CN113139447B - Feature analysis method, device, computer equipment and storage medium - Google Patents
Feature analysis method, device, computer equipment and storage medium Download PDFInfo
- Publication number
- CN113139447B CN113139447B CN202110394351.5A CN202110394351A CN113139447B CN 113139447 B CN113139447 B CN 113139447B CN 202110394351 A CN202110394351 A CN 202110394351A CN 113139447 B CN113139447 B CN 113139447B
- Authority
- CN
- China
- Prior art keywords
- feature
- preset data
- input
- sensitivity
- preset
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000004458 analytical method Methods 0.000 title claims abstract description 26
- 230000035945 sensitivity Effects 0.000 claims abstract description 119
- 238000010801 machine learning Methods 0.000 claims abstract description 69
- 238000000034 method Methods 0.000 claims abstract description 30
- 230000036961 partial effect Effects 0.000 claims description 32
- 238000004590 computer program Methods 0.000 claims description 20
- 238000011156 evaluation Methods 0.000 claims description 12
- 239000008280 blood Substances 0.000 description 6
- 210000004369 blood Anatomy 0.000 description 6
- 238000010586 diagram Methods 0.000 description 6
- 238000004364 calculation method Methods 0.000 description 5
- 238000012163 sequencing technique Methods 0.000 description 5
- 238000013528 artificial neural network Methods 0.000 description 3
- 238000013527 convolutional neural network Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 230000002829 reductive effect Effects 0.000 description 3
- 230000000007 visual effect Effects 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 2
- 238000002372 labelling Methods 0.000 description 2
- 230000000670 limiting effect Effects 0.000 description 2
- 230000000873 masking effect Effects 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 238000012216 screening Methods 0.000 description 2
- 230000003068 static effect Effects 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000007794 visualization technique Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/46—Extracting features or characteristics from the video content, e.g. video fingerprints, representative shots or key frames
-
- A—HUMAN NECESSITIES
- A63—SPORTS; GAMES; AMUSEMENTS
- A63F—CARD, BOARD, OR ROULETTE GAMES; INDOOR GAMES USING SMALL MOVING PLAYING BODIES; VIDEO GAMES; GAMES NOT OTHERWISE PROVIDED FOR
- A63F13/00—Video games, i.e. games using an electronically generated display having two or more dimensions
- A63F13/50—Controlling the output signals based on the game progress
- A63F13/53—Controlling the output signals based on the game progress involving additional visual information provided to the game scene, e.g. by overlay to simulate a head-up display [HUD] or displaying a laser sight in a shooting game
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Software Systems (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Optics & Photonics (AREA)
- Medical Informatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Image Analysis (AREA)
Abstract
The application relates to a feature analysis method, a feature analysis device, computer equipment and a storage medium. The method comprises the following steps: acquiring a preset data set; inputting the input characteristics of each preset datum in the preset data set into a target machine learning model to obtain the predicted output of each preset datum; determining the feature sensitivity of the input feature of each preset data based on the predicted output of each preset data; and determining the average sensitivity of each feature to be analyzed in the feature set to be analyzed corresponding to the target machine learning model based on each feature sensitivity, and sorting each feature to be analyzed based on the average sensitivity of each feature to be analyzed to obtain a sorting result. By adopting the method, the feature ordering accuracy can be improved.
Description
Technical Field
The present disclosure relates to the field of computer technologies, and in particular, to a feature analysis method, a feature analysis device, a computer device, and a storage medium.
Background
With the rapid development of computer technology, artificial Intelligence (AI) based on deep learning has come into the field of view of the public. As a new direction in the field of machine learning, deep learning has achieved effects in search technology, data mining, image recognition, natural language processing, and the like, far exceeding those of the prior related art. When machine learning is used for real-world tasks, the features describing the sample typically need to be designed by human experts, which is called "feature engineering". Feature engineering occupies a considerable place in machine learning.
In the related art, the network weight of the initial layer of the machine learning model is generally used as a coefficient of feature importance, and then the importance of the features of the input model is ordered according to the weight, and feature screening is performed. Since the multi-layer neural network itself has the function of converting the underlying features into high-dimensional features and performing feature selection, the weight of the first layer network can be regarded as a coefficient of the model itself for screening the underlying features. However, since the machine learning model has integrity, only selecting the weight of the first layer network cannot completely reflect the contribution degree of the feature to the overall prediction result of the model, and therefore the result obtained by sequencing is not accurate.
Disclosure of Invention
In view of the foregoing, it is desirable to provide a feature analysis method, apparatus, computer device, and storage medium that can improve feature ordering accuracy.
A method of feature analysis, the method comprising:
acquiring a preset data set;
inputting the input characteristics of each preset datum in the preset data set into a target machine learning model to obtain the predicted output of each preset datum;
determining the feature sensitivity of the input feature of each preset data based on the predicted output of each preset data;
and determining the average sensitivity of each feature to be analyzed in the feature set to be analyzed corresponding to the target machine learning model based on each feature sensitivity, and sorting each feature to be analyzed based on the average sensitivity of each feature to be analyzed to obtain a sorting result.
In some embodiments, the preset data set is a set of game video frames; the input features of the preset data comprise a first input feature and a second input feature; the determining the feature sensitivity of the input feature of each preset data based on the prediction output of each preset data comprises the following steps:
acquiring target field information;
analyzing each game video frame based on the target field information to obtain a first input characteristic of each game video frame;
and calculating based on the target field information and the first input characteristics of each game video frame to obtain the second input characteristics of each game video frame.
In some embodiments, the determining the feature sensitivity of the respective input feature of the respective preset data based on the predicted output of the respective preset data comprises:
for each input feature, calculating a partial derivative for each dimension of the input feature based on a predicted output of preset data corresponding to the input feature;
and calculating norms of the input features based on the partial derivatives corresponding to the input features, and taking the norms as feature sensitivity of the input features.
In some embodiments, each preset data in the preset data set has a corresponding desired output; the calculating partial derivatives of the predicted output based on the preset data corresponding to the input features for each dimension of the input features comprises:
determining a corresponding prediction loss based on a predicted output and an expected output of preset data corresponding to the input feature;
first order partial derivatives of the predictive loss for each dimension of the input feature are calculated.
In some embodiments, the method further comprises:
determining target input characteristics according to the sorting result;
and acquiring preset standard features, comparing the target input features with the preset standard features, and determining an evaluation result of the target machine learning model according to the comparison result.
In some embodiments, the preset data in the preset data set is an image; the input characteristics of the preset data are all pixels in the image; the method further comprises the steps of:
and the calculated characteristic sensitivity of each pixel is sent to a terminal, and the terminal is used for marking each image according to the characteristic sensitivity of each pixel corresponding to each image and visually displaying the marking result.
In some embodiments, the preset data in the preset data set is an image; the input characteristics of the preset data are all pixels in the image; the method further comprises the steps of:
acquiring an image to be processed;
performing pixel shielding on the image to be processed based on the sorting result to obtain a target characteristic image;
and inputting the target characteristic image into the target machine learning model to obtain a predicted output corresponding to the image to be processed.
A feature analysis apparatus, the apparatus comprising:
the data acquisition module is used for acquiring a preset data set;
the data input module is used for inputting the input characteristics of each preset data in the preset data set into a target machine learning model to obtain the prediction output of each preset data;
the sensitivity determining module is used for determining the characteristic sensitivity of the input characteristics of each preset data based on the prediction output of each preset data;
the sorting module is used for determining the average sensitivity of each feature to be analyzed in the feature set to be analyzed corresponding to the target machine learning model based on each feature sensitivity, and sorting each feature to be analyzed based on the average sensitivity of each feature to be analyzed to obtain a sorting result.
A computer device comprising a memory storing a computer program and a processor which when executing the computer program performs the steps of:
acquiring a preset data set;
inputting the input characteristics of each preset datum in the preset data set into a target machine learning model to obtain the predicted output of each preset datum;
determining the feature sensitivity of the input feature of each preset data based on the predicted output of each preset data;
and determining the average sensitivity of each feature to be analyzed in the feature set to be analyzed corresponding to the target machine learning model based on each feature sensitivity, and sorting each feature to be analyzed based on the average sensitivity of each feature to be analyzed to obtain a sorting result.
A computer readable storage medium having stored thereon a computer program which when executed by a processor performs the steps of:
acquiring a preset data set;
inputting the input characteristics of each preset datum in the preset data set into a target machine learning model to obtain the predicted output of each preset datum;
determining the feature sensitivity of the input feature of each preset data based on the predicted output of each preset data;
and determining the average sensitivity of each feature to be analyzed in the feature set to be analyzed corresponding to the target machine learning model based on each feature sensitivity, and sorting each feature to be analyzed based on the average sensitivity of each feature to be analyzed to obtain a sorting result.
According to the method, the device, the computer equipment and the storage medium, the preset data set is obtained, the input features of all preset data in the preset data set are input into the target machine learning model, the prediction output of all preset data is obtained, the feature sensitivity of the input features of all preset data is determined based on the prediction output of all preset data, the average sensitivity of all the features to be analyzed in the feature set to be analyzed corresponding to the target machine learning model is determined based on all the feature sensitivities, all the features to be analyzed are ranked based on the average sensitivity of all the features to be analyzed, and a ranking result is obtained. .
Drawings
FIG. 1 is a diagram of an application environment for a feature analysis method in one embodiment;
FIG. 2 is a flow chart of a feature analysis method in one embodiment;
FIG. 3 is a flow chart illustrating the step of determining feature sensitivity in one embodiment;
FIG. 4 is a schematic diagram of a visual presentation of labeling results according to one embodiment;
FIG. 5 is a block diagram of a feature analysis device in one embodiment;
fig. 6 is an internal structural diagram of a computer device in one embodiment.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the present application will be further described in detail with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the present application.
The feature analysis method provided by the application can be applied to an application environment shown in fig. 1. Wherein the terminal 102 communicates with the server 104 via a network. The method comprises the steps that a server obtains a preset data set, input features of all preset data in the preset data set are input into a target machine learning model to obtain prediction output of all preset data, feature sensitivity of the input features of all preset data is determined based on the prediction output of all preset data, average sensitivity of all the features to be analyzed in a feature set to be analyzed corresponding to the target machine learning model is determined based on the feature sensitivity, all the features to be analyzed are ranked based on the average sensitivity of all the features to be analyzed to obtain a ranking result, the server can further send the ranking result to a terminal, and the terminal can conduct visual display based on the ranking result.
The terminal 102 may be, but not limited to, various personal computers, notebook computers, smartphones, tablet computers, and portable wearable devices, and the server 104 may be implemented by a stand-alone server or a server cluster composed of a plurality of servers.
In one embodiment, as shown in fig. 2, a feature analysis method is provided, and the method is applied to the server in fig. 1 for illustration, and includes the following steps:
step 202, acquiring a preset data set.
Wherein, the preset data set refers to a set composed of preset data. The preset data refers to data prepared in advance and available for performing feature analysis. The feature analysis herein refers to analysis of input features of the target machine learning model. The preset data may be various different types of data based on the difference of the target machine learning model, for example, may be image data, voice data, text data, and the like. In some embodiments, the preset data may be training sample data, test data, or the like for training the target machine learning model.
Specifically, the server may acquire preset data from a database established in advance, or may acquire preset data from a third party. The third party refers to a terminal and a computer device outside a server.
Step 204, inputting the input features of each preset datum in the preset data set into the target machine learning model to obtain the predicted output of each preset datum.
Wherein, the input characteristics of the preset data can be one or more. The input feature of the preset data may be sub-data extracted from the preset data, for example, when the preset data is an image, the pixel position of the image is an input feature, and generally speaking, one feature includes three color channels: r (Red), G (Green), B (Blue ); the input features of the preset data may also be features calculated according to the preset data, for example, the preset data is a game video frame, the server may acquire target field information, analyze the game video frame based on the target field information to obtain a first input feature, and calculate based on the target field information and the first sub-feature to obtain the first input feature, where the first input feature and the second input feature are both input features of the game video frame.
The target machine learning model may be various types of machine learning models, such as a deep neural network (Deep Neural Networks, DNN), a convolutional neural network (Convolutional Neural Networks, CNN), and so on.
Specifically, for each preset datum in the preset data set, the server inputs the input feature of the preset datum into the target machine learning model, so as to obtain the prediction input of the feature to be analyzed.
Step 206, determining the feature sensitivity of the input feature of each preset data based on the predicted output of each preset data.
The feature sensitivity is used for reflecting the importance degree of the input feature, and the higher the feature sensitivity is, the more important the input feature is.
Specifically, for each preset datum in the preset data set, after obtaining the predicted output of the preset datum, the server may determine the feature sensitivity of each input feature in the feature set to be analyzed of the preset datum based on the predicted output.
In some embodiments, for each input feature, the server calculates a partial derivative for each dimension of the input feature based on a predicted output of the preset data corresponding to the input feature, such that the input feature corresponds to a plurality of partial derivatives, the server further calculating a norm of the input feature based on the partial derivatives for each dimension of the input feature, the norm being the feature sensitivity of the input feature. The norms may be, for example, L1 norms, L2 norms, and so on.
And step 208, determining the average sensitivity of each feature to be analyzed in the feature set to be analyzed corresponding to the target machine learning model based on each feature sensitivity, and sorting each feature to be analyzed based on the average sensitivity of each feature to be analyzed to obtain a sorting result.
It can be understood that in the embodiment of the present application, all input features of the target machine learning model need to be analyzed, so that a set of all input features corresponding to the preset data set is a feature set to be analyzed. For example, assume that the preset data set is a game video frame set, including two game video frames, and the input features corresponding to the game video frame 1 are: the input characteristics corresponding to the game video frame 2 are that: and (3) the teammate position and teammate blood volume, and the feature set to be analyzed corresponding to the target machine learning model is { teammate position, enemy position and teammate blood volume }.
Specifically, the server determines a feature sensitivity set corresponding to each feature to be analyzed of the feature set to be analyzed based on all the feature sensitivities obtained through calculation, then averages the feature sensitivity sets to obtain average sensitivity for the feature sensitivity set corresponding to each feature to be analyzed, and finally orders all input features according to the feature sensitivity of each feature to be analyzed to obtain an ordering result, so that feature analysis of the machine learning model is completed. In the above example, assuming that the characteristic sensitivities of the teammate position and the enemy position obtained from the predicted output of the game video frame 1 are respectively X1 and Y1, and the characteristic sensitivities of the teammate position and the teammate blood volume obtained from the predicted output of the game video frame 2 are respectively X2 and Z1, the characteristic sensitivities of the teammate position obtained finally are (x1+x2)/2, the characteristic sensitivities of the enemy position are respectively Y1, and the characteristic sensitivities of the teammate blood volume are respectively Z1.
In the feature analysis method, the preset data set is obtained, the input features of all preset data in the preset data set are input into the target machine learning model to obtain the prediction output of all preset data, the feature sensitivity of the input features of all preset data is determined based on the prediction output of all preset data, the average sensitivity of all the features to be analyzed in the feature set to be analyzed corresponding to the target machine learning model is determined based on all the feature sensitivity, all the features to be analyzed are ranked based on the average sensitivity of all the features to be analyzed to obtain the ranking result, and in the method, all the features to be analyzed in the feature set to be analyzed are ranked based on the feature sensitivity, and the contribution degree of the features to the overall prediction result of the model can be better reflected because the feature sensitivity is determined based on the actual prediction output of the target machine learning model.
In one embodiment, the preset data set is a set of game video frames; the input features of the preset data include a first input feature and a second input feature. Referring to fig. 3, determining feature sensitivities of respective input features of respective preset data based on predicted outputs of the respective preset data, includes:
in step 302, target field information is obtained.
The target field information refers to field information set in the sampling condition. The server may obtain the target field information from one or more sampling conditions.
And step 304, analyzing each game video frame based on the target field information to obtain a first input characteristic of each game video frame.
Specifically, the server analyzes each game video frame to obtain basic field information corresponding to the game video frame, matches the basic field information with target field information to obtain field information corresponding to the target field information, and obtains a first input feature corresponding to the video frame. For example, the server may parse the game video frames to obtain base field information such as a location of each frame, a location of a teammate, a location of an enemy, a blood volume of each teammate, a blood volume of an enemy, and the like.
Step 306, calculating based on the target field information and the first input features of each game video frame to obtain the second input features of each game video frame.
Specifically, in the process of matching the target field information with the basic field information, for the target field information that fails to match, the server may calculate based on the basic field information (i.e., the first input feature) obtained by parsing, and the intermediate result obtained by calculation is used as the second input feature. For example, in a game, the number of teammates, the number of enemies, the total economy of own, and the like are calculated as intermediate results.
In the above embodiment, the video frame is parsed based on the target field information by acquiring the target field information, so as to obtain the first input feature, and the second input feature is obtained by calculating based on the target field information and the first input feature, so that the features of the video frame of the game can be extracted as completely as possible, and the accuracy and completeness of the feature data are improved.
In some embodiments, determining the feature sensitivity of the respective input feature of the respective preset data based on the predicted output of the respective preset data comprises: for each input feature, calculating a partial derivative for each dimension of the input feature based on a predicted output of preset data corresponding to the input feature; the norm of the input feature is calculated based on the respective partial derivatives corresponding to the input feature, and the norm is taken as the feature sensitivity of the input feature.
In some specific embodiments, for each input feature, the server directly calculates a partial derivative of the predicted output of the preset data corresponding to the input feature for each dimension of the input feature, where the calculation formula is as follows (formula 1):
wherein L is the predicted output of the preset data corresponding to the input characteristic,representing the input feature as a feature vector of the ith feature of the preset data and the current dimension as the jth dimension of the feature vector,/th feature of the preset data>The partial derivative is calculated.
In other specific embodiments, since there is a corresponding expected output for each of the preset data in the preset data set, the server may determine a corresponding predicted loss based on the predicted output and the expected output for the preset data corresponding to the input feature, and calculate a first partial derivative of the predicted loss for each dimension of the input feature. Specifically, the server may determine the corresponding prediction loss based on the difference between the predicted output and the desired output.
Further, the server calculates a norm of the input feature based on the partial derivatives of each dimension corresponding to the input feature, and takes the norm as the feature sensitivity of the input feature.
In some specific embodiments, the server may calculate the second norm of the input feature based on the partial derivatives of each dimension to which the input feature corresponds, with the calculation formula being formula (2) below,feature sensitivity for the ith input feature:
in some embodiments, the above method further comprises: determining target input characteristics according to the sorting result; and acquiring preset standard features, comparing the target input features with the preset standard features, and determining an evaluation result of the target machine learning model according to the comparison result.
The preset standard features are features of a target machine learning model which is set according to experience. Specifically, the server may select an input feature with a larger average sensitivity as a target input feature according to the sorting result, compare the target input feature with a preset standard feature, and determine that the target machine learning model accords with the evaluation pass when the matching degree of the comparison result representing the target input feature and the preset standard feature exceeds a preset threshold, that is, the performance of the target machine learning model accords with the expectation; otherwise, when the matching degree of the comparison result representation target input features and the preset standard features does not exceed the preset threshold, determining that the target machine learning model accords with the evaluation, namely that the performance of the target machine learning model does not accord with the expectation.
In some embodiments, after obtaining the comparison result, the server may further obtain accuracy of the target machine learning model, evaluate the target machine learning model based on the comparison result and the accuracy, and determine that the target machine learning model meets the evaluation when the matching degree of the comparison result representing the target input feature and the preset standard feature exceeds a preset threshold and the accuracy exceeds a corresponding preset threshold.
In some embodiments, the preset data in the preset data set is an image; the input characteristics of the preset data are all pixels in the image; the method further comprises the following steps: and the calculated characteristic sensitivity of each pixel is sent to a terminal, and the terminal is used for marking each image according to the characteristic sensitivity of each pixel corresponding to each image and visually displaying the marking result.
Specifically, after the server calculates the feature sensitivity of each feature of each image, the feature sensitivity of each pixel obtained by calculation can be sent to the terminal, the terminal marks each image according to the feature sensitivity of each pixel corresponding to each image, and the marked results are visually displayed.
Fig. 4 is a schematic diagram showing a visual display of labeling results in a specific embodiment. In this embodiment, the terminal marks the coil on the image with a similar contour, and as can be seen from fig. 4, when the machine learning model outputs different prediction results, the position where the sensitivity is concentrated is different, that is, the position where the model focuses on is different. By the visualization method, whether the positions of attention are reasonable when the model makes a decision can be analyzed, and reasons that the current prediction result possibly exists can be made based on the sensitivity interpretation model.
In some embodiments, the preset data in the preset data set is an image; the input characteristics of the preset data are all pixels in the image; the method further comprises the steps of: acquiring an image to be processed; performing pixel shielding on the image to be processed based on the sorting result to obtain a target characteristic image; and inputting the target characteristic image into a target machine learning model to obtain a predicted output corresponding to the image to be processed.
Specifically, after the server obtains the sorting result, the server can perform pixel shielding on the image to be processed based on the sorting result to obtain a target feature image, and the server further inputs the target image into a target machine learning model to obtain a prediction output corresponding to the image to be processed. Because the feature sensitivity reflects the importance degree of the features, the server can select the features important to the model decision based on the sorting result and mask the relatively unimportant features, and because the features are masked, the number of the features input into the target machine learning model is reduced, so that the influence of redundant features on the important features is reduced, and because of the reduction of the features, the number of the processed features is reduced when the server predicts through the target machine model, the prediction efficiency is improved, and the computer resources of the server are saved.
In some embodiments, the server may perform descending order arrangement on the pixels in the image to be processed based on the feature sensitivity, that is, the pixel values with large feature sensitivity are arranged in front and the pixel values with small feature sensitivity are arranged in back, and then set the preset number of pixel values arranged in the back to 0, so as to implement pixel shielding.
In some embodiments, before extracting the features, the server may obtain an evaluation result of the target machine learning model, and when the evaluation result of the target machine learning model characterizes that the performance of the target machine learning model meets the expectations, perform pixel masking on the image to be processed based on the sorting result, so that accuracy of the pixel masking may be improved.
It should be understood that, although the steps in the flowcharts of fig. 2-3 are shown in order as indicated by the arrows, these steps are not necessarily performed in order as indicated by the arrows. The steps are not strictly limited to the order of execution unless explicitly recited herein, and the steps may be executed in other orders. Moreover, at least some of the steps in fig. 2-3 may include multiple steps or stages that are not necessarily performed at the same time, but may be performed at different times, nor does the order in which the steps or stages are performed necessarily performed in sequence, but may be performed alternately or alternately with at least a portion of the steps or stages in other steps or other steps.
In one embodiment, as shown in FIG. 5, there is provided a feature analysis apparatus 500 comprising:
a data acquisition module 502, configured to acquire a preset data set;
the data input module 504 is configured to input features of each preset datum in the preset dataset into the target machine learning model, so as to obtain a predicted output of each preset datum;
a sensitivity determination module 506, configured to determine a feature sensitivity of the input feature of each preset data based on the predicted output of each preset data;
the ranking module 508 is configured to determine an average sensitivity of each feature to be analyzed in the feature set to be analyzed corresponding to the target machine learning model based on each feature sensitivity, and rank each feature to be analyzed based on the average sensitivity of each feature to be analyzed, so as to obtain a ranking result.
In some embodiments, the preset data set is a set of game video frames; the input features of the preset data comprise a first input feature and a second input feature; the sensitivity determination module is further configured to: acquiring target field information; analyzing each game video frame based on the target field information to obtain a first input characteristic of each game video frame; and calculating based on the target field information and the first input characteristics of each game video frame to obtain the second input characteristics of each game video frame.
In some embodiments, the sensitivity determination module is further to: for each input feature, calculating a partial derivative for each dimension of the input feature based on a predicted output of preset data corresponding to the input feature; the norm of the input feature is calculated based on the respective partial derivatives corresponding to the input feature, and the norm is taken as the feature sensitivity of the input feature.
In some embodiments, the sensitivity determination module is further to: determining a corresponding prediction loss based on the predicted output and the expected output of preset data corresponding to the input characteristics; first order partial derivatives of the predicted loss for each dimension of the input feature are calculated.
In some embodiments, the apparatus further comprises: the evaluation module is used for determining target input characteristics according to the sorting result; and acquiring preset standard features, comparing the target input features with the preset standard features, and determining an evaluation result of the target machine learning model according to the comparison result.
In some embodiments, the preset data in the preset data set is an image; the input characteristics of the preset data are all pixels in the image; the device further comprises: the sending module is used for sending the calculated characteristic sensitivity of each pixel to the terminal, and the terminal is used for marking each image according to the characteristic sensitivity of each pixel corresponding to each image and visually displaying the marking result.
In some embodiments, the preset data in the preset data set is an image; the input characteristics of the preset data are all pixels in the image; the device also comprises an extraction module for acquiring the image to be processed; performing pixel shielding on the image to be processed based on the sorting result to obtain a target characteristic image; and inputting the target characteristic image into a target machine learning model to obtain a predicted output corresponding to the image to be processed.
For specific limitations of the feature analysis apparatus, reference may be made to the above description of the feature analysis method, and no further description is given here. The respective modules in the above-described feature analysis apparatus may be implemented in whole or in part by software, hardware, and combinations thereof. The above modules may be embedded in hardware or may be independent of a processor in the computer device, or may be stored in software in a memory in the computer device, so that the processor may call and execute operations corresponding to the above modules.
In one embodiment, a computer device is provided, which may be a server, the internal structure of which may be as shown in fig. 6. The computer device includes a processor, a memory, and a network interface connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database of the computer device is used for storing data such as a preset data set, feature sensitivity and the like. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program is executed by a processor to implement a feature analysis method.
It will be appreciated by those skilled in the art that the structure shown in fig. 6 is merely a block diagram of some of the structures associated with the present application and is not limiting of the computer device to which the present application may be applied, and that a particular computer device may include more or fewer components than shown, or may combine certain components, or have a different arrangement of components.
In one embodiment, a computer device is provided comprising a memory and a processor, the memory having stored therein a computer program, the processor when executing the computer program performing the steps of: acquiring a preset data set;
inputting the input characteristics of each preset data in the preset data set into a target machine learning model to obtain the predicted output of each preset data; determining the feature sensitivity of the input features of each preset datum based on the predicted output of each preset datum; and determining the average sensitivity of each feature to be analyzed in the feature set to be analyzed corresponding to the target machine learning model based on each feature sensitivity, and sequencing each feature to be analyzed based on the average sensitivity of each feature to be analyzed to obtain a sequencing result.
In some embodiments, the preset data set is a set of game video frames; the input features of the preset data comprise a first input feature and a second input feature; determining feature sensitivity of the respective input features of the respective preset data based on the predicted output of the respective preset data, comprising: acquiring target field information; analyzing each game video frame based on the target field information to obtain a first input characteristic of each game video frame; and calculating based on the target field information and the first input characteristics of each game video frame to obtain the second input characteristics of each game video frame.
In some embodiments, determining the feature sensitivity of the respective input feature of the respective preset data based on the predicted output of the respective preset data comprises: for each input feature, calculating a partial derivative for each dimension of the input feature based on a predicted output of preset data corresponding to the input feature; the norm of the input feature is calculated based on the respective partial derivatives corresponding to the input feature, and the norm is taken as the feature sensitivity of the input feature.
In some embodiments, each preset data in the preset data set has a corresponding desired output; calculating the partial derivative for each dimension of the input feature based on the predicted output of the preset data corresponding to the input feature comprises: determining a corresponding prediction loss based on the predicted output and the expected output of preset data corresponding to the input characteristics; first order partial derivatives of the predicted loss for each dimension of the input feature are calculated.
In some embodiments, the processor when executing the computer program further performs the steps of: determining target input characteristics according to the sorting result; and acquiring preset standard features, comparing the target input features with the preset standard features, and determining an evaluation result of the target machine learning model according to the comparison result.
In some embodiments, the preset data in the preset data set is an image; the input characteristics of the preset data are all pixels in the image; the processor when executing the computer program also implements the steps of: and the calculated characteristic sensitivity of each pixel is sent to a terminal, and the terminal is used for marking each image according to the characteristic sensitivity of each pixel corresponding to each image and visually displaying the marking result.
In some embodiments, the preset data in the preset data set is an image; the input characteristics of the preset data are all pixels in the image; the processor when executing the computer program also implements the steps of: acquiring an image to be processed; performing pixel shielding on the image to be processed based on the sorting result to obtain a target characteristic image; and inputting the target characteristic image into a target machine learning model to obtain a predicted output corresponding to the image to be processed.
In one embodiment, a computer readable storage medium is provided having a computer program stored thereon, which when executed by a processor, performs the steps of: acquiring a preset data set; inputting the input characteristics of each preset data in the preset data set into a target machine learning model to obtain the predicted output of each preset data; determining the feature sensitivity of the input features of each preset datum based on the predicted output of each preset datum; and determining the average sensitivity of each feature to be analyzed in the feature set to be analyzed corresponding to the target machine learning model based on each feature sensitivity, and sequencing each feature to be analyzed based on the average sensitivity of each feature to be analyzed to obtain a sequencing result.
In some embodiments, the preset data set is a set of game video frames; the input features of the preset data comprise a first input feature and a second input feature; determining feature sensitivity of the respective input features of the respective preset data based on the predicted output of the respective preset data, comprising: acquiring target field information; analyzing each game video frame based on the target field information to obtain a first input characteristic of each game video frame; and calculating based on the target field information and the first input characteristics of each game video frame to obtain the second input characteristics of each game video frame.
In some embodiments, determining the feature sensitivity of the respective input feature of the respective preset data based on the predicted output of the respective preset data comprises: for each input feature, calculating a partial derivative for each dimension of the input feature based on a predicted output of preset data corresponding to the input feature; the norm of the input feature is calculated based on the respective partial derivatives corresponding to the input feature, and the norm is taken as the feature sensitivity of the input feature.
In some embodiments, each preset data in the preset data set has a corresponding desired output; calculating the partial derivative for each dimension of the input feature based on the predicted output of the preset data corresponding to the input feature comprises: determining a corresponding prediction loss based on the predicted output and the expected output of preset data corresponding to the input characteristics; first order partial derivatives of the predicted loss for each dimension of the input feature are calculated.
In some embodiments, the computer program when executed by the processor further performs the steps of: determining target input characteristics according to the sorting result; and acquiring preset standard features, comparing the target input features with the preset standard features, and determining an evaluation result of the target machine learning model according to the comparison result.
In some embodiments, the preset data in the preset data set is an image; the input characteristics of the preset data are all pixels in the image; the computer program when executed by the processor also performs the steps of: and the calculated characteristic sensitivity of each pixel is sent to a terminal, and the terminal is used for marking each image according to the characteristic sensitivity of each pixel corresponding to each image and visually displaying the marking result.
In some embodiments, the preset data in the preset data set is an image; the input characteristics of the preset data are all pixels in the image; the computer program when executed by the processor also performs the steps of: acquiring an image to be processed; performing pixel shielding on the image to be processed based on the sorting result to obtain a target characteristic image; and inputting the target characteristic image into a target machine learning model to obtain a predicted output corresponding to the image to be processed.
Those skilled in the art will appreciate that implementing all or part of the above described methods may be accomplished by way of a computer program stored on a non-transitory computer readable storage medium, which when executed, may comprise the steps of the embodiments of the methods described above. Any reference to memory, storage, database, or other medium used in embodiments provided herein may include at least one of non-volatile and volatile memory. The nonvolatile Memory may include Read-Only Memory (ROM), magnetic tape, floppy disk, flash Memory, optical Memory, or the like. Volatile memory can include random access memory (RandomAccess Memory, RAM) or external cache memory. By way of illustration, and not limitation, RAM can be in the form of a variety of forms, such as static random access memory (Static Random Access Memory, SRAM) or dynamic random access memory (Dynamic Random Access Memory, DRAM), and the like.
The technical features of the above embodiments may be arbitrarily combined, and all possible combinations of the technical features in the above embodiments are not described for brevity of description, however, as long as there is no contradiction between the combinations of the technical features, they should be considered as the scope of the description.
The above examples merely represent a few embodiments of the present application, which are described in more detail and are not to be construed as limiting the scope of the invention. It should be noted that it would be apparent to those skilled in the art that various modifications and improvements could be made without departing from the spirit of the present application, which would be within the scope of the present application. Accordingly, the scope of protection of the present application is to be determined by the claims appended hereto.
Claims (10)
1. A method of feature analysis, the method comprising:
acquiring a preset data set; the preset data set is a set formed by preset data, and the preset data is any one of images, voices or texts;
inputting the input characteristics of each preset datum in the preset data set into a target machine learning model to obtain the predicted output of each preset datum;
for each input feature, calculating a partial derivative for each dimension of the input feature based on a predicted output of preset data corresponding to the input feature;
calculating norms of the input features based on the partial derivatives corresponding to the input features, and taking the norms as feature sensitivity of the input features;
and determining the average sensitivity of each feature to be analyzed in the feature set to be analyzed corresponding to the target machine learning model based on each feature sensitivity, and sorting each feature to be analyzed based on the average sensitivity of each feature to be analyzed to obtain a sorting result.
2. The method of claim 1, wherein the predetermined data set is a set of game video frames; the input features of the preset data comprise a first input feature and a second input feature; the method further comprises the steps of:
acquiring target field information;
analyzing each game video frame based on the target field information to obtain a first input characteristic of each game video frame;
and calculating based on the target field information and the first input characteristics of each game video frame to obtain the second input characteristics of each game video frame.
3. The method of claim 1, wherein each preset data in the preset dataset has a corresponding desired output; the calculating partial derivatives of the predicted output based on the preset data corresponding to the input features for each dimension of the input features comprises:
determining a corresponding prediction loss based on a predicted output and an expected output of preset data corresponding to the input feature;
first order partial derivatives of the predictive loss for each dimension of the input feature are calculated.
4. The method according to claim 1, wherein the method further comprises:
determining target input characteristics according to the sorting result;
and acquiring preset standard features, comparing the target input features with the preset standard features, and determining an evaluation result of the target machine learning model according to the comparison result.
5. The method of claim 1, wherein the preset data in the preset data set is an image; the input characteristics of the preset data are all pixels in the image; the method further comprises the steps of:
and the calculated characteristic sensitivity of each pixel is sent to a terminal, and the terminal is used for marking each image according to the characteristic sensitivity of each pixel corresponding to each image and visually displaying the marking result.
6. The method of claim 1, wherein the preset data in the preset data set is an image; the input characteristics of the preset data are all pixels in the image; the method further comprises the steps of:
acquiring an image to be processed;
performing pixel shielding on the image to be processed based on the sorting result to obtain a target characteristic image;
and inputting the target characteristic image into the target machine learning model to obtain a predicted output corresponding to the image to be processed.
7. A feature analysis device, the device comprising:
the data acquisition module is used for acquiring a preset data set; the preset data set is a set formed by preset data, and the preset data is any one of images, voices or texts; the data input module is used for inputting the input characteristics of each preset data in the preset data set into a target machine learning model to obtain the prediction output of each preset data;
the sensitivity determination module is used for calculating partial derivatives of each input feature according to the predicted output of preset data corresponding to the input feature; calculating norms of the input features based on the partial derivatives corresponding to the input features, and taking the norms as feature sensitivity of the input features;
the sorting module is used for determining the average sensitivity of each feature to be analyzed in the feature set to be analyzed corresponding to the target machine learning model based on each feature sensitivity, and sorting each feature to be analyzed based on the average sensitivity of each feature to be analyzed to obtain a sorting result.
8. The apparatus of claim 7, wherein the sensitivity determination module is further configured to determine a corresponding prediction loss based on a predicted output and an expected output of preset data corresponding to the input feature; first order partial derivatives of the predictive loss for each dimension of the input feature are calculated.
9. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the method of any of claims 1 to 6 when the computer program is executed.
10. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110394351.5A CN113139447B (en) | 2021-04-13 | 2021-04-13 | Feature analysis method, device, computer equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110394351.5A CN113139447B (en) | 2021-04-13 | 2021-04-13 | Feature analysis method, device, computer equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113139447A CN113139447A (en) | 2021-07-20 |
| CN113139447B true CN113139447B (en) | 2024-01-30 |
Family
ID=76811993
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110394351.5A Active CN113139447B (en) | 2021-04-13 | 2021-04-13 | Feature analysis method, device, computer equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113139447B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111461341A (en) * | 2020-03-18 | 2020-07-28 | 支付宝(杭州)信息技术有限公司 | Method and device for evaluating contribution degree of data source and computer equipment |
| CN111612163A (en) * | 2020-06-28 | 2020-09-01 | 上海优扬新媒信息技术有限公司 | Training method and device based on machine learning model |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP7202973B2 (en) * | 2019-05-29 | 2023-01-12 | 株式会社日立製作所 | Data analysis device, data analysis method, and data analysis program |
-
2021
- 2021-04-13 CN CN202110394351.5A patent/CN113139447B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111461341A (en) * | 2020-03-18 | 2020-07-28 | 支付宝(杭州)信息技术有限公司 | Method and device for evaluating contribution degree of data source and computer equipment |
| CN111612163A (en) * | 2020-06-28 | 2020-09-01 | 上海优扬新媒信息技术有限公司 | Training method and device based on machine learning model |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113139447A (en) | 2021-07-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2020228446A1 (en) | Model training method and apparatus, and terminal and storage medium | |
| EP2806374A1 (en) | Method and system for automatic selection of one or more image processing algorithm | |
| US20230119593A1 (en) | Method and apparatus for training facial feature extraction model, method and apparatus for extracting facial features, device, and storage medium | |
| CN110046706B (en) | Model generation method and device and server | |
| CN110210625B (en) | Modeling method and device based on transfer learning, computer equipment and storage medium | |
| CN110909539A (en) | Word generation method, system, computer device and storage medium of corpus | |
| CN109285105A (en) | Method of detecting watermarks, device, computer equipment and storage medium | |
| CN110135505B (en) | Image classification method and device, computer equipment and computer readable storage medium | |
| CN110503149B (en) | Method and system for classifying local features in image | |
| CN111401192B (en) | Model training method and related device based on artificial intelligence | |
| CN113657087B (en) | Information matching method and device | |
| US20230021551A1 (en) | Using training images and scaled training images to train an image segmentation model | |
| CN110263326A (en) | A kind of user's behavior prediction method, prediction meanss, storage medium and terminal device | |
| CN112784054A (en) | Concept graph processing apparatus, concept graph processing method, and computer-readable medium | |
| CN111582459B (en) | Method for executing operation, electronic equipment, device and storage medium | |
| CN111507137A (en) | Action understanding method and device, computer equipment and storage medium | |
| CN112132278A (en) | Model compression method and device, computer equipment and storage medium | |
| CN110968664A (en) | Document retrieval method, device, equipment and medium | |
| CN110533184B (en) | Network model training method and device | |
| CN115344805A (en) | Material auditing method, computing equipment and storage medium | |
| CN113806564B (en) | Multi-mode informative text detection method and system | |
| CN110909768A (en) | Method and device for acquiring marked data | |
| CN114328942A (en) | Relationship extraction method, apparatus, device, storage medium and computer program product | |
| CN114549849A (en) | Image recognition method and device, computer equipment and storage medium | |
| CN113139447B (en) | Feature analysis method, device, computer equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |