, or delete a certain feature, retrain the model, and then according to the actual effect, inversely deduces whether the feature is important.

However, calculating the Pearson correlation coefficient before the feature and the label can only obtain the common mathematical correlation in theory, and the real environment is more complicated. Therefore, the method of calculating the Pearson correlation coefficient before the feature and the label does not Applicable to realistic models.

Another method is to judge the correlation between features and labels by the final effect of the model after adding or subtracting features. Although this method can be applied in actual scenarios, when the sample contains multiple features, it is necessary to add or subtract each feature. To make predictions, and compare multiple prediction results to determine the weight of the target feature to the target label in the model to be detected. The advantage of this method is that the obtained correlation degree is verified by practice and can be applied in actual models, such as models for predicting real estate quotas, models for predicting video playback traffic, and other network models. However, if you want to analyze the correlation between all features and the target label, you usually need to train the data after removing a certain feature. For example, if there are N features, you need to train N times; and usually N>50, so the application is increased or decreased. The method of judging the correlation between features and labels based on the final effect of the post-feature model will still consume a lot of computing power.

Contents of the invention

The purpose of the embodiments of the present application is to provide a method, device and electronic device for calculating the degree of association between features and tags, so as to obtain the degree of association between features and tags with low computing power consumption. The specific technical scheme is as follows:

In the first aspect of the implementation of the present application, a method for calculating the degree of association between features and tags is firstly provided, the method comprising:

Receive detection parameters from the client component; wherein the detection parameters include parameters required to calculate the degree of association between the target feature and the target label;

According to the detection parameters, call the model to be detected; wherein, the model to be detected is used to predict the value of the target label according to the input target characteristics;

Obtain a verification data set through the detection parameters; wherein, the verification data set includes the value of the target feature, and the verification data set is used to input into the model to be detected;

Inputting the verification data set into the model to be tested to obtain a first prediction result;

Modifying the value of the target feature in the verification data set according to the detection parameters to obtain a modified verification data set;

Inputting the modified verification data set into the model to be tested for prediction to obtain a second prediction result;

Calculate the degree of association between the target feature and the target label in the model to be detected according to the second prediction result and the first prediction result.

In a possible implementation manner, the detection parameters include a disturbance method and a disturbance number; wherein, the disturbance method indicates a method of modifying the value of the target feature, and the number of disturbances indicates the method of modifying the value of the target feature The number of times, the value of the target feature in the verification data set is modified according to the detection parameters, and the modified verification data set is obtained, including:

According to the perturbation method, modify the value of the target feature in each piece of verification data in the verification data set to obtain a modified verification data set;

According to the number of disturbances, repeating the step of modifying the value of the target feature in each piece of verification data in the verification data set according to the disturbance method, to obtain a modified verification data set having the same number of disturbance times as the number of disturbances .

In a possible implementation manner, the perturbation method is perturbation according to a normal distribution; according to the perturbation method, modifying the value of the target feature in each piece of verification data in the verification data set includes:

Calculate the mean value and variance value when the value of the target feature is normally distributed in the verification data set, and obtain the normal distribution curve of the value of the target feature in the verification data set;

Randomly change the value of the target feature in each piece of verification data along the normal distribution curve.

In a possible implementation manner, the calculating the correlation degree between the target feature and the target label in the model to be detected according to the second prediction result and the first prediction result includes:

Calculate the average value of the difference between the second prediction result and the first prediction result to obtain the correlation degree between the target feature and the target label in the model to be detected.

In a possible implementation manner, the method also includes:

The second prediction result of each prediction is fed back to the client component as an intermediate result.

In a possible implementation manner, the method also includes:

storing the degree of association, and/or, the intermediate result in a database;

Receive query information from the client component, the query information includes the target feature to be queried and the target tag name;

The correlation degree corresponding to the target feature and the target tag name is searched from the database, and the query result including the correlation degree is fed back to the client component.

In the second aspect of the implementation of the present application, a device for calculating the degree of association between features and tags is also provided, the device comprising:

A detection parameter receiving module, configured to receive detection parameters from the client component; wherein the detection parameters include parameters required to calculate the degree of association between the target feature and the target tag;

The model to be detected calling module is used to call the model to be detected according to the detection parameters; wherein the model to be detected is used to predict the value of the target label according to the input target characteristics;

A verification data set acquisition module, configured to obtain a verification data set through the detection parameters; wherein, the verification data set includes the value of the target feature, and the verification data set is used for inputting into the model to be detected;

A first prediction result obtaining module, configured to input the verification data set into the model to be tested to obtain a first prediction result;

A verification data set modification module, configured to modify the value of the target feature in the verification data set according to the detection parameters, to obtain a modified verification data set;

A second prediction result obtaining module, configured to input the modified verification data set into the model to be tested for prediction, and obtain a second prediction result;

A degree of association calculation module, configured to calculate the degree of association between the target feature and the target label in the model to be detected according to the second prediction result and the first prediction result.

In a possible implementation manner, the detection parameters include a disturbance method and a disturbance number; wherein, the disturbance method indicates a method of modifying the value of the target feature, and the number of disturbances indicates the method of modifying the value of the target feature Number of times, the verification data set modification module includes:

Modifying the verification data set sub-module, specifically for modifying the value of the target feature in each piece of verification data in the verification data set according to the perturbation method, to obtain a modified verification data set;

The repeated modification submodule is specifically used to repeat the step of modifying the value of the target feature in each piece of verification data in the verification data set according to the perturbation method according to the perturbation times to obtain the same number of perturbations as the perturbation times. The modified validation dataset of the number.

In a possible implementation manner, the perturbation method is perturbation according to a normal distribution; the submodule of modifying the verification data set includes:

The normal distribution curve calculation unit is specifically used to calculate the mean value and variance value when the value of the target feature is normally distributed in the verification data set, and obtain the value of the target feature in the verification data set. the normal distribution curve within the dataset;

The normal distribution modifying unit is specifically used to randomly change the value of the target feature in each piece of verification data along the normal distribution curve.

In a possible implementation manner, the association degree calculation module includes:

The average value calculation sub-module is specifically used to calculate the average value of the difference between the second prediction result and the first prediction result to obtain the correlation degree between the target feature and the target label in the model to be detected.

In a possible implementation manner, the device also includes:

An intermediate result feedback module, configured to feed back the second prediction result of each prediction to the client component as an intermediate result.

In a possible implementation manner, the device also includes:

A feedback result storage module, configured to store the degree of association and/or the intermediate result in a database;

The query information receiving module is used to receive the query information from the client component, the query information includes the target feature to be queried and the target tag name;

The query result feedback module is configured to find the correlation degree corresponding to the target feature and the target tag name from the database, and feed back the query result including the correlation degree to the client component.

In a third aspect of the embodiments of the present application, an electronic device is provided, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory complete communication with each other through the communication bus;

memory for storing computer programs;

The processor is configured to implement any one of the method steps performed in the first aspect of the embodiment of the present application when executing the program stored in the memory.

In yet another aspect of the implementation of the present application, a computer-readable storage medium is also provided, wherein a computer program is stored in the computer-readable storage medium, and when the computer program is executed by a processor, the implementation of the present application is realized. For example, any one of the method steps performed in the first aspect.

In yet another aspect of the implementation of the present application, a computer program product containing instructions is also provided, which, when run on a computer, causes the computer to execute any of the methods for calculating the degree of association between features and tags described above.

A method, device, and electronic device for calculating the degree of correlation between features and tags provided in the embodiments of the present application can detect the degree of correlation between target features and target tags for the trained model to be detected by applying the method of the embodiment of the present application. The detected correlation degree is relevant and accurate to the model to be detected, and can be applied to the model to be detected. By inputting the verification data set into the model to be detected, the target feature value in the verification data set is perturbed multiple times, and the value of the target label is predicted, which realizes the automatic analysis of the correlation between the target feature and the target label in the model to be detected, avoiding manual Do input calculations. In addition, when detecting the correlation between the target label and the target feature, it is only necessary to perturb the value of the target feature, and predict the value of the target label through the value of the target feature, and finally obtain a direct representation of the influence of the target feature on the target label. It is not necessary to predict the influence of all features on the target label, so as to compare the influence of the target features on the target label. Therefore, the application of the method of this application can greatly reduce the amount of calculation, thereby reducing the waste of computing power.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following briefly introduces the drawings that are required in the description of the embodiments or the prior art.

FIG. 1 is a flow chart of a method for calculating the degree of association between features and labels provided by the embodiment of the present application;

FIG. 2 is a schematic diagram of the distribution of target features in a verification data set provided by an embodiment of the present application;

FIG. 3 is a schematic flow chart of a structure of a device for calculating feature and label association provided by an embodiment of the present application;

FIG. 4 is a system block diagram for calculating the degree of correlation between features and tags provided by the embodiment of the present application;

FIG. 5 is a schematic structural diagram of an electronic device provided by an embodiment of the present application.

Detailed ways

The technical solutions in the embodiments of the present application will be described below with reference to the drawings in the embodiments of the present application.

In a related technology, the correlation between the feature and the label can be obtained by calculating the Pearson correlation coefficient before the feature and the label through statistical indicators. The method is mainly to represent the target feature as an array, and represent the target label as an array, and obtain the relationship between the target label and the target feature by calculating the relationship between the two arrays. However, through such a method, the relationship obtained is usually a primary relationship (linear relationship), a quadratic relationship (the relationship between the original value and the square), a cubic relationship (the relationship between the original value and the cubic power), an exponential relationship, and a logarithmic relationship. However, the real scene is complex, and this kind of numerical analysis is difficult to analyze comprehensively, and cannot be applied to the actual model.

In related technologies, there is another method of judging by the final effect after adding or subtracting models. The advantage of this method is that the result is actually verified, but if you want to analyze the correlation between all features and the target label, you usually need to train the data after removing a certain feature, so there are M features, you need to train M times; and usually M>50, so this will consume a lot of computing resources.

In order to solve the technical problems in the above method, in the first aspect of the implementation of the present application, a method for calculating the degree of association between features and tags is firstly provided. The above method includes the steps shown in Figure 1:

Step S101: Receive detection parameters from the client component.

Wherein, the detection parameters include parameters required to calculate the degree of association between the target feature and the target label. For example, the detection parameters may include: information about the model to be detected, information about a verification data set, names of target features and tags, and perturbation mode information. Wherein, the model to be detected is a model that has been trained to obtain the target label, and the information of the model to be detected is used to indicate the acquisition of the model to be detected, wherein the information of the model to be detected can be obtained in various forms, for example, when the user directly upload the model to be tested. At this time, the information of the model to be tested is the file address or file name corresponding to the model to be tested. It can also be the access address corresponding to the model to be tested by the user, such as HDFS/S3 (Hadoop Distributed System/Simple Storage Service , distributed system/simple storage server) address, at this time, the information of the model to be detected is the access address corresponding to the model to be detected.

The validation data set contains multiple pieces of validation data, and each piece of validation data includes the value of the target feature. Similarly, the verification data set information can also be obtained in various forms, for example, it can be obtained through the HDFS/S3 address corresponding to the verification data set or in other ways.

Since there are multiple features input into the model to be detected, the output may also be multiple results. The name of the target feature is usually used by the user to input to the client to indicate which feature is used as the target feature. The name of the target label is used for Input to the client to just have which result as the target label. The target feature is the feature determined by the user according to his own needs to detect the degree of correlation with the target tag. When the user enters the name of the target feature, he can enter the name of one target feature, or input multiple target feature names. The representation of the target features in the validation dataset is usually numerical. The value of the target label is the output result of the model to be tested.

The perturbation method information is a method determined by the user to perturb the value of the target feature in the verification data set, and the perturbation method information may include the method and times of perturbation.

In one example, in the detection parameters received from the client component, it is recognized that the detection parameters include the access address of a network model that has been trained to predict the loan amount, and each piece of verification data includes "lender age" and "loan The access address of the verification data set of multiple features of "person's consumption record", "lender's total asset value" and "lender's ID number". In this example, the name of the target feature is "lender's total asset value", and the target label's The name "loan amount", the perturbation method information is to randomly scramble the order of the value of the target feature in the verification data set 50 times, which means that in this example, the feature "total asset value of the lender" in the verification data set is to be modified value, to calculate the correlation degree between the target feature "total asset value of the lender" and the target label "loan limit" output by the network model in the network model for predicting the loan amount.

Step S102: Call the model to be detected according to the detection parameters.

Wherein, the model to be detected is used to predict the value of the target label according to the input target features; specifically, the model information to be detected in the detection parameters in step S101 can be identified, and the model to be detected can be acquired and invoked according to the information of the model to be detected. For example, when the information of the model to be detected is the access address of the model to be detected, the model to be detected is accessed and invoked through the obtained access address.

Step S103: Obtain a verification data set by detecting parameters.

Among them, the verification data set includes the value of the target feature, and the verification data set is used to input into the model to be tested; when obtaining the verification data set, the same method as step S102 can be used to obtain and call the verification data set according to the information of the verification data set. data set. For example, when the verification data set information is the access address of the verification data set, the verification data set is accessed and called through the obtained access address.

Step S104: Input the verification data set into the model to be tested to obtain the first prediction result.

Step S105: Modify the value of the target feature in the verification data set according to the detection parameters to obtain a modified verification data set.

In practical applications, the disturbance method information in the detection parameters received in step S101 can be identified, and the value of the target feature in the verification data set can be modified according to the disturbance method and the number of disturbances in the disturbance method information to obtain a modified verification data set .

Step S106: input the modified verification data set into the model to be tested for prediction, and obtain a second prediction result.

Step S107: Calculate the correlation degree between the target feature and the target label in the model to be detected according to the second prediction result and the first prediction result.

In practical applications, if the correlation between the target feature and the target label is large, that is, the target feature has a greater impact on the output of the detection model, then when the value of the target feature in the verification data set is modified, the target label output by the detection model The value change is more obvious.

For example, when the model to be tested is a model that predicts the loan amount, the input verification data set includes the two features of the total asset value of the lender and the lender's ID number, and the total asset value of the lender has a greater impact on the output loan amount. is large, when the "total asset value of the lender" is the target feature, the greater the change in the value of the modified "total asset value of the lender", the greater the output loan amount will be; The output loan amount has little influence, then when the "lender ID number" is the target feature, even if the value of the "lender ID number" changes more, the output loan amount will not change significantly. Therefore, the correlation degree between the target feature and the target label in the model to be detected can be calculated according to the first prediction result and the second prediction result.

Specifically, there are many ways to calculate the degree of correlation between target features and target labels in the model to be detected. For example, the difference between the first prediction result and the second prediction result can be calculated to obtain the target feature and target label in the model to be detected The correlation degree of the target feature and the target label in the model to be detected can also be obtained by calculating the ratio of the first prediction result and the second prediction result.

Apply the method of the embodiment of the present application to detect the degree of correlation between the target feature and the target label on the model to be detected that has been trained. The detected degree of correlation is related to the model to be detected and reliable, and can be applied to the model to be detected. in the model. The verification data set can be input into the model to be detected, and the target feature value in the verification data set can be disturbed to predict the value of the target label, which can automatically analyze the correlation between the target feature and the target label in the model to be detected, and avoid manual input calculate. In addition, when detecting the degree of correlation between the target label and the target feature, it is only necessary to perturb the value of the target feature to predict the value of the target label, and obtain a correlation degree that can directly represent the influence of the target feature on the target label without predicting The influence of all features on the target label is compared to compare the influence of the target features on the target label. Therefore, the application of the method of the present application can greatly reduce the amount of calculation, thereby reducing the waste of computing power.

In practical applications, due to the single use of the server for calculation, the amount of calculation will be large and the efficiency will be low. Therefore, the method of this application can be applied to the network side, wherein the network side includes server components and training resource components. The above-mentioned steps of calculating the degree of association in this application are executed in the component, managed by the server component, and the degree of association is fed back to the client.

Specifically, the training resource component may include multiple container platforms, each container platform has a corresponding identifier, and the training resource component is matched to the preset container platform corresponding to the preset container platform identifier through the preset container platform identifier. Users can determine one or more container platforms to implement the method of this application according to their own needs. Therefore, when users input detection parameters on the client terminal, the detection parameters can include the preset container platform identifier used to determine the preset container platform. The preset container platform identifier may be various types of identifiers. For example, the preset container platform identifier may be an access address corresponding to the preset container platform, or may be a name of the preset container platform. The method of the embodiment of the present application can also be realized through the following steps:

Step 1: The server component receives the detection parameters from the client component, identifies the preset container platform identifier in the detection parameters, and sends an invocation instruction to the training resource component.

Step 2: The training resource component invokes the container platform corresponding to the preset container platform identifier and the model to be tested corresponding to the model information to be tested according to the received call instruction, and executes steps S102-S107; and feeds back the degree of association to the server components.

Step 3: The server component receives the feedback result fed back by the training resource component, including the correlation degree between the target feature and the target label, and sends the feedback result to the client component.

Wherein, the feedback result includes the degree of association between the target feature and the target label, and the degree of association can be fed back in the form of a dictionary index of {feature: degree of association}. The degree of association indicates the influence of the target feature on the target label. A larger value indicates that the target feature has a greater influence on the target label, and conversely, a smaller correlation degree indicates that the target feature has a smaller influence on the target label.

By applying the method of the embodiment of the present application, the correlation degree between the target feature and the target label in the model to be detected can be calculated in the training resource component, and the correlation degree can be fed back to the client component through the server component to reduce the computing power of the server component , to improve the efficiency of calculating the correlation degree. And after obtaining the correlation degree of the target feature to the target label, when using the above-mentioned detection model to predict the target label, the features with a small correlation degree with the target label can be eliminated, thereby reducing the amount of calculation.

The disturbance mode information in the detection parameters of the client component received in the above step S101 includes the disturbance method and the number of disturbances; in the above step S105, the value of the target feature in the verification data set is modified according to the detection parameters, and the modified verification data set is obtained. Include the following steps:

Step 1. According to the perturbation method, modify the value of the target feature in each piece of verification data in the verification data set to obtain a modified verification data set;

Step 2: Repeat step 1 according to the number of perturbations to obtain a modified verification data set with the same number of perturbations.

Among them, the disturbance method refers to the method of modifying the value of the target feature. There are many disturbance methods, such as disturbance according to the normal distribution method, disturbance according to the uniform distribution method, disturbance according to the method of disrupting the value of the target feature value, and disturbance by directly assigning a constant value. , users can determine the perturbation method according to their own needs. The number of disturbances indicates the number of times to modify the value of the target feature, and it can also be a preset number of times according to business requirements. The more the number of disturbances, the more accurate the final correlation degree will be.

Applying the method of the embodiment of the present application, the value of the target label can be predicted by only perturbing the value of the target feature, and then by calculating the change of the value of the target label after multiple perturbations to the value of the target feature, to obtain the value to be detected The degree of correlation between the target feature and the target feature in the model reduces the amount of calculation on the network side and achieves the technical effect of reducing the waste of computing power.

In the following, it will be described by taking the perturbation method as perturbation according to the normal distribution as an example. At this time, in the above step 1, modify the value of the target feature in each piece of verification data in the verification data set according to the perturbation method, including the following steps:

Step 1, calculate the mean value and variance value when the value of the target feature is normally distributed in the verification data set, and obtain the normal distribution curve of the value of the target feature in the verification data set;

Step 2. Randomly change the value of the target feature in each piece of verification data along the normal distribution curve.

In this case, the value of the target feature is randomly perturbed according to the normal distribution, so as to automatically obtain different verification data sets according to the normal method. As shown in Figure 2, in practical applications, the value of the target feature in the verification data set may be randomly distributed as shown in the histogram in Figure 2, and its normal distribution is calculated as shown in the curve in Figure 2, then the values in the verification data set can be The values of the target feature are modified to fit this normal distribution curve. Applying the method of the embodiment of the present application can provide a method for disturbing the value of the target feature, and increase the diversity of methods for disturbing the value of the target feature.

In the following, description will be made by taking the perturbation method as perturbation in a uniform distribution manner as an example. At this time, the above step 1 modifies the value of the target feature in each piece of verification data in the verification data set according to the perturbation method, including the following steps:

Step 1, calculate the mean value of the winning feature value of each item of verification data in the verification data set;

Step 2, modify the value of the target feature in each piece of verification data to the mean value.

In this case, the value of the target feature is randomly perturbed in a uniform distribution manner, so as to automatically obtain different verification data sets in a normal manner. Applying the method of the embodiment of the present application can provide a method for disturbing the value of the target feature, and increase the diversity of methods for disturbing the value of the target feature.

In the following, the perturbation method is perturbed according to the method of disturbing the value of the target feature as an example for illustration. At this time, the above step 1 modifies the value of the target feature in each piece of verification data in the verification data set according to the perturbation method, including the following steps:

Keep the values of the target features in the validation dataset unchanged, and shuffle the order of the target features.

Applying the method of the embodiment of the present application can provide a method for disturbing the value of the target feature, and increase the diversity of methods for disturbing the value of the target feature.

In the following, the disturbance method is directly assigned a constant value as an example for illustration. At this time, the above step 1 modifies the value of the target feature in each piece of verification data in the verification data set according to the selected perturbation method, including the following steps:

Assigns the values of all target features in the validation dataset to a constant.

Applying the method of the embodiment of the present application can provide a method for disturbing the value of the target feature, and increase the diversity of methods for disturbing the value of the target feature.

In the above step S107, the degree of correlation between the target feature and the target label in the model to be detected is calculated according to the second prediction result and the first prediction result, including:

Calculate the average value of the difference between one or more second prediction results and the first prediction result to obtain the correlation degree between the target feature and the target label in the model to be detected.

If the correlation degree between the target feature and the target label is large, the predicted prediction result will change greatly after each disturbance to the value of the target feature. Since there are multiple pieces of verification data in the verification data set, the prediction results obtained from each piece of verification data can be averaged to obtain the first prediction result.

In one example, the verification data set is input into the model to be tested, and the first prediction result obtained is avg(R). After multiple perturbations are performed on the value of the target feature in the verification data set, the modified verification data sets are respectively input into In the model to be tested, a plurality of second prediction results avg(R ₁ '), avg(R ₂ '), avg(R ₃ ')...avg(R _N '), where N represents the number of disturbances, is to be The correlation between target features and target labels in the detection model can be calculated by the following formula:

D=((avg(R)-avg(R ₁ '))+(avg(R)-avg(R ₂ '))+(avg(R)-(R ₃ '))+…

+(avg(R)-vg(R _N ')))/

Among them, D represents the correlation degree between the target feature and the target label in the model to be detected; R represents the prediction result obtained for each verification data in the verification data set; R ₁ ^' represents each verification data in the verification data set after the first perturbation The obtained prediction results; R ₂ 'represents the prediction results obtained for each piece of verification data in the verification data set after the second perturbation; R ₃ ' represents the prediction results obtained for each piece of verification data in the verification data set after the third perturbation ; R _N ' represents the prediction result obtained for each piece of verification data in the verification data set after the Nth perturbation; N represents the number of perturbations.

In an example, when the container platform is preset to perform the task of calculating the correlation between target features and target labels, the following pseudocode can be used to run:

Obtain parameters model_dir, dataset_dir, feature_names, function, epochs//Receive the detection parameters sent by the client

Load model model=load_model(model_dir)//Load the model to be tested through the information of the model to be tested

Load dataset_dir and make a prediction R=model.predict(dataset)//Load the verification data set through the verification data set information; and input the verification data set into the model to be tested for prediction

For fn in feature_names

Analyze the data distribution of fn in dataset_dir//Calculate the value distribution of the target features in the verification data set

For i in range(epoches)

Modify the value of the fn feature in dataset_dir//disturb the value of the target feature

Load new_dataset_dir data as new_dataset_i//Update the modified verification data set and re-enter it into the model to be tested for prediction

R'.append(mean(model.predict(new_dataset_i)))Record the average predicted value in R'//Calculate the predicted value corresponding to each modified verification data set

fn_R'=Mean(R') to get the average predicted value of feature fn after multiple executions//calculate the average value of the predicted value changes after multiple modifications

The degree of association between fn and Label is fn_R’-R//Get the degree of association between the target label and the target feature

Return all {fn:fn_R’-R}//Return the intermediate results to the server component

Applying the method of the embodiment of the present application can provide a method for quantifying the relationship between the target feature and the target label in the detection model, and obtain a correlation degree that can directly represent the influence of the target feature on the target label without predicting all features The impact on the target label, so as to compare the influence of the target feature on the target label. Therefore, the application of the method of this application can reduce the amount of calculation, thereby reducing the waste of computing power.

In the above-mentioned embodiment, if the container platform in the training resource component performs the step of calculating the correlation degree between the target feature and the target label in the model to be detected, the pre-stored container platforms in the training resource component include self-built container platforms or public container platforms. Among them, the self-built container platform only supports designated user access, and the public container platform can support all user access.

Specifically, the specified user may be an authenticated user, and the authentication may determine whether the user is the specified user through the user's IP and network information. For example, a user connected to a local area network can be authenticated as a designated user, and the user can only access the self-built container platform when connected to the local area network. The self-built container platform is a self-built resource pool or computer room. The self-built container platform only accepts access from designated users and has better security. For example, the self-built container platform can be a k8s cluster or a mesos cluster. Public container platforms can support access by multiple users, such as AWS (Amazon web service, Amazon Cloud), Tencent Cloud, Huawei Cloud, and Alibaba Cloud. When using a public container platform, it is necessary to have an interface that adapts to the public container platform.

When performing the task of calculating the correlation degree in the preset container platform, it is necessary to perturb the value of the target feature in the verification data set multiple times, and then perform multiple predictions on the modified verification data set, which is computationally intensive and time-consuming. Therefore, you can choose to perform computing tasks on the GPU (Graphic Processing Unit, graphics processing unit).

Applying the method of the embodiment of the present application can reduce the computational load of the CPU (Central Processing Unit, central processing unit) by presetting the container platform in the GPU to carry out the calculation of the target feature and the target label correlation degree on the detection model, thereby improving the calculation efficiency of association.

In this embodiment of the present application, the second prediction result of each prediction may also be fed back to the client component as an intermediate result.

Specifically, the intermediate result may include the value after each perturbation of the value of the target feature in the verification data set, and the second prediction result corresponding to the modified verification data set predicted by the model to be detected after each perturbation. After the training resource component feeds back the intermediate results to the server component, the server component can directly feed it back to the client.

By applying the method of the embodiment of the present application, the intermediate results in the execution of the prediction task can also be fed back to the server component, thereby facilitating the user to inquire about the process of task execution.

In practical application, the server component also includes a database, which can store the calculated correlation degree between the target feature and the target label in the model to be detected, and/or the intermediate results in the above embodiments. The method of the embodiment of the present application can also be realized through the following steps:

Step 1, storing the degree of association and/or intermediate results in the database.

Step 2, receiving query information from the client component, the query information includes the target feature to be queried and the name of the target tag.

Step 3: Find the correlation degree corresponding to the target feature and the target label name from the database, and feed back the query result including the correlation degree to the client component.

Specifically, after receiving the query information from the client component, the target features and target tag names in the query information can be matched with the target features and target tag names in the database to find the target features and target tag names in the query information Corresponding degree of association, and feed back the query result including the degree of association to the client component.

Applying the method of the embodiment of the present application can enable the user to query the correlation degree between the target feature and the target label and the operation result, so that the user can watch the intermediate prediction result, and can also perform manual checking or calculation when the system fails. detection, which is more user-friendly.

Specifically, the query result may be a feedback result including the correlation degree of the target feature and the target label in the model to be detected, an intermediate result fed back by the training resource component, and a result of commenting on the correlation degree. Among them, the result of commenting on the degree of association can be determined according to the size of the degree of association. When the degree of association is large, the evaluation result is that the target feature has a greater impact on the target label. If the degree of association is small, the evaluation result Because the target feature has less impact on the target label, it is recommended to remove it.

By applying the method of the embodiment of the present application, the received results can be stored in the database for the client to query, so that the user can see the task running results more intuitively.

The second aspect of the embodiment of the present application also provides a device for calculating the degree of association between features and tags, as shown in Figure 3:

The detection parameter receiving module 301 is used to receive detection parameters from the client component; wherein the detection parameters include parameters required to calculate the degree of association between the target feature and the target tag;

The model to be detected calling module 302 is used to call the model to be detected according to the detection parameters; wherein, the model to be detected is used to predict the value of the target label according to the input target feature;

The verification data set acquisition module 303 is used to obtain the verification data set through the detection parameters; wherein, the verification data set includes the value of the target feature, and the verification data set is used for inputting into the model to be detected;

The first prediction result obtaining module 304 is used to input the verification data set into the model to be tested to obtain the first prediction result;

The verification data set modification module 305 is used to modify the value of the target feature in the verification data set according to the detection parameters to obtain a modified verification data set;

The second prediction result obtaining module 306 is used to input the modified verification data set into the model to be tested for prediction to obtain the second prediction result;

The correlation calculation module 307 is configured to calculate the correlation between the target feature and the target label in the model to be detected according to the second prediction result and the first prediction result.

Apply the device of the embodiment of the present application to detect the degree of correlation between the target feature and the target label on the trained model to be detected. The detected degree of correlation is related to the model to be detected and reliable, and can be applied to the model to be detected. in the model. The verification data set can be input into the model to be detected, and the target feature value in the verification data set can be disturbed to predict the value of the target label, which can automatically analyze the correlation between the target feature and the target label in the model to be detected, and avoid manual input calculate. In addition, when detecting the degree of correlation between the target label and the target feature, it is only necessary to perturb the value of the target feature to predict the value of the target label, and obtain a correlation degree that can directly represent the influence of the target feature on the target label without predicting The influence of all features on the target label is compared to compare the influence of the target features on the target label. Therefore, the application of the method of the present application can greatly reduce the amount of calculation, thereby reducing the waste of computing power.

In a possible implementation, the detection parameters include a disturbance method and a disturbance number; wherein, the disturbance method indicates a method for modifying the value of the target feature, and the disturbance number indicates the number of times the value of the target feature is modified, and the verification data set modification module includes:

Modify the verification data set sub-module, which is specifically used to modify the value of the target feature in each piece of verification data in the verification data set according to the perturbation method, and obtain the modified verification data set;

The repeated modification sub-module is specifically used to repeat the step of modifying the value of the target feature in each piece of verification data in the verification data set according to the perturbation times and according to the perturbation method, so as to obtain a modified verification data set with the same number of perturbations.

Applying the device of the embodiment of the present application, it is possible to predict the value of the target label by only perturbing the value of the target feature, and then by calculating the change of the value of the target label after multiple perturbations to the value of the target feature, to obtain the value to be detected The degree of correlation between the target feature and the target feature in the model reduces the amount of calculation on the network side and achieves the technical effect of reducing the waste of computing power.

In a possible implementation, the perturbation method is perturbation according to a normal distribution; modifying the verification data set submodule includes:

The normal distribution curve calculation unit is specifically used to calculate the mean value and variance value when the value of the target feature is normally distributed in the verification data set, and obtain the normal distribution curve of the value of the target feature in the verification data set ;

The normal distribution modification unit is specifically used to randomly change the value of the target feature in each piece of verification data along the normal distribution curve.

Applying the device of the embodiment of the present application can provide a method for disturbing the value of the target feature, and increase the diversity of methods for disturbing the value of the target feature.

In a possible implementation manner, the correlation calculation module includes:

The average value calculation sub-module is specifically used to calculate the average value of the difference between the second prediction result and the first prediction result to obtain the correlation degree between the target feature and the target label in the model to be tested.

Applying the device of the embodiment of the present application can provide a method for quantifying the relationship between the target feature and the target label in the model to be detected, and obtain a correlation degree that can directly represent the influence of the target feature on the target label without predicting all features The impact on the target label, so as to compare the influence of the target feature on the target label. Therefore, the application of the method of this application can reduce the amount of calculation, thereby reducing the waste of computing power.

In a possible implementation manner, the device also includes:

The intermediate result feedback module is configured to feed back the second prediction result of each prediction to the client component as an intermediate result.

By applying the device of the embodiment of the present application, the intermediate results in the execution of the prediction task can also be fed back to the server component, so as to facilitate the user to inquire about the process of task operation.

In a possible implementation manner, the device also includes:

The feedback result storage module is used to store the degree of association and/or intermediate results in the database;

The query information receiving module is used to receive the query information from the client component, and the query information includes the target feature to be queried and the target tag name;

The query result feedback module is used to find the correlation degree corresponding to the target feature and the target tag name from the database, and feed back the query result including the correlation degree to the client component.

Applying the device of the embodiment of the present application can enable the user to inquire about the correlation degree between the target feature and the target label and the operation result, so that the user can watch the intermediate prediction result, and can also perform manual checking or calculation when the system fails. detection, which is more user-friendly.

In practical applications, if the method for calculating the degree of association in the first aspect of the embodiment of the present application is applied to the training resource component, the third aspect of the application can also provide a system for calculating the degree of association between features and tags, specifically , the system includes as shown in Figure 4:

Client component 401 (Client end, client end), is used for obtaining the detection parameter of user input, and detection parameter is sent to server component, receives and displays the feedback result that server component sends and comprises target feature and target label association degree, wherein , the detection parameters include the information of the model to be detected, the information of the verification data set, the identification of the preset container platform, the name of the target feature and target label, and the information of the disturbance method;

The server component 402 (Server end, server end) is used to receive the detection parameters sent by the client, according to the preset container platform identifier, send a call instruction to the training resource component, and receive the feedback from the training resource component, including the degree of correlation between the target feature and the target label The feedback result, and send the feedback result to the client component;

The training resource component 403 is used to call the container platform corresponding to the preset container platform identifier and the model to be tested corresponding to the model information to be tested according to the received calling instruction, obtain the verification data set through the verification data set information, and convert the verification data Input the data set into the model to be tested to obtain the first prediction result; modify the value of the target feature in the verification data set according to the disturbance mode information to obtain the modified verification data set; input the modified verification data set into the model to be tested for prediction, and obtain , the second prediction result; calculate the correlation degree between the target feature and the target label in the model to be detected according to the second prediction result and the first prediction result; and feed back the correlation degree to the server component.

Apply the system of the embodiment of the present application to detect the correlation degree between the target feature and the target label on the trained model to be detected. The detected correlation degree is related to the model to be detected and is reliable, and can be applied to the model to be detected. in the model. By inputting the detection parameters to the network side, the network side inputs the verification data set into the model to be detected through the preset container platform, performs multiple perturbations on the target feature values in the verification data set, predicts the value of the target label, and realizes automatic analysis The degree of correlation between target features and target labels in the model to be detected avoids manual input calculations. In addition, when detecting the degree of correlation between the target label and the target feature, it is only necessary to perturb the value of the target feature to predict the value of the target label, and obtain a correlation degree that can directly represent the influence of the target feature on the target label without predicting The influence of all features on the target label is compared to compare the influence of the target features on the target label. Therefore, the application of the method of the present application can greatly reduce the amount of calculation, thereby reducing the waste of computing power.

As shown in FIG. 4 above, the client component 401 includes the following modules:

The command line tool module (Cli, command line tool) is used to obtain the detection parameters input by the user through the command line tool, and send the detection parameters to the server component.

The URL module (Web, URL) is used to receive and display the feedback result sent by the server component, including the correlation degree between the target feature and the target label.

Wherein, the URL module can also provide a URL of the system, and the user can click the URL to enter the client interface of the system, and can input the detection parameters in the interface, and the interface can also display the information sent by the server component, including target features and target parameters. Feedback result of tag association degree. The command-line tool module provides a command-line tool for the user, and the user can input detection parameters in the form of the command-line tool.

By applying the system provided by the embodiment of the present application, the user can input detection parameters in multiple ways through the website module and the command line tool module, which increases the diversity of user input modes.

As shown in FIG. 4 , the client component 401 also includes an interface call input module (Restful API, REST-style API), which is used to establish a connection with a third-party system and realize integration with the third-party system.

Specifically, the interface call input module can provide advanced users with an interface that can access the third-party system, and the advanced user can integrate the third-party system with the system by inputting specific parameters of the interface, so that the detection of the third-party system can be detected through the interface. Parameters are input to the client. Among them, advanced users are users with third-party systems.

The system applying the embodiment of the present application can integrate with the third-party system by providing an API interface, so that the user can directly interact with the system of the embodiment of the present application through the third-party system, increasing the diversity of user use.

In a possible implementation manner, the command line tool module is also used to receive query information sent by the user, where the query information includes the target feature to be queried and the target tag name;

The URL module is also used to receive and display the query result including the degree of relevance sent by the server component.

Among them, the URL module actually provides a visual interface through which not only the final feedback results and intermediate feedback results of the correlation between target features and target tags can be seen, but also the detection parameters input by the user can be seen through this interface, and Analyze the running status of tasks and use resource status information.

Wherein, the running state of the analysis task includes three states: running, running successfully, and running failed. When the client component displays the running status, it can also display the intermediate results of calculating the correlation between the target feature and the target label in the model to be detected according to the user's instruction. The resource status information includes the calling status of the preset container platform and the memory operation status of the GPU. Wherein, the calling situation of the preset container platform may include the number of called preset container platforms and the operating environment of the preset container platform.

By applying the method of the embodiment of this application, the calculation of the correlation degree between the target features and the target tags in the model to be detected can be displayed on the client, so that the user can directly view various information, so that the user feels more intuitive and more convenient to use .

As shown in FIG. 4 above, the server component 402 includes the following modules:

The interface service module (API server) is specifically used to receive the detection parameters sent by the client component, and send an execution task request including the detection parameters to the manager module;

The manager module (Manager) is specifically used to receive the execution task request sent by the interface service module, and send the call instruction to the cloud interface call module according to the execution task request; receive the feedback from the cloud interface call module, including the target feature and the target tag correlation degree Feedback the result, and send the feedback result to the client component through the interface service module;

The cloud interface calling module (Cloud Adaptor) is specifically used to receive the calling instruction sent by the manager module; call the training resource component according to the calling instruction, and receive the feedback result returned by the training resource component.

Applying the system of the embodiment of the present application, the user only needs to input the detection parameters to the server component, and the server component calls the training resource component to calculate the correlation degree between the target feature and the target label in the model to be detected, avoiding manual prediction of multiple prediction results The calculation of the correlation degree realizes the technical effect of automatically analyzing the correlation degree between the target feature and the target label in the model to be detected.

As shown in FIG. 4 above, the server component 402 also includes a data saving module, that is, a database (DB, Datebase), which is used to receive and save the feedback result sent by the manager module.

Specifically, the manager module can write the received results into the database. The feedback result sent by the manager module may include the correlation degree between the target feature and the target label in the model to be detected, the intermediate result fed back by the training resource component, and the result of commenting on the correlation degree.

Applying the system of the embodiment of the present application, the received results can be stored in the database for the client to query, so that the user can see the task running results more intuitively.

In the above-mentioned server component 402, the interface service module is also used to receive query information from the client component, the query information includes the target feature and target tag name to be queried; and send a task execution request containing the query information to the manager module.

The manager module is used to find the correlation degree corresponding to the target feature and the target label name from the database according to the task execution request, and feed back the query result including the correlation degree to the client component through the interface service module.

Applying the system of the embodiment of the present application can enable the user to inquire about the correlation degree between the target feature and the target label and the operation result, which is more user-friendly.

The embodiment of the present application also provides an electronic device, as shown in FIG. complete the mutual communication,

Memory 503, used to store computer programs;

When the processor 501 is used to execute the program stored on the memory 503, the following steps are implemented:

Receive detection parameters from the client component; where the detection parameters include parameters required to calculate the degree of association between the target feature and the target label;

According to the detection parameters, the model to be detected is called; wherein, the model to be detected is used to predict the value of the target label according to the input target characteristics;

Obtain a verification data set by detecting parameters; wherein, the verification data set includes the value of the target feature, and the verification data set is used to input into the model to be detected;

Input the verification data set into the model to be tested to obtain the first prediction result;

Modify the value of the target feature in the verification data set according to the detection parameters to obtain the modified verification data set;

Inputting the modified verification data set into the model to be tested for prediction to obtain a second prediction result;

The degree of correlation between the target feature and the target label in the model to be detected is calculated according to the second prediction result and the first prediction result.

The communication bus mentioned by the terminal above may be a Peripheral Component Interconnect (PCI for short) bus or an Extended Industry Standard Architecture (EISA for short) bus or the like. The communication bus can be divided into an address bus, a data bus, a control bus, and the like. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

The communication interface is used for communication between the terminal and other devices.

The memory may include a random access memory (Random Access Memory, RAM for short), and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory. Optionally, the memory may also be at least one storage device located far away from the aforementioned processor.

The above-mentioned processor can be a general-purpose processor, including a central processing unit (Central Processing Unit, referred to as CPU), a network processor (Network Processor, referred to as NP), etc.; it can also be a digital signal processor (Digital Signal Processor, referred to as DSP) , Application Specific Integrated Circuit (ASIC for short), Field Programmable Gate Array (Field-Programmable Gate Array, FPGA for short) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

In yet another embodiment provided by the present application, a computer-readable storage medium is also provided. A computer program is stored in the computer-readable storage medium. When the computer program is executed by a processor, the server component executes any one of the above-mentioned embodiments. A method for calculating the degree of association between features and labels.

In yet another embodiment provided by the present application, a computer-readable storage medium is also provided. A computer program is stored in the computer-readable storage medium. When the computer program is executed by a processor, the training resource component implements any of the above-mentioned embodiments. A method for calculating the degree of association between features and labels.

In yet another embodiment provided by the present application, a computer program product including instructions is also provided, which, when run on a computer, causes the computer to execute the method for calculating the degree of association between a feature and a tag in any of the above embodiments.

In the above embodiments, all or part of them may be implemented by software, hardware, firmware or any combination thereof. When implemented using software, it may be implemented in whole or in part in the form of a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on the computer, the processes or functions according to the embodiments of the present application will be generated in whole or in part. A computer can be a general purpose computer, special purpose computer, computer network, or other programmable device. Computer instructions may be stored in or transmitted from one computer-readable storage medium to another computer-readable storage medium, e.g. Coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.) to another website site, computer, server or data center. The computer-readable storage medium may be any available medium that can be accessed by a computer, or a data storage device such as a server, a data center, etc. integrated with one or more available media. Available media may be magnetic media (eg, floppy disk, hard disk, magnetic tape), optical media (eg, DVD), or semiconductor media (eg, Solid State Disk (SSD)).

It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is a relationship between these entities or operations. There is no such actual relationship or order between them. Furthermore, the term "comprises", "comprises" or any other variation thereof is intended to cover a non-exclusive inclusion such that a process, method, article, or apparatus comprising a set of elements includes not only those elements, but also includes elements not expressly listed. other elements of or also include elements inherent in such a process, method, article, or device. Without further limitations, an element defined by the phrase "comprising a ..." does not preclude the presence of additional identical elements in the process, method, article, or apparatus that includes the element.

Each embodiment in this specification is described in a related manner, the same and similar parts of each embodiment can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device and system embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and for relevant parts, please refer to part of the description of the method embodiments.

The above are only preferred embodiments of the present application, and are not intended to limit the protection scope of the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of this application are included within the protection scope of this application.

Claims (10)

1. A method for calculating relevance of features to tags, the method comprising:

receiving detection parameters from a client component; the detection parameters comprise parameters required for calculating the association degree between the target features and the target labels;

calling a model to be detected according to the detection parameters; the model to be detected is used for predicting the value of the target label according to the input target characteristics;

acquiring a verification data set through the detection parameters; the verification data set comprises values of the target features, and is used for being input into the model to be detected;

inputting the verification data set into the model to be detected to obtain a first prediction result;

modifying the value of the target feature in the verification data set according to the detection parameter to obtain a modified verification data set;

inputting the modified verification data set into the model to be detected for prediction to obtain a second prediction result;

and calculating the association degree of the target feature and the target label in the model to be detected according to the second prediction result and the first prediction result.

2. The method of claim 1, wherein the detection parameters include a perturbation method and a number of perturbations; wherein the disturbance method represents a method for modifying the target feature value, the disturbance times represent times for modifying the target feature value, and the modifying the value of the target feature in the verification data set according to the detection parameter to obtain a modified verification data set includes:

modifying the value of the target feature in each piece of verification data in the verification data set according to the disturbance method to obtain a modified verification data set;

and according to the disturbance times, repeating the step of modifying the value of the target feature in each piece of verification data in the verification data set according to the disturbance method to obtain the modified verification data sets with the same number as the disturbance times.

3. The method of claim 2, wherein the perturbation method is perturbation in a normal distribution manner; modifying the value of the target feature in each piece of verification data in the verification data set according to the perturbation method, including:

calculating the mean value and the variance value of the target characteristic in the verification data set in the normal distribution mode to obtain a normal distribution curve of the value of the target characteristic in the verification data set;

and randomly changing the value of the target characteristic in each piece of verification data along the normal distribution curve.

4. The method according to claim 1, wherein the calculating the association degree between the target feature and the target label in the model to be detected according to the second prediction result and the first prediction result comprises:

and calculating the average value of the difference values of the second prediction result and the first prediction result to obtain the association degree of the target characteristics and the target label in the model to be detected.

5. The method of claim 1, further comprising:

feeding back a second prediction result of each prediction as an intermediate result to the client component.

6. The method of claim 5, further comprising:

storing the degree of association, and/or the intermediate result, in a database;

receiving query information from the client component, wherein the query information comprises target characteristics to be queried and a target label name;

and searching the association degree corresponding to the target characteristic and the target label name from the database, and feeding back a query result containing the association degree to the client component.

7. An apparatus for calculating a degree of association between a feature and a tag, the apparatus comprising:

the detection parameter receiving module is used for receiving detection parameters from the client component; the detection parameters comprise parameters required for calculating the association degree between the target features and the target labels;

the model to be detected calling module is used for calling the model to be detected according to the detection parameters; the model to be detected is used for predicting the value of the target label according to the input target characteristics;

the verification data set acquisition module is used for acquiring a verification data set through the detection parameters; the verification data set comprises values of the target features, and is used for being input into the model to be detected;

the first prediction result obtaining module is used for inputting the verification data set into the model to be detected to obtain a first prediction result;

the verification data set modification module is used for modifying the value of the target feature in the verification data set according to the detection parameter to obtain a modified verification data set;

the second prediction result obtaining module is used for inputting the modified verification data set into the model to be detected for prediction to obtain a second prediction result;

and the association degree calculation module is used for calculating the association degree between the target feature and the target label in the model to be detected according to the second prediction result and the first prediction result.

8. The apparatus of claim 7, wherein the detection parameters include a perturbation method and a number of perturbations; wherein the perturbation method represents a method of modifying the target feature value, the perturbation times represent times of modifying the target feature value, and the verification data set modification module includes:

a modified verification data set sub-module, which is specifically configured to modify a value of a target feature in each piece of verification data in the verification data set according to the perturbation method, so as to obtain a modified verification data set;

and the repeated modification submodule is specifically used for repeating the step of modifying the value of the target feature in each piece of verification data in the verification data set according to the disturbance method according to the disturbance times to obtain the modified verification data sets with the same number as the disturbance times.

9. An electronic device is characterized by comprising a processor, a communication interface, a memory and a communication bus, wherein the processor and the communication interface are used for realizing mutual communication by the memory through the communication bus;

a memory for storing a computer program;

a processor for implementing the steps performed by the server component of the method of any one of claims 1 to 6 when executing the program stored in the memory.

10. A computer-readable storage medium, characterized in that a computer program is stored in the computer-readable storage medium, which computer program, when being executed by a processor, carries out the method steps of any one of the claims 1-6.

Images