Feature screening method, device, storage medium and electronic device

Abstract

The present disclosure relates to a feature screening method, device, storage medium and electronic device. The method determines a first feature weight and a second feature weight corresponding to each condition feature according to specified feature data of multiple target diagnosis objects; determines a target feature weight of the condition feature according to the first feature weight and the second feature weight corresponding to each condition feature, and the target feature weight is used to characterize the importance of the condition feature for determining that the target diagnosis object is the target case object or the non-target case object; determines the target condition feature from multiple condition features according to the target feature weight corresponding to each condition feature, which can effectively improve the reliability of the feature screening results and effectively reduce the amount of calculation required for feature screening, thereby effectively improving the feature screening rate and providing a reliable data basis for later diagnosis and later model training of medical researchers.

Description

Technical Field

[0001] The present disclosure relates to the field of data processing, and in particular, to a feature screening method, device, storage medium and electronic device. Background Art

[0002] Feature screening is an important data preprocessing process in the field of machine learning. The purpose of feature screening is to screen out feature data that has a greater impact on model recognition results from feature data of many dimensions when the data dimension is large. Especially in the relevant research on psoriasis diagnosis, it is found that psoriasis patients have more lesion characteristics, such as joint lesions, tendon lesions, and bursal lesions. Among them, the joint lesion characteristics include sacroiliac joint effusion status characteristics, sacroiliac joint blood flow status characteristics, sacroiliac joint osteophyte status characteristics, hip joint effusion status characteristics, hip wrist blood flow status characteristics, hip osteophyte status characteristics. Characteristics of wrist effusion, wrist blood flow, wrist osteophyte, elbow effusion, elbow blood flow, elbow blood flow, etc. In addition, tendon lesions and bursa lesions also include effusion, blood flow, osteophyte and other condition characteristics. This leads to a huge number of undiagnosed features involved in the entire psoriasis diagnosis process. Such a large amount of undiagnosed feature data is obviously not conducive to medical researchers' diagnosis of psoriasis, nor is it conducive to the training of relevant human body state judgment models. Therefore, it is necessary to screen out relatively important features from the numerous undiagnosed features.

[0003] However, current feature screening methods usually have problems such as large amount of computation required for the feature screening process and low feature screening efficiency, or poor reliability of the screening results and easy omission of important features. Summary of the invention

[0004] The purpose of the present disclosure is to provide a feature screening method, device, storage medium and electronic device.

[0005] In order to achieve the above object, the present disclosure provides a feature screening method in a first aspect, the method comprising:

[0006] Acquire designated feature data of each of the target diagnostic objects among a plurality of target diagnostic objects, the designated feature data including a plurality of condition features and category labeling data corresponding to the target diagnostic object, the category labeling data being used to label the target diagnostic object as a target case object or a non-target case object;

[0007] Determining, according to the specified feature data of the plurality of target diagnostic objects, a first feature weight and a second feature weight corresponding to each of the condition features;

[0008] Determine a target feature weight of the condition feature according to the first feature weight and the second feature weight corresponding to each condition feature, wherein the target feature weight is used to characterize the importance of the condition feature for determining that the target diagnosis object is the target case object or the non-target case object;

[0009] The target situation feature is determined from the plurality of situation features according to the target feature weight corresponding to each situation feature.

[0010] A second aspect of the present disclosure provides a feature screening device, the device comprising:

[0011] an acquisition module, configured to acquire designated feature data of each of a plurality of target diagnostic objects, the designated feature data including a plurality of condition features, and category labeling data corresponding to the target diagnostic object, the category labeling data being used to label the target diagnostic object as a target case object or a non-target case object;

[0012] A first determining module is configured to determine a first feature weight and a second feature weight corresponding to each of the condition features according to the specified feature data of the plurality of target diagnostic objects;

[0013] A second determination module is configured to determine a target feature weight of the condition feature according to the first feature weight and the second feature weight corresponding to each condition feature, wherein the target feature weight is used to characterize the importance of the condition feature for determining that the target diagnosis object is the target case object or the non-target case object;

[0014] The third determination module is configured to determine the target situation feature from the multiple situation features according to the target feature weight corresponding to each situation feature.

[0015] A third aspect of the present disclosure provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect above.

[0016] A fourth aspect of the present disclosure provides an electronic device, including:

[0017] a memory having a computer program stored thereon;

[0018] A processor is used to execute the computer program in the memory to implement the steps of the method described in the first aspect above.

[0019] The above technical scheme can determine the first feature weight and the second feature weight of each of the condition features according to the specified feature data of the multiple target diagnostic objects, and then determine the target feature weight of the condition feature according to the first feature weight and the second feature weight, and determine the target condition feature from the multiple condition features according to the target feature weight corresponding to each of the condition features. It can effectively improve the reliability of the feature screening results and effectively reduce the amount of calculation required for feature screening, thereby not only effectively improving the feature screening rate, but also providing a reliable data basis for the later diagnosis and later model training of medical researchers.

[0020] Other features and advantages of the present disclosure will be described in detail in the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The accompanying drawings are used to provide a further understanding of the present disclosure and constitute a part of the specification. Together with the following specific embodiments, they are used to explain the present disclosure but do not constitute a limitation of the present disclosure. In the accompanying drawings:

[0022] Figure 1 is a flow chart of a feature screening method shown in an exemplary embodiment of the present disclosure;

[0023] Figure 2 According to this disclosure Figure 1 A flow chart of a feature screening method shown in the illustrated embodiment;

[0024] Figure 3 According to this disclosure Figure 1 A flow chart of another feature screening method shown in the illustrated embodiment;

[0025] Figure 4 According to this disclosure Figure 1 A flow chart of another feature screening method shown in the illustrated embodiment;

[0026] Figure 5 is a block diagram of a feature screening device shown in an exemplary embodiment of the present disclosure;

[0027] Figure 6 is a block diagram of an electronic device according to an exemplary embodiment;

[0028] Figure 7 is a block diagram of another electronic device according to an exemplary embodiment. DETAILED DESCRIPTION

[0029] The specific implementation of the present disclosure is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described herein is only used to illustrate and explain the present disclosure, and is not used to limit the present disclosure.

[0030] Before introducing the specific implementation methods of the present disclosure in detail, the application scenarios of the present disclosure are first described as follows. The present disclosure can be applied to the process of feature screening for diagnostic features of a certain case, for example, the process of feature screening for human feature data to obtain condition features that have a greater impact on the diagnosis of psoriasis. It can also be applied to the process of preprocessing a large number of human feature data before training a human state judgment model to obtain condition features that have a greater impact on human state judgment. Here, the process of feature screening for human feature data to obtain condition features that have a greater impact on the diagnosis of psoriasis is used as an example for explanation. In the relevant research on psoriasis diagnosis, it was found that psoriasis patients have more lesion features, such as joint lesions, tendon lesions, and bursal lesions, among which the joint lesions The lesion characteristics include sacroiliac joint effusion status characteristics, sacroiliac joint blood flow status characteristics, sacroiliac joint osteophyte status characteristics, hip joint effusion status characteristics, hip and wrist blood flow status characteristics, hip joint osteophyte status characteristics, wrist joint effusion status characteristics, wrist joint blood flow status characteristics, wrist joint osteophyte status characteristics, elbow joint effusion status characteristics, elbow joint blood flow status characteristics, elbow joint blood flow status characteristics, etc. Tendon lesions and burseal lesions also include effusion, blood flow, osteophyte and other condition characteristics. This leads to a huge number of undiagnosed features involved in the entire psoriasis diagnosis process. Such a large amount of undiagnosed feature data is obviously not conducive to medical researchers' diagnosis of psoriasis, nor is it conducive to the training of related disease diagnosis models. Therefore, it is necessary to screen out relatively important features from the numerous undiagnosed features. However, current feature screening usually either uses the important features in people's subjective impressions as feature screening results, or trains models for different features and then determines whether the features used in the training model are important features based on the accuracy of the model recognition results. However, the inventors have found that the above method of using the important features in people's subjective impressions as feature screening results has the problem of strong subjective factors, poor reliability of screening results, and easy omission of important features. Another method of determining important features through the accuracy of model recognition results has the problem of large computational complexity and low feature screening efficiency.

[0031] In order to solve the above technical problems, the present disclosure provides a feature screening method, device, storage medium and electronic device, which obtains designated feature data of each target diagnostic object among multiple target diagnostic objects, wherein the designated feature data includes multiple condition features and category annotation data corresponding to the target diagnostic object, and the category annotation data is used to mark the target diagnostic object as a target case object or a non-target case object; according to the designated feature data of the multiple target diagnostic objects, a first feature weight and a second feature weight corresponding to each condition feature are determined; according to the first feature weight and the second feature weight corresponding to each condition feature, a target feature weight of the condition feature is determined, and the target feature weight is used to characterize the importance of the condition feature for determining that the target diagnostic object is the target case object or the non-target case object; according to the target feature weight corresponding to each condition feature, the target condition feature is determined from multiple condition features, which can effectively improve the reliability of the feature screening results and effectively reduce the amount of calculation required for feature screening, thereby not only effectively improving the feature screening rate, but also providing a reliable data basis for the later diagnosis and later model training of medical researchers.

[0032] The technical solution of the present disclosure is described below in conjunction with specific embodiments.

[0033] Figure 1 is a flow chart of a feature screening method shown in an exemplary embodiment of the present disclosure; Figure 1 As shown, the method may include:

[0034] Step 101 : obtaining designated feature data of each target diagnosis object among a plurality of target diagnosis objects, wherein the designated feature data includes a plurality of status features and category labeling data corresponding to the target diagnosis object.

[0035] Among them, the category annotation data is used to label the target diagnosis object as a target case object or a non-target case object. The target diagnosis object can be image data, audio data or text data, or it can be a specified device, animal, or human body.

[0036] For example, the target diagnostic object may be a human object, the multiple target diagnostic objects may include human object A, human object B, human object C, etc., the target case object may be a psoriasis patient, and the non-target case object may be a non-psoriasis patient.

[0037] Step 102: Determine a first feature weight and a second feature weight corresponding to each of the condition features according to the designated feature data of the multiple target diagnostic objects.

[0038] Among them, the first feature weight can be a description of the importance of the status feature for determining that the target diagnosis object is the target case object or the non-target case object from the perspective of a single factor feature; the second feature weight can be a description of the importance of the status feature for determining that the target diagnosis object is the target case object or the non-target case object from the perspective of multidimensional feature fusion.

[0039] By way of example, the designated feature data may include sacroiliac joint effusion condition characteristics, sacroiliac joint blood flow condition characteristics, sacroiliac joint osteophyte condition characteristics, hip joint effusion condition characteristics, hip and wrist blood flow condition characteristics, hip joint osteophyte condition characteristics, wrist joint effusion condition characteristics, wrist joint blood flow condition characteristics, wrist joint osteophyte condition characteristics, elbow joint effusion condition characteristics, elbow joint blood flow condition characteristics, elbow joint blood flow condition characteristics, tendon effusion condition characteristics, tendon blood flow condition characteristics, tendon osteophyte condition characteristics, bursa effusion condition characteristics, bursa blood flow condition characteristics, and bursa osteophyte condition characteristics. When the target case object is a psoriasis patient and the non-target case object is a non-psoriasis patient, the first feature weight may be a description of the importance of the condition feature for determining that the human object is the psoriasis patient from the perspective of a single factor feature; the second feature weight may be a description of the importance of the condition feature for determining that the human object is the psoriasis patient from the perspective of a multidimensional feature fusion.

[0040] Step 103: determine the target feature weight of each situation feature according to the first feature weight and the second feature weight corresponding to each situation feature.

[0041] The target feature weight is used to characterize the importance of the condition feature in determining whether the target diagnosis object is the target case object or the non-target case object.

[0042] In this step, the first feature weight and the second feature weight can be weighted and summed to obtain the target feature weight. When performing the weighted summation, the weights corresponding to the first feature weight and the second feature weight can be pre-set. For example, when determining the target feature weight W corresponding to the i-th condition feature, the first feature weight corresponding to the i-th condition feature is represented by WA, and the second feature weight corresponding to the i-th condition feature is represented by WB, W=aWA+bWA, where a is the weight corresponding to the first feature weight WA, and b is the weight corresponding to the second feature weight WB. a and b can be pre-set.

[0043] In addition, in this step, when the target case object is a psoriasis patient and the non-target case object is a non-psoriasis patient, the target feature weight is used to characterize the importance of the condition feature in determining that the human subject is the psoriasis patient.

[0044] Step 104 , determining the target situation feature from the plurality of situation features according to the target feature weight corresponding to each situation feature.

[0045] In this step, a preset number of situation features with the largest weights of the target feature may be used as the target situation feature.

[0046] The above technical scheme can determine the first feature weight and the second feature weight of each condition feature according to the specified feature data of the multiple target diagnostic objects, and then determine the target feature weight of the condition feature according to the first feature weight and the second feature weight, and determine the target condition feature from the multiple condition features according to the target feature weight corresponding to each condition feature. It can effectively improve the reliability of the feature screening results and effectively reduce the amount of calculation required for feature screening, thereby not only effectively improving the feature screening rate, but also providing a reliable data basis for the later diagnosis and later model training of medical researchers.

[0047] Figure 2 According to this disclosure Figure 1 The embodiment shown is a flow chart of a feature screening method; Figure 2 As shown, the step 102 of determining the first feature weight corresponding to each condition feature according to the specified feature data of the multiple target diagnostic objects may include the following steps:

[0048] Step 1021 : obtaining the Euclidean distance between a specified target diagnosis object among the plurality of target diagnosis objects and other target diagnosis objects among the plurality of target diagnosis objects except the specified target diagnosis object.

[0049] For example, when multiple target diagnostic objects include human body object AZ, the Euclidean distance between a specified target diagnostic object in human body object AZ (the specified target diagnostic object may be any one of human body objects AZ) and each target diagnostic object in other target diagnostic objects except the specified target diagnostic object is obtained respectively. For example, when the specified target diagnostic object is human body object A, the Euclidean distance between human body object A and each human body object in human body object BZ can be obtained; when the specified target diagnostic object is human body object B, the Euclidean distance between human body object B and each human body object in human body object A and human body object CZ can be obtained; when the specified target diagnostic object is human body object Z, the Euclidean distance between human body object Z and each human body object in human body object AY can be obtained.

[0050] Step 1022: The target diagnostic object whose Euclidean distance with the other target diagnostic objects is less than a first preset distance threshold is taken as an element in a first target diagnostic object set corresponding to the designated target diagnostic object.

[0051] For example, in the case where the designated target diagnostic object is human object A, if the target diagnostic objects whose Euclidean distance between human object A and each human object in human object BZ is greater than the first preset distance threshold include human object B, human object E, human object F and human object G, that is, human object B, human object E, human object F, and human object G are elements in the first target diagnostic object set corresponding to human object A. Similarly, the elements in the first target diagnostic object set corresponding to each human object in human object BZ can be obtained.

[0052] Step 1023, when it is determined that the first target diagnosis object set includes multiple target case objects and multiple non-target case objects, the designated target diagnosis object corresponding to the first target diagnosis object set is used as the pending target diagnosis object to obtain a second target diagnosis object set consisting of multiple pending target diagnosis objects.

[0053] Among them, whether each target diagnostic object is a target case object can be determined through the category annotation data corresponding to each target diagnostic object, and whether each target diagnostic object is a non-target case object can also be determined through the category annotation data corresponding to each target diagnostic object.

[0054] For example, still taking the example shown in the above step 1022 as an example for explanation, the human object B, human object E, human object F, and human object G are elements in the first target diagnosis object set corresponding to the human object A, the target case object is a psoriasis patient, and the non-target case object is a non-psoriasis patient. If the human object B is a non-psoriasis patient, the human object E is a psoriasis patient, the human object F is a non-psoriasis patient, and the human object G is a psoriasis patient, then the first target diagnosis object set corresponding to the human object A includes 2 psoriasis patients and 2 non-psoriasis patients, so the human object A can be used as a target case to be diagnosed. A target diagnostic object is determined, that is, as an element in the second target diagnostic object set. Similarly, it is determined whether the first target diagnostic object set corresponding to each human object in the human object BZ includes multiple psoriasis patients and multiple non-psoriasis patients. When it is determined that the first target diagnostic object set corresponding to the human object includes multiple psoriasis patients and multiple non-psoriasis patients, the designated target diagnostic object is used as the to-be-determined target diagnostic object, thereby determining whether the human object BZ is an element in the second target diagnostic object set, thereby obtaining each element in the second target diagnostic object set, namely forming the second target diagnostic object set.

[0055] Step 1024, obtaining target similar objects and target non-similar objects corresponding to each of the pending target diagnosis objects in the first target diagnosis object set.

[0056] In this step, when the target diagnosis object to be determined is a target case object, a first target distance between the target diagnosis object to be determined and a target case object in the first target diagnosis object set is obtained, and a second target distance between the target diagnosis object to be determined and a non-target case object in the first target diagnosis object set is obtained, and a preset number of target case objects in the first target diagnosis object set with the smallest first target distance are taken as the target similar objects, and a preset number of non-target case objects in the first target diagnosis object set with the smallest second target distance are taken as the target non-similar objects;

[0057] In the case that the target diagnosis object to be determined is a non-target case object, a third target distance between the target diagnosis object to be determined and a target case object in the first target diagnosis object set, and a fourth target distance between the target diagnosis object to be determined and a non-target case object in the first target diagnosis object set are obtained, and a preset number of non-target case objects in the first target diagnosis object set with the smallest fourth target distance are taken as the target similar objects, and a preset number of target case objects in the first target diagnosis object set with the smallest third target distance are taken as the target non-similar objects.

[0058] Step 1025, obtaining the first average distance between each condition feature of each target diagnostic object to be determined and the condition feature of the target similar object corresponding to the target diagnostic object to be determined, and the second average distance between each condition feature of each target diagnostic object to be determined and the condition feature of the target non-similar object corresponding to the target diagnostic object to be determined.

[0059] For example, if the human object A is a psoriasis patient, the target similar objects of the human object A include human object E and human object G, and the target non-similar objects of the human object A include human object B and human object F. Each human object may include 20 condition features, and the 20 condition features may be sacroiliac joint effusion condition features, sacroiliac joint blood flow condition features, sacroiliac joint osteophyte condition features, hip joint effusion condition features, hip joint wrist blood flow condition features, hip joint osteophyte condition features, wrist joint effusion condition features, wrist joint blood flow condition features, wrist joint osteophyte condition features, elbow joint effusion condition features, elbow joint blood flow condition features, elbow joint blood flow condition features, tendon effusion condition features, tendon blood flow condition features, tendon osteophyte condition features, bursal effusion condition features, bursal blood flow condition features, bursal osteophyte condition features, height features, weight features, etc., wherein each condition feature of the human object A is represented by A. i Indicates that each condition feature of human object B is represented by B i Indicates that each condition feature of human object E is represented by E i Indicates that each condition feature of the human object F is represented by F iIndicates that each condition feature of the human object G is represented by G i Represents, where i = 1, 2…20; the first average distance corresponding to the nth condition feature of human object A is The second average distance corresponding to the nth condition feature of the human object A is

[0060] Step 1026, determining the first feature weight corresponding to each condition feature of each of the pending target diagnosis objects in the second target diagnosis object set in turn according to the first average distance and the second average distance corresponding to each condition feature of the pending target diagnosis objects.

[0061] In this step, the first preset initial feature weight corresponding to each condition feature can be obtained, and each condition feature of each target diagnostic object to be determined in the second target diagnostic object set is traversed, and the following weight adjustment step is performed to obtain the updated first preset initial feature weight of each condition feature, and the updated first preset initial feature weight of each condition feature is used as the first feature weight corresponding to the condition feature;

[0062] The weight adjustment step includes:

[0063] When the first average distance is smaller than the second average distance, the first preset initial feature weight of the current condition feature is increased by a first preset weight threshold; when the first average distance is larger than the second average distance, the first preset initial feature weight of the current condition feature is reduced by a second preset weight threshold; when the first average distance is equal to the second average distance, the first preset initial feature weight of the current condition feature is maintained unchanged.

[0064] It should be noted that the first preset weight threshold corresponding to each situation feature may be the difference between the first average distance and the second average distance corresponding to the situation feature, or may be other preset values.

[0065] For example, when a certain condition feature of the human object A corresponds to the current first preset initial feature weight WU, the updated first preset initial feature weight can be expressed as:

[0066]

[0067] Among them, H is the set of target similar objects. Since the target similar objects corresponding to the human object A are human object E and human object G, there are 2 elements in the set H. M is the set of target non-similar objects. The target non-similar objects corresponding to the human object A are human object B and target human object F. Therefore, there are 2 elements in the set M. i is the serial number representation of the status feature.

[0068] The above technical scheme can effectively obtain the first feature weight corresponding to each condition feature, that is, describe the importance of each condition feature for determining the target case object or non-target case object from a single factor perspective, and can provide reliable data preparation for subsequent feature screening.

[0069] Figure 3 According to this disclosure Figure 1 The embodiment shown is a flow chart of another feature screening method; Figure 3 As shown, the step 102 of determining the second feature weight corresponding to each condition feature according to the specified feature data of the multiple target diagnostic objects may include the following steps:

[0070] Step 1027, obtaining a preset initial model, where the preset initial model includes a second preset initial feature weight for each condition feature.

[0071] The preset initial model may be:

[0072] y pre =w1d1+w2d2+…+w N d N ,

[0073] In the above formula, w i is the second preset initial weight, d i Characteristics of the situation.

[0074] For example, still taking the target diagnosis object as a human object, the target case object as a psoriasis patient, and the non-target case object as a non-psoriasis patient as an example, the specified feature data may include the following 20 condition features, of which d1 is a sacroiliac joint effusion condition feature, d2 is a sacroiliac joint blood flow condition feature, d3 is a sacroiliac joint osteophyte condition feature, d4 is a hip joint effusion condition feature, d5 is a hip joint osteophyte condition feature, d6 is a wrist joint effusion condition feature, d7 is a wrist joint blood flow condition feature, d8 is a wrist joint osteophyte condition feature, d9 is an elbow joint effusion condition feature, and d10 is a sacroiliac joint effusion condition feature. 10 is the characteristic of blood flow in elbow joint, d 11 is the characteristic of blood flow in elbow joint, d 12 Characteristic of tendon effusion, d 13 Characteristic of tendon blood flow. 14 Characteristic of tendon osteophyte condition, d 15 Characteristic of bursal effusion, d 16 Characterizes the blood flow of the bursa, and d 17 Characteristic of synovial osteophyte condition, d 18 Characteristic of blood flow in hip joint and wrist 19 is the height feature, d 20is the weight characteristic, w i is the condition characteristic d i The corresponding second preset initial weight.

[0075] Step 1028, using the designated feature data of the multiple target diagnostic objects as training data, iteratively train the preset initial model to obtain a target judgment model, wherein the target judgment model includes the second feature weight corresponding to each condition feature.

[0076] In this step, you can Figure 4 The steps S1 to S4 shown in the figure are implemented, Figure 4 According to this disclosure Figure 1 The embodiment shown is a flowchart of another feature screening method, such as Figure 4 As shown, the implementation process of step 1028 may include:

[0077] S1, taking a designated number of target case objects among the multiple target diagnosis objects as reference objects.

[0078] S2, obtaining a target difference between each condition feature in each target diagnosis object except the reference object and the condition feature in the reference object.

[0079] For example, if the multiple target diagnostic objects include human object A, human object M, human object N, human object L and human object K, and the reference objects are human object M, human object N, human object L and human object K, then the target difference between each condition feature of human object A and human object M, human object N, human object L and human object K is obtained, for example, A i -M i , A i -N i , A i -L i , A i -K i , i represents the i-th condition feature.

[0080] S3, determining a fusion feature corresponding to each of the situation features according to the target difference of the situation features.

[0081] In this step, a preset activation function can be obtained; the target difference value of each condition feature is input into the preset activation function to obtain the activation function value corresponding to each target difference value; the target and value of the activation function values ​​corresponding to the condition features of multiple target diagnostic objects are obtained; and the target and value of the condition feature are used as the fusion feature corresponding to the condition feature.

[0082] For example, the preset activation function is f(x), then the fusion feature corresponding to the i-th condition feature can be expressed as: f(A i -M i )+f(A i -N i )+f(A i -L i )+f(A i -K i ) = d i .

[0083] S4, iteratively training the preset initial model according to the fusion feature corresponding to each condition feature and the preset loss function to obtain the target judgment model.

[0084] In this step, the fused feature corresponding to each condition feature can be input into the preset initial model to obtain the objective function value; the current loss value is determined by the preset loss function according to the objective function value; it is determined whether the current loss value is less than or equal to the preset loss value threshold; when the current loss value is greater than the preset loss value threshold, the current second preset initial feature weight is updated according to the gradient descent method to obtain the updated preset initial model, and the step of inputting the fused feature corresponding to each condition feature into the preset initial model to obtain the objective function value to determining whether the current loss value is less than or equal to the preset loss value threshold is executed repeatedly, until when the current loss value is less than or equal to the preset loss value threshold, the current second preset initial feature weight of each condition feature is used as the second feature weight corresponding to the condition feature.

[0085] Among them, the preset loss function can be:

[0086] L=(y pre -y true ) 2 +α||W||

[0087] In the above function, ||W|| is w1 to w N The second norm of y true is the labeled value, y pre The function value calculated for the current model.

[0088] The above technical scheme can effectively obtain the second feature weight corresponding to each condition feature, that is, describe the importance of each condition feature for determining the target case object or non-target case object from the perspective of multi-factor fusion, and provide reliable data preparation for subsequent feature screening.

[0089] Figure 5 is a block diagram of a feature screening device shown in an exemplary embodiment of the present disclosure; see Figure 5, the device may include:

[0090] The acquisition module 501 is configured to acquire the specified characteristic data of each target diagnosis object among multiple target diagnosis objects, wherein the specified characteristic data includes multiple condition characteristics and the category labeling data corresponding to the target diagnosis object, wherein the category labeling data is used to label the target diagnosis object as a target case object or a non-target case object;

[0091] A first determining module 502 is configured to determine a first feature weight and a second feature weight corresponding to each condition feature according to the specified feature data of the multiple target diagnostic objects;

[0092] The second determination module 503 is configured to determine the target feature weight of the condition feature according to the first feature weight and the second feature weight corresponding to each condition feature, and the target feature weight is used to represent the importance of the condition feature for determining that the target diagnosis object is the target case object or the non-target case object;

[0093] The third determination module 504 is configured to determine the target situation feature from the multiple situation features according to the target feature weight corresponding to each situation feature.

[0094] The above technical scheme can determine the first feature weight and the second feature weight of each condition feature according to the specified feature data of the multiple target diagnostic objects, and then determine the target feature weight of the condition feature according to the first feature weight and the second feature weight, and determine the target condition feature from the multiple condition features according to the target feature weight corresponding to each condition feature. It can effectively improve the reliability of the feature screening results and effectively reduce the amount of calculation required for feature screening, thereby not only providing a reliable data basis for the later model training, but also effectively improving the feature screening rate.

[0095] Optionally, the first determining module 502 is configured to:

[0096] Obtaining the Euclidean distance between a specified target diagnosis object among the plurality of target diagnosis objects and other target diagnosis objects among the plurality of target diagnosis objects except the specified target diagnosis object;

[0097] The target diagnostic object whose Euclidean distance with the other target diagnostic objects is less than a first preset distance threshold is taken as an element in the first target diagnostic object set corresponding to the designated target diagnostic object;

[0098] When it is determined that the first target diagnosis object set includes a plurality of target case objects and a plurality of non-target case objects, the designated target diagnosis object corresponding to the first target diagnosis object set is used as a pending target diagnosis object to obtain a second target diagnosis object set consisting of a plurality of the pending target diagnosis objects;

[0099] Obtaining target similar objects and target non-similar objects corresponding to each of the pending target diagnostic objects in the first target diagnostic object set;

[0100] Obtaining a first average distance between each condition feature of each target object to be diagnosed and the condition feature of a target object of the same type corresponding to the target object to be diagnosed, and a second average distance between each condition feature of each target object to be diagnosed and the condition feature of a target object of a different type corresponding to the target object to be diagnosed;

[0101] The first feature weight corresponding to each condition feature of each of the target diagnostic objects to be determined in the second target diagnostic object set is determined in sequence according to the first average distance and the second average distance corresponding to each condition feature of the target diagnostic objects to be determined.

[0102] Optionally, the first determining module 502 is configured to:

[0103] In the case where the target diagnosis object to be determined is a target case object, a first target distance between the target diagnosis object to be determined and a target case object in the first target diagnosis object set is obtained, and a second target distance between the target diagnosis object to be determined and a non-target case object in the first target diagnosis object set is obtained, and a preset number of target case objects in the first target diagnosis object set with the smallest first target distances are taken as the target similar objects, and a preset number of non-target case objects in the first target diagnosis object set with the smallest second target distances are taken as the target non-similar objects;

[0104] In the case of a non-target case object of the target diagnostic object to be determined, a third target distance between the target diagnostic object to be determined and the target case object in the first target diagnostic object set, and a fourth target distance between the target diagnostic object to be determined and the non-target case object in the first target diagnostic object set are obtained, and a preset number of non-target case objects with the smallest fourth target distance in the first target diagnostic object set are taken as the target similar objects, and a preset number of target case objects with the smallest third target distance in the first target diagnostic object set are taken as the target non-similar objects.

[0105] Optionally, the first determining module 502 is configured to:

[0106] Obtaining a first preset initial feature weight corresponding to each condition feature;

[0107] Traversing each condition feature of each of the pending target diagnosis objects in the second target diagnosis object set, performing the following weight adjustment steps to obtain the first preset initial feature weight after each of the condition features is updated, and using the first preset initial feature weight after each of the condition features is updated as the first feature weight corresponding to the condition feature;

[0108] The weight adjustment step includes:

[0109] When the first average distance is smaller than the second average distance, the first preset initial feature weight of the current condition feature is increased by a first preset weight threshold; when the first average distance is larger than the second average distance, the first preset initial feature weight of the current condition feature is reduced by a second preset weight threshold; when the first average distance is equal to the second average distance, the first preset initial feature weight of the current condition feature is maintained unchanged.

[0110] The above technical scheme can effectively obtain the first feature weight corresponding to each condition feature, that is, describe the importance of each condition feature for determining the target case object or non-target case object from a single factor perspective, and can provide reliable data preparation for subsequent feature screening.

[0111] Optionally, the first determining module 502 is configured to:

[0112] Acquire a preset initial model, the preset initial model including a second preset initial feature weight for each condition feature;

[0113] The specified feature data of the multiple target diagnosis objects are used as training data, and the preset initial model is iteratively trained to obtain a target judgment model, which includes the second feature weight corresponding to each condition feature.

[0114] Optionally, the first determining module 502 is configured to:

[0115] Using a specified number of target case objects among the multiple target diagnosis objects as reference objects;

[0116] Obtaining a target difference between each condition feature in each target diagnosis object among the multiple target diagnosis objects except the reference object and the condition feature in the reference object;

[0117] Determine the fusion feature corresponding to each condition feature according to the target difference of each condition feature;

[0118] The preset initial model is iteratively trained according to the fusion feature corresponding to each condition feature and the preset loss function to obtain the target judgment model.

[0119] Optionally, the first determining module 502 is configured to:

[0120] Inputting the fusion feature corresponding to each condition feature into the preset initial model to obtain an objective function value;

[0121] Determine the current loss value through a preset loss function according to the objective function value;

[0122] Determine whether the current loss value is less than or equal to a preset loss value threshold;

[0123] When the current loss value is greater than the preset loss value threshold, the current second preset initial feature weight is updated according to the gradient descent method to obtain the updated preset initial model, and the step of inputting the fusion feature corresponding to each situation feature into the preset initial model to obtain the objective function value to determine whether the current loss value is less than or equal to the preset loss value threshold is executed repeatedly, until the current loss value is less than or equal to the preset loss value threshold, and the current second preset initial feature weight of each situation feature is used as the second feature weight corresponding to the situation feature.

[0124] Optionally, the first determining module 502 is configured to:

[0125] Get the preset activation function;

[0126] Inputting the target difference value of each condition feature into the preset activation function to obtain an activation function value corresponding to each target difference value;

[0127] Obtaining targets and values ​​of the activation function values ​​corresponding to the condition features of the target diagnosis objects;

[0128] The target and value of the situation feature are used as the fusion feature corresponding to the situation feature.

[0129] The above technical scheme can effectively obtain the second feature weight corresponding to each condition feature, that is, describe the importance of each condition feature for determining the target case object or non-target case object from the perspective of multi-factor fusion, and provide reliable data preparation for subsequent feature screening.

[0130] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0131] Figure 6 FIG. 1 is a block diagram of an electronic device according to an exemplary embodiment. Figure 6As shown, the electronic device 700 may include: a processor 701 and a memory 702. The electronic device 700 may also include one or more of a multimedia component 703, an input / output (I / O) interface 704, and a communication component 705.

[0132] Among them, the processor 701 is used to control the overall operation of the electronic device 700 to complete all or part of the steps in the above-mentioned feature screening method. The memory 702 is used to store various types of data to support the operation of the electronic device 700, and these data may include instructions for any application or method used to operate on the electronic device 700, and application-related data, such as contact data, messages sent and received, pictures, audio, video, etc. The memory 702 can be implemented by any type of volatile or non-volatile storage object or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read-Only Memory, referred to as EPROM), programmable read-only memory (Programmable Read-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic memory, flash memory, disk or optical disk. The multimedia component 703 may include a screen and an audio component. The screen may be, for example, a touch screen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone for receiving external audio signals. The received audio signal may be further stored in the memory 702 or sent via the communication component 705. The audio component also includes at least one speaker for outputting audio signals. The I / O interface 704 provides an interface between the processor 701 and other interface modules, and the other interface modules may be keyboards, mice, buttons, etc. These buttons may be virtual buttons or physical buttons. The communication component 705 is used for wired or wireless communication between the electronic device 700 and other objects. Wireless communication, such as Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G or 4G, or a combination of one or more thereof, so the corresponding communication component 705 may include: Wi-Fi module, Bluetooth module, NFC module.

[0133] In an exemplary embodiment, the electronic device 700 can be implemented by one or more application specific integrated circuits (ASIC), digital signal processors (DSP), digital signal processing devices (DSPD), programmable logic devices (PLD), field programmable gate arrays (FPGA), controllers, microcontrollers, microprocessors or other electronic components to execute the above-mentioned feature screening method.

[0134] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided, and when the program instructions are executed by a processor, the steps of the above-mentioned feature screening method are implemented. For example, the computer-readable storage medium can be the above-mentioned memory 702 including program instructions, and the above-mentioned program instructions can be executed by the processor 701 of the electronic device 700 to complete the above-mentioned feature screening method.

[0135] Figure 7 1 is a block diagram of another electronic device according to an exemplary embodiment. For example, the electronic device 1900 may be provided as a server. Figure 7 The electronic device 1900 includes a processor 1922, which may be one or more, and a memory 1932 for storing a computer program executable by the processor 1922. The computer program stored in the memory 1932 may include one or more modules each corresponding to a set of instructions. In addition, the processor 1922 may be configured to execute the computer program to perform the above-mentioned feature screening method.

[0136] In addition, the electronic device 1900 may further include a power supply component 1926 and a communication component 1950. The power supply component 1926 may be configured to perform power management of the electronic device 1900, and the communication component 1950 may be configured to implement communication of the electronic device 1900, for example, wired or wireless communication. In addition, the electronic device 1900 may further include an input / output (I / O) interface 1958. The electronic device 1900 may operate based on an operating system stored in the memory 1932, such as Windows Server 2003. TM , Mac OS X TM , Unix TM , Linux TM etc.

[0137] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided, and when the program instructions are executed by a processor, the steps of the above-mentioned feature screening method are implemented. For example, the computer-readable storage medium can be the above-mentioned memory 1932 including program instructions, and the above-mentioned program instructions can be executed by the processor 1922 of the electronic device 1900 to complete the above-mentioned feature screening method.

[0138] The preferred embodiments of the present disclosure are described in detail above in conjunction with the accompanying drawings. However, the present disclosure is not limited to the specific details of the above embodiments. Within the technical concept of the present disclosure, the technical solution of the present disclosure can be subjected to a variety of simple modifications, and these simple modifications all belong to the protection scope of the present disclosure.

[0139] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present disclosure will not further describe various possible combinations.

[0140] In addition, various embodiments of the present disclosure may be arbitrarily combined, and as long as they do not violate the concept of the present disclosure, they should also be regarded as the contents disclosed by the present disclosure.

Claims

1. A feature screening method, characterized in that: The method comprises: Acquire designated feature data of each of the target diagnostic objects among a plurality of target diagnostic objects, the designated feature data including a plurality of condition features and category labeling data corresponding to the target diagnostic object, the category labeling data being used to label the target diagnostic object as a target case object or a non-target case object; According to the designated feature data of the multiple target diagnostic objects, determine the first feature weight and the second feature weight corresponding to each of the condition features, wherein the first feature weight describes the importance of the condition feature for determining that the target diagnostic object is the target case object or the non-target case object from the perspective of single-factor features; the second feature weight describes the importance of the condition feature for determining that the target diagnostic object is the target case object or the non-target case object from the perspective of multi-dimensional feature fusion; Determine a target feature weight of the condition feature according to the first feature weight and the second feature weight corresponding to each condition feature, wherein the target feature weight is used to characterize the importance of the condition feature for determining that the target diagnosis object is the target case object or the non-target case object; A target situation feature is determined from the plurality of situation features according to the target feature weight corresponding to each situation feature.

2. The method according to claim 1, characterized in that: Determining a first feature weight corresponding to each of the condition features according to the specified feature data of the multiple target diagnostic objects includes: Obtaining the Euclidean distance between a specified target diagnosis object among the plurality of target diagnosis objects and other target diagnosis objects among the plurality of target diagnosis objects except the specified target diagnosis object; The target diagnostic object whose Euclidean distance with the other target diagnostic objects is less than a first preset distance threshold is taken as an element in a first target diagnostic object set corresponding to the designated target diagnostic object; In the case where it is determined that the first target diagnosis object set includes a plurality of target case objects and a plurality of non-target cases, the designated target diagnosis object corresponding to the first target diagnosis object set is used as a pending target diagnosis object to obtain a second target diagnosis object set consisting of a plurality of the pending target diagnosis objects; Obtaining target similar objects and target non-similar objects corresponding to each of the pending target diagnostic objects in the first target diagnostic object set; Obtaining a first average distance between each of the condition characteristics of each of the pending target diagnosis objects and the condition characteristics of the target similar object corresponding to the pending target diagnosis object, and a second average distance between each of the condition characteristics of each of the pending target diagnosis objects and the condition characteristics of the target non-similar object corresponding to the pending target diagnosis object; The first feature weight corresponding to each condition feature of each of the pending target diagnosis objects in the second target diagnosis object set is determined in sequence according to the first average distance and the second average distance corresponding to each condition feature.

3. The method according to claim 2, characterized in that The obtaining of target similar objects and target non-similar objects corresponding to each of the pending target diagnostic objects in the first target diagnostic object set includes: In the case where the target diagnosis object to be determined is a target case object, a first target distance between the target diagnosis object to be determined and a target case object in the first target diagnosis object set is obtained, and a second target distance between the target diagnosis object to be determined and a non-target case object in the first target diagnosis object set is obtained, and a preset number of target case objects with the smallest first target distance in the first target diagnosis object set are taken as the target similar objects, and a preset number of non-target case objects with the smallest second target distance in the first target diagnosis object set are taken as the target non-similar objects; In the case that the target diagnosis object to be determined is a non-target case object, a third target distance between the target diagnosis object to be determined and a target case object in the first target diagnosis object set, and a fourth target distance between the target diagnosis object to be determined and a non-target case object in the first target diagnosis object set are obtained, and a preset number of non-target case objects with the smallest fourth target distance in the first target diagnosis object set are taken as the target similar objects, and a preset number of target case objects with the smallest third target distance in the first target diagnosis object set are taken as the target non-similar objects.

4. The method according to claim 2, characterized in that: The determining, in sequence according to the first average distance and the second average distance corresponding to each condition feature of each of the pending target diagnosis objects in the second target diagnosis object set, a first feature weight corresponding to each of the condition features includes: Obtaining a first preset initial feature weight corresponding to each of the status features; Traversing each condition feature of each of the pending target diagnosis objects in the second target diagnosis object set, performing the following weight adjustment steps to obtain the first preset initial feature weight after each of the condition features is updated, and using the first preset initial feature weight after each of the condition features is updated as the first feature weight corresponding to the condition feature; Wherein, the weight adjustment step includes: When the first average distance is less than the second average distance, the first preset initial feature weight of the current situation feature is increased by a first preset weight threshold; when the first average distance is greater than the second average distance, the first preset initial feature weight of the current situation feature is reduced by a second preset weight threshold; when the first average distance is equal to the second average distance, the first preset initial feature weight of the current situation feature is maintained unchanged.

5. The method according to claim 1, characterized in that Determining a second feature weight corresponding to each of the condition features according to the specified feature data of the multiple target diagnostic objects includes: Acquire a preset initial model, wherein the preset initial model includes a second preset initial feature weight for each condition feature; The preset initial model is iteratively trained using the designated feature data of the multiple target diagnosis objects as training data to obtain a target judgment model, wherein the target judgment model includes the second feature weight corresponding to each condition feature.

6. The method according to claim 5, characterized in that The predetermined feature data of the plurality of target diagnostic objects are used as training data to iteratively train the preset initial model to obtain a target judgment model, wherein the target judgment model includes the second feature weight corresponding to each condition feature, including: Using a specified number of target case objects among the multiple target diagnosis objects as reference objects; Obtaining a target difference between each condition feature in each of the target diagnostic objects among the multiple target diagnostic objects except the reference object and the condition feature in the reference object; Determine the fusion feature corresponding to each of the status features according to the target difference value of the status feature; The preset initial model is iteratively trained according to the fusion features corresponding to each situation feature and the preset loss function to obtain the target judgment model.

7. The method according to claim 6, characterized in that The iterative training of the preset initial model according to the fusion feature corresponding to each condition feature and the preset loss function to obtain the target judgment model includes: Inputting the fusion feature corresponding to each of the condition features into the preset initial model to obtain an objective function value; Determine the current loss value through a preset loss function according to the objective function value; Determine whether the current loss value is less than or equal to a preset loss value threshold; When the current loss value is greater than the preset loss value threshold, the current second preset initial feature weight is updated according to the gradient descent method to obtain the updated preset initial model, and the step of inputting the fusion feature corresponding to each of the situation features into the preset initial model to obtain the objective function value to determine whether the current loss value is less than or equal to the preset loss value threshold is executed repeatedly, until the current loss value is less than or equal to the preset loss value threshold, the current second preset initial feature weight of each of the situation features is used as the second feature weight corresponding to the situation feature.

8. The method according to claim 6, characterized in that The determining, according to the target difference of each of the situation features, a fusion feature corresponding to the situation feature comprises: Get the preset activation function; Inputting the target difference value of each of the condition characteristics into the preset activation function to obtain the activation function value corresponding to each of the target difference values; Obtaining targets and values ​​of the activation function values ​​corresponding to the condition features of a plurality of the target diagnostic objects; The target and value of the situation feature are used as the fusion feature corresponding to the situation feature.

9. A feature screening device, characterized in that: The device comprises: an acquisition module, configured to acquire designated feature data of each of a plurality of target diagnosis objects, wherein the designated feature data includes a plurality of condition features, and category labeling data corresponding to the target diagnosis object, wherein the category labeling data is used to label the target diagnosis object as a target case object or a non-target case object; A first determination module is configured to determine a first feature weight and a second feature weight corresponding to each of the condition features according to the specified feature data of the multiple target diagnosis objects, wherein the first feature weight describes the importance of the condition feature for determining that the target diagnosis object is the target case object or the non-target case object from the perspective of a single factor feature; the second feature weight describes the importance of the condition feature for determining that the target diagnosis object is the target case object or the non-target case object from the perspective of a multi-dimensional feature fusion; A second determination module is configured to determine a target feature weight of the condition feature according to the first feature weight and the second feature weight corresponding to each condition feature, wherein the target feature weight is used to characterize the importance of the condition feature for determining that the target diagnosis object is the target case object or the non-target case object; The third determination module is configured to determine a target situation feature from the plurality of situation features according to the target feature weight corresponding to each situation feature.

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

11. An electronic device, characterized in that: include: a memory having a computer program stored thereon; A processor, configured to execute the computer program in the memory to implement the steps of the method according to any one of claims 1 to 8.

Images