Algorithm arrangement method and related device

Abstract

The invention discloses an algorithm arrangement method and a related device, and the algorithm arrangement method comprises the steps: receiving a plurality of nodes selected by a user from a node library, and receiving a node sequence set by the user for the plurality of nodes, the plurality of nodes comprise a plurality of general algorithm nodes and a plurality of scene adaptive nodes, a scene adaptation node is associated with at least one general algorithm node; according to the node sequence, the multiple nodes are combined to form a target algorithm, the target algorithm is used for data classification of a target scene, in the data classification process, the universal algorithm nodes are used for data classification of the to-be-classified data, and a first classification result is obtained; and the scene adaptation node is used for correcting the first classification result obtained by the associated general algorithm node to obtain a second classification result. According to the scheme, the data classification effect of the target algorithm in the target scene can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, in particular to an algorithm arrangement method and related device. BACKGROUND

[0002] A single data classification algorithm can only realize a single-dimensional data classification function. In order to realize a multi-dimensional data classification function, a complex data classification algorithm needs to be constructed. One implementation way of constructing a complex data classification algorithm is algorithm arrangement.

[0003] The algorithm arrangement method in the related art can be described as follows: a node library is constructed in advance, and the node library provides reusable algorithm nodes; a plurality of algorithm nodes selected by a user from the node library are received, and a node order set by the user for the plurality of algorithm nodes is received, and the plurality of algorithm nodes are combined according to the node order to obtain a complex data classification algorithm.

[0004] However, the complex algorithm obtained by the algorithm arrangement method in the related art has insufficient adaptability to actual application scenarios, and therefore the data classification effect (recall rate, accuracy rate, etc.) when applied to actual application scenarios is not good. SUMMARY

[0005] The present application provides an algorithm arrangement method and related device, which can solve the problem that the data classification effect of the complex algorithm obtained by the algorithm arrangement method in the related art is not good in actual application scenarios.

[0006] The present application provides an algorithm arrangement method, which includes: receiving a plurality of nodes selected by a user from a node library, and receiving a node order set by the user for the plurality of nodes, wherein the plurality of nodes include a plurality of general algorithm nodes and a plurality of scene adaptation nodes, and one scene adaptation node is associated with at least one general algorithm node; combining the plurality of nodes according to the node order to form a target algorithm, and the target algorithm is used for data classification in a target scene, in the data classification process, the general algorithm node is used for data classification on to-be-classified data to obtain a first classification result, and the scene adaptation node is used for correcting the first classification result obtained by its associated general algorithm node to obtain a second classification result.

[0007] The application provides an algorithm arrangement device, comprising a receiving module and a combination module. The receiving module is configured to receive a plurality of nodes selected by a user from a node library and receive a node order set by the user for the plurality of nodes, wherein the plurality of nodes comprise a plurality of general algorithm nodes and a plurality of scene adaptation nodes, and one scene adaptation node is associated with at least one general algorithm node. The combination module is configured to combine the plurality of nodes to form a target algorithm according to the node order, and the target algorithm is used for data classification of a target scene. In the data classification process, the general algorithm node is used for data classification of to-be-classified data to obtain a first classification result, and the scene adaptation node is used for correcting the first classification result obtained by the associated general algorithm node to obtain a second classification result.

[0008] The application provides an electronic device comprising a memory and a processor, wherein the processor is configured to execute program instructions stored in the memory to implement the above method.

[0009] The application provides a computer readable storage medium, wherein program instructions are stored on the computer readable storage medium, and the program instructions are executed by a processor to implement the above method.

[0010] The application provides a computer program product comprising a computer program, wherein the computer program is executed by a processor to implement the above method.

[0011] The above scheme adds selectable scene adaptation nodes in the node library. In the algorithm arrangement process, the user selects a plurality of nodes including a plurality of general algorithm nodes and at least one scene adaptation node associated with the general algorithm node from the node library, and receives a node order of the plurality of nodes. The target algorithm is combined according to the node order. The scene adaptation node is adapted to the target scene, and in the data classification process of the target scene, the scene adaptation node is used for correcting the first classification result of the associated general algorithm node to obtain the second classification result. Thus, on the one hand, the target algorithm can be quickly generated for the target scene. On the other hand, the correction of the first classification result by the scene adaptation node with a higher adaptation degree to the target scene can avoid the negative influence of the data classification effect of the general algorithm node in the target scene on the overall data classification effect, thereby improving the adaptation degree of the target algorithm to the target scene and further improving the overall data classification effect. On the other hand, the manufacturer only needs to add the scene adaptation node with simple processing logic in the node library, so the development cost is lower and the development period is shorter compared with the development of an algorithm node for data classification of the target scene, and the difficulty is lower compared with the optimization of the general algorithm node.

[0012] It should be understood that the foregoing general description and the following detailed description are only exemplary and explanatory, rather than limiting the application. BRIEF DESCRIPTION OF DRAWINGS

[0013] The drawings incorporated in and forming a part of the specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the application.

[0014] Figure 1 is a flowchart of an embodiment of the algorithm arrangement method provided by the present application; Figure 2 is a flowchart of another embodiment of the algorithm arrangement method provided by the present application; Figure 3 is a flowchart of a specific example of the algorithm arrangement method of the present application; Figure 4 is a structural diagram of an algorithm arrangement system provided by the present application; Figure 5 is a structural diagram of an embodiment of an algorithm arrangement device provided by the present application; Figure 6 is a structural diagram of an embodiment of an electronic device provided by the present application; Figure 7 is a structural diagram of an embodiment of a computer readable storage medium provided by the present application. DETAILED DESCRIPTION

[0015] The schemes of the embodiments of the present application will be described in detail below with reference to the drawings.

[0016] In the following description, specific details are set forth in order to provide a thorough understanding of the present application. The present application can be practiced without resorting to the details specific.

[0017] The term "and / or" herein merely describes an associated relationship between associated objects, which means that there can be three relationships, for example, A and / or B can mean that A exists alone, A and B exist together, and B exists alone. In addition, the character " / " herein generally represents an "or" relationship between the front and rear associated objects. In addition, "multiple" herein means two or more than two. In addition, the term "at least one" herein means any combination of any one or more of the plurality or at least two of the plurality, for example, including at least one of A, B and C can mean including any one or more elements selected from the set consisting of A, B and C. In addition, the term "several" herein means any integer greater than 0, for example, 1, 2, 3, 4, 5, ….

[0018] A single data classification algorithm can only realize a single-dimensional data classification function. In order to realize a multi-dimensional data classification function, a complex data classification algorithm needs to be constructed. One implementation way of constructing a complex data classification algorithm is algorithm arrangement.

[0019] The algorithm composition refers to that a node library is constructed in advance, reusable algorithm nodes are provided in the node library, the algorithm nodes can be regarded as algorithm atoms, a user selects a plurality of algorithm nodes from the node library, sets node orders of the selected algorithm nodes, and combines the selected algorithm nodes according to the node orders to obtain a composite algorithm.

[0020] The inventors of the present application have found through long-term research that the composite algorithm obtained by the algorithm composition method in the related art has insufficient adaptability to actual application scenarios, and therefore the data classification effect (recall rate, accuracy rate, etc.) when used in actual application scenarios is not good, and it is difficult to meet the requirements in actual application scenarios.

[0021] The inventors of the present application have further found that the reason why the composite algorithm has insufficient adaptability to actual application scenarios is that the algorithm nodes in the node library provided by the manufacturer are general algorithm nodes trained by using general data sets, and the training sample data in the general data sets is not training sample data obtained in actual application scenarios, so the composite algorithm composed of the general algorithm nodes has insufficient adaptability to actual application scenarios.

[0022] In order to improve the adaptability of the composite algorithm to actual application scenarios, one solution is that the manufacturer optimizes the general algorithm nodes for actual application scenarios, and the optimization of the general algorithm nodes needs to fine-tune the general algorithm nodes by using training sample data obtained in actual application scenarios, but the manufacturer cannot obtain the training sample data due to its security and privacy, so the solution has great implementation difficulty. Another solution is that the manufacturer develops algorithm nodes adapted to actual application scenarios from scratch, which has a long cycle and high economic cost.

[0023] Therefore, the present application proposes a new algorithm composition method. The embodiments thereof are introduced as follows: Figure 1 is a flowchart of an embodiment of the algorithm composition method provided by the present application. As shown in Figure 1 the algorithm composition method can include the following steps in the embodiment: S110: receiving a plurality of nodes selected by a user from a node library, and receiving node orders set by the user for the plurality of nodes.

[0024] The plurality of nodes include a plurality of general algorithm nodes and a plurality of scene adaptation nodes, and one scene adaptation node is associated with at least one general algorithm node.

[0025] The execution subject of the embodiment is an algorithm composition device, which can be any electronic device with algorithm composition capability.

[0026] The several general algorithm nodes can be one general algorithm node or multiple general algorithm nodes. Different general algorithm nodes can implement data classification functions of different dimensions. For example, the general algorithm node 1 is used for classifying whether the action of data is normative, and the general algorithm node 2 is used for classifying whether the dressing of data is normative.

[0027] The general algorithm node associated with one scene adaptation node can be one or multiple. The scene adaptation node associated with one general algorithm node is one.

[0028] In some embodiments, the user can select nodes and set node orders on the visualization canvas by clicking, dragging, and the like.

[0029] S120: Combining the plurality of nodes in the node order to form a target algorithm.

[0030] The target algorithm is used for data classification of a target scene. In the data classification process, the general algorithm node is used for data classification of the data to be classified to obtain a first classification result, and the scene adaptation node is used for correcting the first classification result obtained by the general algorithm node associated therewith to obtain a second classification result.

[0031] The data to be classified can be images, videos, texts, and the like. The target scene is an actual application scene. The scene adaptation node associated with the general algorithm node has a higher degree of adaptation to the target scene. The general algorithm node is used for data classification of the data to be classified, and the scene adaptation node is used for correcting the first classification result of the general algorithm node associated therewith to obtain a second classification result having a higher degree of adaptation to the target scene. Since the scene adaptation node corrects the first classification result rather than classifies data, the processing logic of the scene adaptation node is simpler than that of the general algorithm node associated therewith.

[0032] The above scheme adds selectable scene adaptation nodes to the node library. In the algorithm arrangement process, the user selects multiple nodes from the node library, including several general algorithm nodes and at least one scene adaptation node associated with the general algorithm nodes, and receives the node order of the multiple nodes. The target algorithm is combined according to the node order. The scene adaptation node adapts to the target scene, and is used to correct the first classification result of the associated general algorithm node to obtain a second classification result in the data classification process of the target scene. Thus, on the one hand, the target algorithm can be quickly generated for the target scene. On the other hand, the correction of the first classification result by the scene adaptation node with a higher adaptation degree to the target scene can avoid the negative impact of the poor data classification effect of the general algorithm node in the target scene on the overall data classification effect, thereby improving the adaptation degree of the target algorithm to the target scene and further improving the overall data classification effect. On the other hand, the manufacturer only needs to add scene adaptation nodes with simple processing logic to the node library, so the development cost is lower and the development cycle is shorter compared with the development of algorithm nodes for data classification of the target scene, and the difficulty is lower compared with the optimization of general algorithm nodes.

[0033] In some embodiments, the general algorithm node is a neural network large model. The scene adaptation node is a lightweight neural network micro model in the target scene. The neural network micro model is more lightweight in structure than the neural network large model, and requires fewer training sample data and shorter training time.

[0034] In some embodiments, for the scene adaptation node, in addition to supporting the user to select the scene adaptation node to compose the target algorithm, the node library also supports the user to upload the training sample data of the target scene to at least one of train and fine-tune the scene adaptation node, so that the adaptation degree of the scene adaptation node to the target scene reaches the expectation, further improves the adaptation degree of the scene adaptation node to the target scene, enables the scene adaptation node to have the correction ability of the first classification result, and the like. The training of the scene adaptation node can be performed before being selected to compose the target algorithm. The fine-tuning of the scene adaptation node can be performed before being selected to compose the target algorithm, or can be performed after being selected to compose the target algorithm, but in the case that the trial target algorithm finds that the data classification effect does not meet the preset condition (such as the first preset condition and the second preset condition mentioned later).

[0035] In some embodiments, for the scene adaptation node, the node library also supports the user to upload trial sample data, training sample data, and the like, and supports the user to label the training sample data, such as the training sample data in the training phase / fine-tuning phase of the scene adaptation node, supporting the user to label the real second classification result, and the training sample data in the fine-tuning phase of the general algorithm phase, supporting the user to label the real first classification result. For detailed description of the training phase and the fine-tuning phase, please refer to the embodiments later.

[0036] In some embodiments, the node library also supports, for the scene adaptation stage, user deletion of original scene adaptation nodes, addition of custom scene adaptation nodes, modification of the structure of cause scene adaptation nodes, and the like.

[0037] In some embodiments, the input of the scene adaptation node includes the first classification result and correction reference information. The correction reference information includes at least one of the to-be-classified data, the features of the to-be-classified data, and the classification-related region in the to-be-classified data. The features of the to-be-classified data and the classification-related region in the to-be-classified data are obtained by the general algorithm node in the process of obtaining the first classification result.

[0038] For example, the to-be-classified data is a traffic scene image, the general algorithm node is a universal classification model, and the associated scene adaptation node is a classification correction model for traffic scenes. After inputting the traffic scene image into the universal classification model, the process of obtaining the first classification result by the universal classification model is as follows: feature extraction is performed on the traffic scene image to obtain the features of the traffic scene image, and based on the features of the traffic scene image, the classification-related region in the traffic scene image and the first classification result corresponding to the classification-related region are obtained. Then, the classification-related region in the traffic scene image and the first classification result are input into the classification correction model to obtain the second classification result.

[0039] For another example, the to-be-classified data is a biological scene image, the general algorithm node is a universal classification model, and the associated scene adaptation node is a classification correction model for biological scenes. After inputting the biological scene image into the universal classification model, the process of obtaining the first classification result by the universal classification model is as follows: feature extraction is performed on the biological scene image to obtain the features of the biological scene image, and based on the features of the biological scene image, the first classification result is obtained. Then, the features of the biological scene image and the first classification result are input into the classification correction model to obtain the second classification result.

[0040] In some embodiments, the process of obtaining the second classification result by the scene adaptation node is as follows: it is determined whether the first classification result is correct in combination with the correction reference information; in response to the first classification result being correct, the first classification result is correct is taken as the second classification result; and in response to the first classification result being incorrect, the first classification result is incorrect is taken as the second classification result.

[0041] For example, the first classification result is that the classification-related region in the traffic scene image belongs to the abnormal traffic behavior category. In the case where the scene adaptation node determines that the first classification result is correct, the second classification result is the first classification result correct, i.e., indicating that the classification-related region in the traffic scene image belongs to the abnormal traffic behavior category. In the case where the scene adaptation node determines that the first classification result is incorrect, the second classification result is the first classification result incorrect, i.e., indicating that the classification-related region in the traffic scene image does not belong to the abnormal traffic behavior category.

[0042] For example, the first classification result is that the biological scene image belongs to the cat category. In a case where the scene adaptation node determines that the first classification result is correct, the second classification result is that the biological scene image belongs to the cat category. In a case where the scene adaptation node determines that the first classification result is incorrect, the second classification result is that the biological scene image does not belong to the cat category.

[0043] In some embodiments, the scene adaptation node obtains the second classification result by determining whether the first classification result is correct in combination with the correction reference information; in response to the first classification result being correct, taking the first classification result as the second classification result; and in response to the first classification result being incorrect, taking the opposite result of the first classification result as the second classification result.

[0044] For example, the first classification result is that the classification-related region in the traffic scene image belongs to the abnormal traffic behavior category. In a case where the scene adaptation node determines that the first classification result is correct, the second classification result is that the classification-related region in the traffic scene image belongs to the abnormal traffic behavior category. In a case where the scene adaptation node determines that the first classification result is incorrect, the second classification result is that the classification-related region in the traffic scene image does not belong to the abnormal traffic behavior category.

[0045] For example, the first classification result is that the biological scene image belongs to the abnormal traffic behavior category. In a case where the scene adaptation node determines that the first classification result is correct, the second classification result is that the biological scene image belongs to the cat category. In a case where the scene adaptation node determines that the first classification result is incorrect, the second classification result is that the biological scene image does not belong to the cat category.

[0046] In some embodiments, in the training phase, the training process of the scene adaptation node is as follows: obtaining a training sample data set, the training sample data set including a plurality of training sample data, the training sample data including a training first classification result and corresponding training correction reference information, the training first classification result being obtained by the general algorithm node, and the training sample data being labeled with a true second classification result; inputting the training sample data into the scene adaptation node to make the scene adaptation node obtain a predicted second classification result; constructing a loss function based on the difference between the true second classification result and the predicted second classification result; adjusting the parameters of the scene adaptation node based on the loss function; and repeating the foregoing process until a training expectation is reached.

[0047] In some embodiments, the training sample data set in the training phase includes at least part of the training sample data of the target scene uploaded by the user.

[0048] In some embodiments, before S110, the method further comprises: receiving training sample data of the target scene uploaded by the user; and training the scene adaptation node by using the training sample data of the target scene. Thus, the training authority of the scene adaptation node is given to the user, so that the user can upload the training sample data of the target scene to the training node, and train the scene adaptation node by using the training sample data of the target scene uploaded by the user. Not only can the adaptation degree of the scene adaptation node to the target scene reach the expectation, but also there is no problem of data security and privacy.

[0049] Figure 2 is a flowchart of another embodiment of the algorithm arrangement method provided in the present application. As shown in Figure 2 In the present embodiment, after S120, the method further comprises the following steps: S130: verifying the data classification effect of the target algorithm by using the trial sample data of the target scene uploaded by the user.

[0050] The trial sample data is trial classification data under the target scene. The trial classification data is similar to the data to be classified.

[0051] In some embodiments, the data classification effect can be measured by accuracy, recall rate and the like.

[0052] S140: optimizing the target algorithm in response to the data classification effect not satisfying the first preset condition.

[0053] In some embodiments, the first preset condition comprises that the accuracy is greater than a first accuracy threshold, and the recall rate is greater than a first recall rate threshold. For example, the first accuracy threshold is 80%, and the first recall rate threshold is 90%.

[0054] The above scheme gives the trial authority of the target algorithm to the user, so that the user can verify the data classification effect of the target algorithm by using the trial sample data of the target scene uploaded by the user before applying the target algorithm to the data classification of the target scene, so as to optimize the target algorithm in the case that the first preset condition is not satisfied. Thus, not only the effect verification for the target scene can be completed, but also there is no problem of data security and privacy.

[0055] In some embodiments, S140 comprises S141. S141: optimizing the target algorithm by using a first optimization manner. The first optimization manner is fine-tuning the scene adaptation node by using the training sample data of the target scene uploaded by the user.

[0056] The fine-tuning stage of the scene adaptation stage is performed on the basis of the training stage, and the scene adaptation node can be further optimized by fine-tuning. The processing process of the scene adaptation node in the fine-tuning stage is similar to the processing process in the training stage. Details are not described herein.

[0057] The above scheme gives the user the fine-tuning right of the scene adaptation node, so that the user can upload the training sample data of the target scene in the fine-tuning stage and fine-tune the scene adaptation node through the training sample data, thereby not only improving the adaptation degree of the scene adaptation node to the target scene and realizing the optimization of the target algorithm, but also not existing data security and privacy problems. In addition, compared with the general algorithm node, the training sample data required for fine-tuning the scene adaptation node is less, the complexity is lower, and the cycle is shorter.

[0058] In some embodiments, S140 includes S142. S142: optimizing the target algorithm by using a second optimization manner, the second optimization manner including at least one of node modification on the target algorithm and classification prompt word modification of modifying the general algorithm node.

[0059] The node modification on the target algorithm can be adding a node in the target algorithm or replacing a node in the target algorithm. The addition or replacement can be for all types of nodes in the target algorithm, such as the general algorithm node, the scene adaptation node, and the logic conversion node. For details of the logic conversion node, please refer to the following embodiments, which are not repeated here.

[0060] The replacement of the general algorithm node can be replacing the general algorithm node with a scene algorithm node or replacing the general algorithm node with another general algorithm node that can achieve the same data classification function. The scene algorithm node is an algorithm node adapted to the target scene. For example, the general algorithm node can classify any target object, and the target scene is to classify vehicles. First, determine whether there is a scene algorithm node for classifying vehicles. If there is, replace the general algorithm node with a scene algorithm node for classifying vehicles. If not, replace the general algorithm node with another general algorithm node.

[0061] The classification prompt word is used to provide prior knowledge related to data classification, clarify data classification targets, data classification boundaries, etc. for the general algorithm node, and help the general algorithm node obtain a better first classification result.

[0062] The above scheme can improve the data classification effect of the general algorithm node.

[0063] In some embodiments, S142 includes: in response to a node modification operation of the user, replacing at least one node in the target algorithm.

[0064] In some embodiments, S142 includes: in response to a prompt word modification operation of the user, modifying the classification prompt word of at least one general algorithm node.

[0065] In some embodiments, the step of optimizing the target algorithm in the first optimization manner is performed when it is confirmed that the training sample data exists, and the step of optimizing the target algorithm in the second optimization manner is performed when it is confirmed that the training sample data does not exist.

[0066] It can be understood that the first optimization manner is more effective than the second optimization manner. The above scheme adopts a double-layer optimization strategy, that is, the first optimization manner is used preferentially, and the second optimization manner is used when the user does not have training sample data, so as to maximize the effectiveness of optimization while ensuring normal optimization.

[0067] In some embodiments, S140 comprises: optimizing the target algorithm in a third optimization manner, the third optimization manner being fine-tuning the general algorithm node by using the training sample data of the target scene uploaded by the user.

[0068] In the fine-tuning stage of the general algorithm node, the training sample data is similar to the data to be classified and is labeled with a real first classification result. The fine-tuning process of the general algorithm node comprises: extracting features of the training sample data by using the general algorithm node; obtaining a predicted first classification result based on the features of the training sample data by using the general algorithm node; constructing a loss function based on the difference between the predicted first classification result and the real first classification result; adjusting the parameters of the general algorithm node based on the loss function; and repeating the foregoing process until the fine-tuning expectation is reached.

[0069] In some embodiments, after S140, there further comprises: re-verifying the data classification effect of the target algorithm by using the trial sample data of the target scene uploaded by the user. In addition, in response to the re-verified data classification effect not satisfying the first preset condition, the optimization of the target algorithm is continued until the data classification effect satisfies the first preset condition.

[0070] In some embodiments, after S140, there further comprises S150-S160.

[0071] S150: applying the optimized target algorithm to data classification of the target scene.

[0072] S160: in response to the data classification effect of the optimized target algorithm not satisfying a second preset condition, optimizing the optimized target algorithm again.

[0073] The optimization manner of the optimized target algorithm in the application stage is similar to the optimization manner of the target algorithm in the trial stage. For example, the optimization manner is that, in a case where it is confirmed that the user has new training sample data, the first optimization manner is adopted to optimize the optimized target algorithm based on the new training sample data again, or in a case where it is confirmed that the user has no new training sample data, the second optimization manner is adopted to optimize the optimized target algorithm again.

[0074] The second preset condition is stricter than the first preset condition. In some embodiments, the first preset condition includes that the accuracy is greater than a second accuracy threshold, the recall rate is greater than a second recall rate threshold, and the like. The second accuracy threshold is greater than the first accuracy threshold, and the second recall rate threshold is greater than the first recall rate threshold. For example, the second accuracy threshold is 90%, and the second recall rate threshold is 95%.

[0075] The above scheme can further optimize the optimized target algorithm again by using the first optimization manner or the second optimization manner if the user wants a better data classification effect in the application process of the optimized target algorithm.

[0076] In some embodiments, before S110, the method further includes: in response to an algorithm node creation request of the user, displaying an algorithm node creation interface to the user; receiving a custom general algorithm node created by the user in the algorithm node creation interface; and saving the custom general algorithm node in the node library.

[0077] The above scheme supports the user to create a custom general algorithm node according to the actual needs of the user.

[0078] In some embodiments, the plurality of nodes further include a plurality of logical conversion nodes, and the logical conversion nodes are configured to perform logical conversion on the at least one first classification result and / or the at least one second classification result to obtain a third classification result in the data classification process.

[0079] The logical conversion node can be a combination of at least two of the logical conversion node, the logical conversion node, and the non-logical conversion node.

[0080] The logical conversion of the AND logic conversion node is that the third classification result is a specific category when and only when the multiple first / second classification results satisfy the corresponding category condition after the AND operation is performed on the multiple first / second classification results. For example, the AND logic conversion node performs logical conversion on two first classification results, one of which is that a target is detected, and the other of which is that the target has abnormal behavior, and the third classification result is that an abnormal event occurs. For another example, both of the two first classification results are that there is action abnormality in the to-be-classified data, and the third classification result is that there is action abnormality in the to-be-classified data. For another example, both of the two first classification results are the cat category, and the third classification result is the cat category.

[0081] The logical conversion of the OR logic conversion node is that the third classification result is a specific category as long as at least one of the multiple first / second classification results is the specific category after the OR operation is performed on the multiple first / second classification results. For example, the OR logic conversion node performs logical conversion on one first classification result and one second classification result, the first classification result is that there is first abnormal behavior and / or the second classification result is that there is second abnormal behavior, and the third classification result is that there is abnormal behavior.

[0082] The logical conversion of the non-logic conversion node is that the third classification result is the negation result of the corresponding first / second classification result after the negation operation is performed on the one or more first / second classification results. For example, the first classification result is that there is action abnormality, and the corresponding third classification result is that there is no action abnormality.

[0083] In some embodiments, before S110, the method further includes: in response to a node expansion request of a user, displaying a node expansion interface to the user; receiving an existing logic conversion node selected by the user in the node expansion interface; receiving a node expansion file uploaded by the user; and expanding the function of the existing logic conversion node by using the node expansion file.

[0084] The node expansion file can be a logic conversion script or other executable file. The node expansion file complies with a standard expansion format specification.

[0085] The above scheme supports the user to visually expand the function of the existing logic conversion node to adapt to the actual needs of the user.

[0086] In some embodiments, before S110, the method further includes: in response to a logic node creation request of a user, displaying a logic node creation interface to the user; receiving a custom logic conversion node created by the user in the logic node creation interface; and saving the custom logic conversion node in the node library.

[0087] The description of the user-created custom logical conversion node can be received, and a logical conversion function file of the user-created custom logical conversion node can be received. The description of the custom logical conversion node and the logical conversion function file are stored in the node library. The logical conversion function file can be a logical conversion script or other executable file. The logical conversion function file complies with a standard creation format specification.

[0088] The above scheme supports the user to visually create a custom logical conversion node, and realizes extension of the logical conversion nodes of the node library to adapt to actual needs of the user.

[0089] For ease of understanding, the algorithm arrangement method provided in the present application is described in the form of a specific example as follows.

[0090] Figure 3 is a flowchart of a specific example of the algorithm arrangement method of the present application. As shown in Figure 3 , the algorithm arrangement method comprises: I. Arranging a target algorithm.

[0091] 1. In response to an algorithm arrangement request, the node library is displayed on an algorithm arrangement interface.

[0092] 2. A plurality of nodes selected by the user from the node library are received, and a node order set by the user for the plurality of nodes is received.

[0093] The plurality of nodes comprise a plurality of general algorithm nodes, a plurality of scene adaptation nodes, and a plurality of logical conversion nodes.

[0094] Among them, a general algorithm node is used to perform one-dimensional data classification on to-be-classified data to obtain a first classification result. A scene adaptation node is used to correct the first classification result obtained by one or more general algorithm nodes associated therewith to obtain a second classification result. A logical conversion node can be a logical conversion node, or a logical conversion node, or a non-logical conversion node, etc., and is used to perform logical conversion on at least one first classification result and / or at least one second classification result.

[0095] Among them, when the node library does not provide a general algorithm node and a logical conversion node required by the user, the user is supported to customize.

[0096] 3. The plurality of nodes are combined to form a target algorithm according to the node order.

[0097] II. Trial of the target algorithm.

[0098] The trial sample data of the target scene uploaded by the user is used to verify the data classification effect of the target algorithm. A trial report about the data classification effect is generated and displayed to the user, so that the user can determine whether the data classification result meets a first preset condition according to the trial report.

[0099] In response to the data classification effect not satisfying the first preset condition, the target algorithm is optimized. In response to the first preset condition being satisfied, the target algorithm can be directly applied to data classification of the target scene.

[0100] III. Optimizing the target algorithm.

[0101] It is determined whether the user has training sample data of the target scene. In response to the training sample data of the target scene being present, the target algorithm is optimized by using a first optimization manner. In response to the training sample data of the target scene being absent, the target algorithm is optimized by using a second optimization manner. In addition, the optimized target algorithm can be tested again until the target algorithm satisfies the first preset condition.

[0102] IV. Applying the target algorithm.

[0103] The target algorithm is applied to data classification of the target scene. After being put into application, if the user wants the algorithm effect to be better, it is determined whether the data classification effect satisfies a second preset condition. In response to the data classification effect satisfying the second preset condition, the target algorithm is continuously applied. In response to the data classification effect not satisfying the second preset condition, the target algorithm is optimized again.

[0104] V. Optimizing the target algorithm again.

[0105] It is determined whether the user has new training sample data of the target scene. In response to the new training sample data of the target scene being present, the target algorithm is optimized again by using the first optimization manner. In response to the new training sample data of the target scene being absent, the target algorithm is optimized again by using the second optimization manner. In addition, the target algorithm optimized again can be tested again until the target algorithm satisfies the second preset condition.

[0106] The algorithm arrangement method provided in the application can be applied to an algorithm arrangement system.

[0107] Figure 4 is a structural schematic diagram of the algorithm arrangement system provided in the application. As shown in Figure 4 The algorithm application module, the algorithm management module, and the algorithm training module can be software modules or hardware modules. If the modules are software modules, different software modules can run on one physical machine or multiple physical machines. If the modules are hardware modules, different hardware modules can belong to one physical machine or different physical machines. Among them: The algorithm application module can be used to: obtain a target algorithm from the algorithm management module; apply the target algorithm; receive and transmit test sample data and training sample data uploaded by a user to the algorithm training module, and the like.

[0108] The algorithm management module can be configured to create a custom general algorithm node, a custom logical conversion node, receive a user operation to orchestrate a target algorithm, modify a node in the target algorithm, modify a classification prompt word of the general algorithm node, and obtain an optimized scene adaptation node from the algorithm training module.

[0109] The algorithm training module can be configured to obtain trial sample data and training sample data uploaded by a user from the algorithm application module, test a data classification effect of a target algorithm according to the trial sample data uploaded by the user, and optimize a scene adaptation node according to the training sample data uploaded by the user.

[0110] Figure 5 FIG. 1 is a structural schematic diagram of an embodiment of an algorithm orchestration apparatus provided by the present application. As shown in FIG. 1, the algorithm orchestration apparatus includes a receiving module and a combination module. Wherein: Figure 5 The receiving module is configured to receive a plurality of nodes selected by a user from a node library and receive a node order set by the user for the plurality of nodes, wherein the plurality of nodes include a plurality of general algorithm nodes and a plurality of scene adaptation nodes, and one scene adaptation node is associated with at least one general algorithm node. The combination module is configured to combine the plurality of nodes to form a target algorithm according to the node order, and the target algorithm is used for data classification of a target scene. In the data classification process, the general algorithm node is used for data classification of to-be-classified data to obtain a first classification result, and the scene adaptation node is used for correcting the first classification result obtained by its associated general algorithm node to obtain a second classification result.

[0111] For other detailed descriptions of this embodiment, please refer to the previous embodiments, which will not be described here.

[0112]

[0113] FIG. 2 is a structural schematic diagram of an embodiment of an electronic device provided by the present application. As shown in FIG. 2, the electronic device 50 includes a memory 51 and a processor 52, and the processor 52 is configured to execute program instructions stored in the memory 51 to implement the steps in any method embodiment described above. In one specific implementation scenario, the electronic device 50 can include but is not limited to a microcomputer, a server, and in addition, the electronic device 50 can also include a notebook computer, a tablet computer, and other carrying devices, which are not limited here. Figure 6 Figure 6

[0114] ​​In particular, the processor 52 is configured to control itself and the memory 51 to implement the steps in any of the above method embodiments. The processor 52 can also be referred to as a CPU (Central Processing Unit). The processor 52 can be an integrated circuit chip with processing capability. The processor 52 can also be a general purpose processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. The general purpose processor can be a microprocessor or the processor can be any conventional processor. In addition, the processor 52 can be a combination of the integrated circuit chip and the general purpose processor.

[0115] Please refer to Figure 7 , Figure 7 is a structural diagram of an embodiment of the computer readable storage medium of the present application. The computer readable storage medium 60 stores program instructions 601, and the program instructions 601 are executed by a processor to implement the steps in any of the above method embodiments.

[0116] The present application also provides a computer program product, which contains a computer program. When the computer program is executed by a processor, the steps in the method described in any of the above embodiments can be implemented. Specifically, the computer program product can be software or a program product containing a computer program, which can be run on a computing device or stored in any available medium.

[0117] In some embodiments, the apparatus provided by the present application has functions or contains modules which can be used to implement the method described in the above method embodiments. The specific implementation can refer to the description of the above method embodiments, and for the sake of brevity, it will not be repeated here.

[0118] The above description of each embodiment tends to emphasize the differences between each embodiment, and the same or similar parts can be mutually referred to. For the sake of brevity, it will not be repeated here.

[0119] In several embodiments provided in the present application, it should be understood that the disclosed methods and apparatuses can be implemented in other manners. For example, the division of the apparatus embodiments described above is merely a logical function division, and there can be another division manner in actual implementation. For example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In another image location, the displayed or discussed coupling or direct coupling or communication connection between the units can be indirect coupling or communication connection through some interfaces, and can be electrical, mechanical or in other forms.

[0120] In addition, each function unit in the various embodiments of the present application can be integrated into a processing unit, or each unit can exist physically independently, or two or more units can be integrated into one unit. The integrated unit can be implemented in the form of hardware or in the form of a software function unit. If the integrated unit is implemented in the form of a software function unit and sold or used as an independent product, it can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application, essentially or in other words, the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor (processor) to execute all or part of the steps of the various embodiments of the method of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various other media that can store program codes.

Claims

1. An algorithm orchestration method, characterized in that, include: The system receives multiple nodes selected by a user from a node library and receives the node order set by the user for the multiple nodes, wherein the multiple nodes include several general algorithm nodes and several scene adaptation nodes, and one of the scene adaptation nodes is associated with at least one of the general algorithm nodes. According to the node order, the multiple nodes are combined to form a target algorithm. The target algorithm is used for data classification in a target scene. During the data classification process, the general algorithm node is used to classify the data to be classified and obtain a first classification result. The scene adaptation node is used to correct the first classification result obtained by the associated general algorithm node and obtain a second classification result.

2. The method according to claim 1, characterized in that, After combining the multiple nodes according to the node order to form the target algorithm, the algorithm further includes: The data classification effect of the target algorithm is tested using the trial sample data of the target scenario uploaded by the user. In response to the data classification effect not meeting the first preset condition, the target algorithm is optimized.

3. The method according to claim 2, characterized in that, The optimization of the target algorithm includes: The target algorithm is optimized using a first optimization method, which involves fine-tuning the scene adaptation nodes using training sample data of the target scene uploaded by the user; or... The target algorithm is optimized using a second optimization method, which includes at least one of modifying the nodes of the target algorithm or modifying the classification prompt words of the general algorithm nodes.

4. The method according to claim 3, characterized in that, The step of optimizing the target algorithm using the first optimization method is performed when it is confirmed that the user has the training sample data; the step of optimizing the target algorithm using the second optimization method is performed when it is confirmed that the user does not have the training sample data. And / or, the plurality of nodes further includes several logical transformation nodes, and the optimization of the target algorithm using the second optimization method includes at least one of the following steps: In response to the user's node modification operation, at least one of the nodes in the target algorithm is replaced; In response to the user's prompt word modification operation, the classification prompt words of at least one of the general algorithm nodes are modified.

5. The method according to claim 2, characterized in that, After optimizing the target algorithm, the method further includes: The optimized target algorithm is applied to the data classification of the target scenario; If the data classification effect of the optimized target algorithm does not meet the second preset condition, the optimized target algorithm is optimized again.

6. The method according to claim 1, characterized in that, Before receiving the multiple nodes selected by the user from the node library and receiving the node order set by the user for the multiple nodes, the method further includes: In response to the user's algorithm node creation request, an algorithm node creation interface is displayed to the user; Receive custom generic algorithm nodes created by users in the algorithm node creation interface; The custom general algorithm node is stored in the node library.

7. The method according to claim 1, characterized in that, The plurality of nodes also includes several logical transformation nodes. During the data classification process, the logical transformation nodes are used to perform logical transformation on at least one of the first classification results and / or at least one of the second classification results to obtain a third classification result.

8. The method according to claim 7, characterized in that, Before receiving the multiple nodes selected by the user from the node library and receiving the node order set by the user for the multiple nodes, the method further includes: In response to the user's node expansion request, a node expansion interface is displayed to the user. Receive the existing logical transformation node selected by the user in the node extension interface; Receive the node extension file uploaded by the user; The existing logical transformation nodes are extended using the node extension file; And / or, In response to the user's logical node creation request, a logical node creation interface is displayed to the user. Receive custom logical transformation nodes created by the user in the logical node creation interface; The custom logic transformation node is stored in the node library.

9. An electronic device, characterized in that, It includes a memory and a processor, the processor being configured to execute program instructions stored in the memory to implement the method of any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, It stores program instructions that, when executed by a processor, implement the method of any one of claims 1-8.

11. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1-8.

Images