Service binding method and apparatus for machine learning tasks

Abstract

The application discloses a service binding method and device for a machine learning task. The service binding method comprises the following steps: acquiring service information data generated by each service capability operator in an operation process of the machine learning task, wherein the machine learning task comprises multiple service stages, and each service stage comprises multiple service capability operators; determining a first health state of each service capability operator according to the service information data generated by each service capability operator and according to a pre-set first state checking rule; determining a second health state of each service stage according to the first health state of each service capability operator and according to a pre-set second state checking rule; and triggering at least one of an alarm operation and a service operation of a corresponding service stage according to the second health state of at least one service stage in the multiple service stages and according to a pre-set operation triggering rule. The service binding method can realize service binding of each stage of the machine learning task.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the field of machine learning, and more particularly, to a service binding method and apparatus for a machine learning task. BACKGROUND

[0002] Machine learning operations (MLOps) defines the whole life cycle of machine learning, from data pipeline to model training, from model training to model serving, and from model training to model monitoring. Each part is now mostly independent or fragmented. Due to the fragmentation of each part and the connection between each part, once a problem occurs in a certain part, it brings great challenges to tasks such as optimizing the model, iterating the model, and troubleshooting. SUMMARY

[0003] The present disclosure provides a service binding method and apparatus for a machine learning task, which at least solves the above problems, and can also not solve the above problems.

[0004] According to an aspect of the present disclosure, a service binding method for a machine learning task is provided, the service binding method comprising: obtaining service information data generated by each service capability operator in an operation process of the machine learning task, wherein the machine learning task comprises a plurality of service stages, and each service stage comprises a plurality of service capability operators; determining a first health state of each service capability operator based on the service information data generated by each service capability operator and according to a pre-set first state checking rule; determining a second health state of each service stage based on the first health state of each service capability operator and according to a pre-set second state checking rule; and triggering at least one of an alarm operation and a service operation of a corresponding service stage according to a pre-set operation triggering rule based on the second health state of at least one service stage of the plurality of service stages.

[0005] Optionally, the service information data comprises at least one of standard index data and business index data, wherein the standard index data represents index data commonly used in various machine learning tasks, and the business index data represents index data used in a specific machine learning task.

[0006] Optionally, the second state checking rule comprises a default checking rule and a backup checking rule, and the determining the second health state of each service stage according to the preset second state checking rule comprises: determining the second health state of each service stage according to the default checking rule; switching the second state checking rule from the default checking rule to the backup checking rule when the default checking rule is invalid, and determining the second health state of each service stage according to the backup checking rule.

[0007] Optionally, the first health state and the second health state comprise at least one of a running state, a success state, a failure state and a warning state, wherein the running state represents a state that the current node is running, the success state represents a state that the current node has completed running, the failure state represents a state that the current node fails to run, and the warning state represents a state that the current node abnormally runs, and the current node comprises at least one of a current service capability operator and a current service stage.

[0008] Optionally, the service binding method further comprises: representing the running state as a first symbol, representing the success state as a second symbol, representing the failure state as a third symbol, and representing the warning state as a fourth symbol through a user interface, so as to display the first health state of each service capability operator and the second health state of each service stage.

[0009] Optionally, the triggering at least one of the alarm operation and the service operation of the corresponding service stage according to the preset operation triggering rule based on the second health state of at least one of the plurality of service stages comprises: triggering at least one of the alarm operation and the service operation of the corresponding service stage according to the preset operation triggering rule based on the second health state of at least one of the plurality of service stages being the warning state or the failure state.

[0010] Optionally, the service stage of the machine learning task comprises at least one of a data workflow stage, a model training stage, a model serving stage and a model monitoring stage.

[0011] Optionally, the service capability operator of the model monitoring stage comprises a traffic replication operator, wherein the traffic replication operator is used to capture traffic features of online real-time traffic data of the model serving stage.

[0012] Optionally, the service capability operator of the model monitoring stage further comprises at least one of a data drift operator and a model drift operator, wherein the data drift operator is used to determine whether data drift occurs through the traffic features, and the model drift operator is used to determine whether model drift occurs through the traffic features.

[0013] Optionally, triggering at least one of the alarm operation and the service operation of the corresponding service stage according to the pre-set operation triggering rule based on the second health status of at least one of the plurality of service stages comprises: when the second health status of the model monitoring stage is a failure state and the data drift operator determines that data drift occurs, triggering at least one of the alarm operation and the A / B traffic distribution operation of the model service stage.

[0014] Optionally, triggering at least one of the alarm operation and the service operation of the corresponding service stage according to the pre-set operation triggering rule based on the second health status of at least one of the plurality of service stages comprises: when the second health status of the model monitoring stage is a failure state and the model drift operator determines that model drift occurs, triggering at least one of the alarm operation and the model retraining operation of the model training stage.

[0015] According to another aspect of the present disclosure, a service binding apparatus for a machine learning task is provided, the service binding apparatus comprising: a data acquisition unit configured to acquire service information data generated by each service capability operator in an operation process of the machine learning task, wherein the machine learning task comprises a plurality of service stages, and each service stage comprises a plurality of service capability operators; a first checking unit configured to determine a first health status of each service capability operator according to a pre-set first state checking rule based on the service information data generated by each service capability operator; a second checking unit configured to determine a second health status of each service stage according to a pre-set second state checking rule based on the first health status of each service capability operator; and an operation triggering unit configured to trigger at least one of an alarm operation and a service operation of a corresponding service stage according to a pre-set operation triggering rule based on the second health status of at least one of the plurality of service stages.

[0016] Optionally, the service information data comprises at least one of standard indicator data and business indicator data, wherein the standard indicator data represents indicator data commonly used in various machine learning tasks, and the business indicator data represents indicator data used in a specific machine learning task.

[0017] Optionally, the second state checking rule comprises a default checking rule and a backup checking rule, wherein the second checking unit is configured to determine the second health status of each service stage according to the default checking rule; when the default checking rule is invalid, switching the second state checking rule from the default checking rule to the backup checking rule, and determining the second health status of each service stage according to the backup checking rule.

[0018] Optionally, the first health status and the second health status comprise at least one of a running status, a success status, a failure status, and a warning status, wherein the running status represents a status that the current node is running, the success status represents a status that the current node has completed running, the failure status represents a status that the current node fails to run, and the warning status represents a status that the current node abnormally runs, and wherein the current node comprises at least one of a current service capability operator and a current service stage.

[0019] Optionally, the service binding apparatus further comprises a status display unit configured to display the first health status of each service capability operator and the second health status of each service stage through a user interface by representing the running status as a first symbol, representing the success status as a second symbol, representing the failure status as a third symbol, and representing the warning status as a fourth symbol.

[0020] Optionally, the operation triggering unit is configured to trigger at least one of an alarm operation and a service operation of a corresponding service stage according to a pre-set operation triggering rule based on the second health status of at least one of the plurality of service stages being the warning status or the failure status.

[0021] Optionally, the service stage of the machine learning task comprises at least one of a data workflow stage, a model training stage, a model serving stage, and a model monitoring stage.

[0022] Optionally, the service capability operator of the model monitoring stage comprises a traffic replication operator, wherein the traffic replication operator is configured to capture a traffic feature of online real-time traffic data of the model serving stage.

[0023] Optionally, the service capability operator of the model monitoring stage further comprises at least one of a data drift operator and a model drift operator, wherein the data drift operator is configured to determine whether data drift occurs based on the traffic feature, and the model drift operator is configured to determine whether model drift occurs based on the traffic feature.

[0024] Optionally, the operation triggering unit is configured to trigger at least one of an alarm operation and an A / B traffic distribution operation of the model serving stage when the second health status of the model monitoring stage is the failure status and the data drift operator determines that data drift occurs.

[0025] Optionally, the operation triggering unit is configured to trigger at least one of an alarm operation and a model retraining operation of the model training stage when the second health status of the model monitoring stage is the failure status and the model drift operator determines that model drift occurs.

[0026] According to another aspect of the present disclosure, there is provided a computer-readable storage medium storing instructions, wherein the instructions, when executed by at least one computing device, cause the at least one computing device to perform the service binding method for a machine learning task as described above.

[0027] According to another aspect of the present disclosure, there is provided a system including at least one computing device and at least one storage device storing instructions, wherein the instructions, when executed by the at least one computing device, cause the at least one computing device to perform the service binding method for a machine learning task as described above.

[0028] The service binding method and apparatus for a machine learning task according to the exemplary embodiments of the present disclosure can define each part of a machine learning task as a stage by using the Operator technology provided by Kubernetes, and connect data processing, model training, model service, and model monitoring in a whole link by combining with Event Driver, so as to realize service binding of each stage on the basis of having information of a whole flow of a machine learning task, and bring convenience to tasks such as optimizing a model, iterating a model, and troubleshooting.

[0029] Additional aspects and / or advantages of the general inventive concept will be set forth in part in the description which follows, and in part will be obvious from the description, or can be learned by practice of the general inventive concept. BRIEF DESCRIPTION OF DRAWINGS

[0030] These and / or other aspects and advantages of the present disclosure will become apparent and more readily appreciated from the following description, taken in conjunction with the accompanying drawings, in which:

[0031] Figure 1 FIG. 1 is a flowchart illustrating a service binding method for a machine learning task according to an exemplary embodiment of the present disclosure;

[0032] Figure 2 FIG. 2 is a user interface demonstration diagram illustrating a service binding method for a machine learning task according to an exemplary embodiment of the present disclosure;

[0033] Figure 3 FIG. 3 is a block diagram illustrating a service binding apparatus for a machine learning task according to an exemplary embodiment of the present disclosure. DETAILED DESCRIPTION

[0034] The following description with reference to the accompanying drawings is provided to assist in a comprehensive understanding of embodiments of the application as defined by the claims and their equivalents. Various specific details are included to assist in understanding but are not intended to limit the application. Therefore, one of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the application. In addition, descriptions of well-known functions and constructions are omitted for clarity and conciseness.

[0035] It should be noted that "at least one of a plurality of items" appearing in the present disclosure means that three types of alternatives are included: "any one of the plurality of items", "a combination of any two or more of the plurality of items", and "all of the plurality of items". For example, "including at least one of A and B" includes three alternatives: (1) including A alone, (2) including B alone, and (3) including both A and B. Also, for example, "performing at least one of step 1 and step 2" includes three alternatives: (1) performing step 1 alone, (2) performing step 2 alone, and (3) performing both step 1 and step 2.

[0036] The service binding method and apparatus for machine learning tasks according to the exemplary embodiments of the present disclosure can define each part of the machine learning task as a stage by the Operator technology provided by Kubernetes, and connect the data workflow, model training, model service, and model monitoring full link together in combination with the Event Driver, thereby having information of the whole process of the machine learning task, and bringing convenience to the tasks of optimizing the model, iterating the model, and troubleshooting problems. Here, the data workflow can represent an automated pipeline processing service for data.

[0037] Reference will now be made to Figures 1 to 3 The service binding method and apparatus for machine learning tasks according to the exemplary embodiments of the present disclosure are described in detail.

[0038] Figure 1 is a flowchart illustrating a service binding method for machine learning tasks according to the exemplary embodiments of the present disclosure. The service binding method for machine learning tasks according to the exemplary embodiments of the present disclosure can be implemented in a computing device having sufficient computing power.

[0039] Reference will now be made to Figure 1 In step S101, service information data generated in the operation process of each service capability operator of the machine learning task can be acquired. Here, the machine learning task can include a plurality of service stages, and each service stage can include a plurality of service capability operators. Further, the number of service stages and the number of service capability operators can be determined by those skilled in the art according to the actual situation of the current machine learning task.

[0040] According to exemplary embodiments of the present disclosure, the service information data can include at least one of standard indicator data and business indicator data. Here, the standard indicator data can represent indicator data commonly used in various machine learning tasks, and the business indicator data can represent indicator data used in a specific machine learning task. Further, the specific machine learning task can be a regression task, a classification task, a clustering task, or a dimension reduction task, etc., and the standard indicator data and the business indicator data can be configured by those skilled in the art according to a unified standard specification.

[0041] Next, in step S102, a first health state of each service capability actor can be determined based on the service information data generated by each service capability actor, according to a pre-set first state checking rule. As an example, the service capability actor can record service information data of the entire calculation process, and after the calculation is completed, the service information data can include data indicating whether the calculation is completed normally, and then the first health state of the corresponding service capability actor can be determined based on the service information data and according to the pre-set first state checking rule. Here, the first state checking rule can be set by those skilled in the art according to actual conditions.

[0042] Next, in step S103, a second health state of each service stage can be determined based on the first health state of each service capability actor, according to a pre-set second state checking rule. As an example, as described above, each service stage can include multiple service capability actors, and for a certain service stage, the calculation failure or abnormal calculation process of certain service capability actors does not affect the overall state of the service stage, in other words, the calculation failure or abnormal calculation process of certain service capability actors does not affect the result, so the second state checking rule can be configured to exclude the influence of the first health state of these service capability actors on the second health state of the service stage to which they belong.

[0043] According to exemplary embodiments of the present disclosure, the second state checking rule can include a default checking rule and a backup checking rule. On this basis, the second health state of each service stage can be determined according to the default checking rule, and when the default checking rule fails, the second state checking rule can be switched from the default checking rule to the backup checking rule, and the second health state of each service stage can be determined according to the backup checking rule. Here, the default checking rule and the backup checking rule can be set by those skilled in the art according to actual conditions.

[0044] According to an example embodiment of the present disclosure, the first health status and the second health status can include at least one of a running status, a success status, a failure status, and a warning status. Here, the running status can represent a status that the current node is running, the success status can represent a status that the current node has completed running, the failure status can represent a status that the current node fails to run, and the warning status can represent a status that the current node runs abnormally (e.g., high load or monitoring index deviation, etc.). Further, the current node can include at least one of a current service capability operator and a current service stage. As an example, for the first health status, the current node can be the current service capability operator; and for the second health status, the current node can be the current service stage.

[0045] According to an example embodiment of the present disclosure, the running status can be represented as a first symbol, the success status can be represented as a second symbol, the failure status can be represented as a third symbol, and the warning status can be represented as a fourth symbol through a user interface (UI), so as to display the first health status of each service capability operator and the second health status of each service stage. Here, different types of second health status can be distinguished through the first symbol, the second symbol, the third symbol, and the fourth symbol.

[0046] Next, in step S104, at least one of an alarm operation and a service operation of a corresponding service stage can be triggered according to a pre-set operation triggering rule based on the second health status of at least one service stage of the plurality of service stages. Here, at least one of the alarm operation and the service operation of the corresponding service stage can be triggered according to the pre-set operation triggering rule based on that the second health status of at least one service stage of the plurality of service stages is the warning status or the failure status. Further, an alarm information can be sent to an operation and maintenance personnel of the machine learning task through triggering the alarm operation, so that the operation and maintenance personnel can timely manually intervene in the problem, and the problem can also be automatically processed through triggering the service operation of the corresponding service stage. Still further, the service operation can include a model retraining operation or an A / B traffic distribution operation, but is not limited thereto, and the specific operation content of the service operation can be set by those skilled in the art according to the actual configuration of each service stage.

[0047] According to exemplary embodiments of the present disclosure, the service stage of the machine learning task can include at least one of a data workflow stage, a model training stage, a model serving stage, and a model monitoring stage. Here, the service capability operator of the model monitoring stage can include a traffic replication operator. Further, the traffic replication operator can be used to capture traffic features of online real-time traffic data of the model serving stage. As an example, the service capability operator of the model monitoring stage can also include at least one of a data drift operator and a model drift operator. Here, the data drift operator can be used to determine whether data drift occurs by traffic features, and the model drift operator can be used to determine whether model drift occurs by traffic features. Further, data drift can represent a case where statistical properties of data change over time, and model drift can represent a case where a model becomes unstable over time, resulting in a prediction result that is constantly wrong. Further, the way in which the data drift operator determines whether data drift occurs and the way in which the model drift operator determines whether model drift occurs can be set by those skilled in the art according to the specific model trained for the current machine learning task.

[0048] According to exemplary embodiments of the present disclosure, when the second health status of the model monitoring stage is a failure state, and the data drift operator determines that data drift occurs, at least one of an alarm operation and an A / B traffic distribution operation of the model serving stage can be triggered.

[0049] According to exemplary embodiments of the present disclosure, when the second health status of the model monitoring stage is a failure state, and the model drift operator determines that model drift occurs, at least one of an alarm operation and a model retraining operation of the model training stage is triggered. Referring to Figure 2 The user interface display effect of the service binding method of the machine learning task according to exemplary embodiments of the present disclosure is described in detail.

[0050] Figure 2 is a user interface demonstration diagram showing the service binding method of the machine learning task according to exemplary embodiments of the present disclosure.

[0051] Referring to Figure 2As an example, the full flow of the machine learning task in the user interface is illustrated to include four service stages of data, model, service and monitoring as shown in the figure, each service capability operator of the respective service stage records service information data of the entire computing process, and the service information data is counted by the Operator technology provided by Kubernetes, so as to determine the health status of each service capability operator and each service stage. Specifically, the symbols corresponding to the data stage and the model stage in the figure represent the success state, the symbol corresponding to the service stage represents the warning state, and the symbol corresponding to the monitoring stage represents the failure state. The overall operation of the machine learning task can be determined by the second health status of each service stage. In addition, the monitoring stage includes a model service operator, a traffic replication operator, a feature extraction operator, a feature store operator, a data drift operator and a model drift operator. After the model service operator is selected by the mouse, detailed information (such as service information data) of the operator can also be displayed. The first health status of each service capability operator described above is a success state, and based on the service capability operators described above, the data features of the online real-time traffic data of the service stage can be captured, and then it is calculated whether data drift or model drift occurs. When drift occurs, an alarm or model retraining operation can be triggered according to the operation triggering rule.

[0052] Figure 3 is a block diagram illustrating a service binding apparatus of a machine learning task according to an example embodiment of the present disclosure. The service binding apparatus of the machine learning task according to the example embodiment of the present disclosure can be implemented in a computing apparatus with sufficient computing power.

[0053] Referring to Figure 3 The service binding apparatus 300 of the machine learning task according to the example embodiment of the present disclosure can include a data acquisition unit 301, a first checking unit 302, a second checking unit 303 and an operation triggering unit 304.

[0054] The data acquisition unit 301 can acquire service information data generated by each service capability operator of the machine learning task in the computing process. As described above, the machine learning task can include a plurality of service stages, and each service stage can include a plurality of service capability operators.

[0055] According to the example embodiment of the present disclosure, the service information data can include at least one of standard indicator data and business indicator data. Here, the standard indicator data can represent indicator data commonly used in various machine learning tasks, and the business indicator data can represent indicator data used in a specific machine learning task.

[0056] The first checking unit 302 can determine a first health state of each service capability operator based on the service information data generated by each service capability operator according to a pre-set first state checking rule.

[0057] The second checking unit 303 can determine a second health state of each service stage based on the first health state of each service capability operator according to a pre-set second state checking rule.

[0058] According to an example embodiment of the present disclosure, the second state checking rule can include a default checking rule and a backup checking rule. On this basis, the second checking unit 303 can determine the second health state of each service stage according to the default checking rule, switch the second state checking rule from the default checking rule to the backup checking rule when the default checking rule is invalid, and determine the second health state of each service stage according to the backup checking rule.

[0059] According to an example embodiment of the present disclosure, the first health state and the second health state can include at least one of a running state, a success state, a failure state and a warning state. As described above, the running state can represent a state that the current node is running, the success state can represent a state that the current node has completed running, the failure state can represent a state that the current node fails to run, and the warning state can represent a state that the current node abnormally runs. Here, the current node can include at least one of a current service capability operator and a current service stage.

[0060] According to an example embodiment of the present disclosure, the service binding apparatus 300 can further include a state display unit. The state display unit can display the first health state of each service capability operator and the second health state of each service stage through a user interface by representing the running state as a first symbol, representing the success state as a second symbol, representing the failure state as a third symbol, and representing the warning state as a fourth symbol.

[0061] The operation triggering unit 304 can trigger at least one of an alarm operation and a service operation of a corresponding service stage according to a pre-set operation triggering rule based on the second health state of at least one service stage of the plurality of service stages.

[0062] According to an example embodiment of the present disclosure, the operation triggering unit 304 can trigger at least one of an alarm operation and a service operation of a corresponding service stage according to a pre-set operation triggering rule based on the second health state of at least one service stage of the plurality of service stages being a warning state or a failure state.

[0063] According to an example embodiment of the present disclosure, the service stage of the machine learning task can include at least one of a data processing stage, a model training stage, a model serving stage, and a model monitoring stage. As described above, the service capability operator of the model monitoring stage can include a traffic replication operator. Here, the traffic replication operator can be used to capture traffic features of online real-time traffic data of the model serving stage. Further, the service capability operator of the model monitoring stage can also include at least one of a data drift operator and a model drift operator. Still further, the data drift operator can be used to determine whether data drift occurs by the traffic features, and the model drift operator can be used to determine whether model drift occurs by the traffic features.

[0064] According to an example embodiment of the present disclosure, when the second health status of the model monitoring stage is a failure status, and the data drift operator determines that data drift occurs, the operation triggering unit 304 can trigger at least one of an alarm operation and an A / B traffic distribution operation of the model serving stage.

[0065] According to an example embodiment of the present disclosure, when the second health status of the model monitoring stage is a failure status, and the model drift operator determines that model drift occurs, the operation triggering unit 304 can trigger at least one of an alarm operation and a model retraining operation of the model training stage.

[0066] The service binding method and apparatus of the machine learning task according to an example embodiment of the present disclosure can define each part of the machine learning task as a stage by the Operator technology provided by Kubernetes, and connect the data processing, the model training, the model serving, and the model monitoring full link by combining the Event Driver, so as to realize the service binding of each stage on the basis of having the machine learning task full flow information, and bring convenience to the tasks such as optimizing the model, iterating the model, and troubleshooting problems.

[0067] The service binding method and apparatus of the machine learning task according to an example embodiment of the present disclosure are described above with reference to Figures 1 to 3 The service binding method and apparatus of the machine learning task according to an example embodiment of the present disclosure are described above with reference to

[0068] Figure 3 Each unit in the service binding apparatus of the machine learning task shown can be configured as software, hardware, firmware, or any combination of the above, which performs a certain function. For example, each unit can correspond to a dedicated integrated circuit, can also correspond to pure software code, and can also correspond to a module combined by software and hardware. In addition, one or more functions implemented by each unit can also be uniformly executed by components in a physical entity device (for example, a processor, a client, or a server, etc.).

[0069] In addition, the service binding method and apparatus of the machine learning task according to an example embodiment of the present disclosure are described above with reference to Figure 1The service binding method of the machine learning task described can be implemented by a program (or instructions) recorded on a computer readable storage medium. For example, according to an exemplary embodiment of the present disclosure, a computer readable storage medium storing instructions can be provided, wherein when the instructions are run by at least one computing device, the at least one computing device is caused to perform the service binding method of the machine learning task according to the present disclosure.

[0070] The computer program in the computer readable storage medium described above can be run in an environment deployed in a computer device such as a client, a host, a proxy device, a server, etc. It should be noted that the computer program can also be used to perform additional steps other than the steps described above or perform more specific processing when performing the steps described above, and the content of the additional steps and further processing has been described above with reference to the Figure 1 During the description of the related method, the steps mentioned above are not repeated here.

[0071] It should be noted that each unit in the service binding device of the machine learning task according to an exemplary embodiment of the present disclosure can be completely implemented by running a computer program to realize the corresponding function, i.e., each unit corresponds to each step in the functional architecture of the computer program, so that the entire system is called by a special software package (e.g., lib library) to realize the corresponding function.

[0072] On the other hand, Figure 3 Each unit shown can also be implemented by hardware, software, firmware, middleware, microcode or any combination thereof. When implemented by software, firmware, middleware or microcode, the program code or code segment for performing the corresponding operation can be stored in a computer readable medium such as a storage medium, so that the processor can perform the corresponding operation by reading and running the corresponding program code or code segment.

[0073] For example, the exemplary embodiments of the present disclosure can also be implemented as a computing device including a storage component and a processor, the storage component storing a set of computer executable instructions, when the set of computer executable instructions is executed by the processor, performing the service binding method of the machine learning task according to the exemplary embodiments of the present disclosure.

[0074] Specifically, the computing device can be deployed in a server or a client, or on a node device in a distributed network environment. In addition, the computing device can be a PC computer, a tablet device, a personal digital assistant, a smart phone, a web application or other devices capable of executing the above-mentioned instruction set.

[0075] Here, the computing device need not be a single computing device, but can be a collection of devices or circuits that individually or jointly execute the instructions (or sets of instructions) described above. The computing device can also be part of an integrated control system or system manager, or be configured as a portable electronic device that interfaces with a local or remote (e.g., via wireless transmission) interface.

[0076] In the computing device, the processor can include a central processing unit (CPU), a graphics processing unit (GPU), a programmable logic device, a dedicated processor system, a microcontroller, or a microprocessor. By way of example and not limitation, the processor can also include an analog processor, a digital processor, a microprocessor, a multi-core processor, a processor array, a network processor, etc.

[0077] Certain operations described in the service binding method of a machine learning task according to the exemplary embodiments of the present disclosure can be implemented by software, certain operations can be implemented by hardware, and in addition, these operations can be implemented by a combination of software and hardware.

[0078] The processor can execute instructions or codes stored in one of the storage components, where the storage components can also store data. The instructions and data can also be sent and received over a network via the network interface device, which can employ any known transmission protocol.

[0079] The storage components can be integrated with the processor, for example, RAM or flash memory disposed within an integrated circuit microprocessor, etc. In addition, the storage components can include separate devices, such as external disk drives, storage arrays, or other storage devices usable by any database system. The storage components and the processor can be operatively coupled or can communicate with each other, for example, through I / O ports, network connections, etc., so that the processor can read files stored in the storage components.

[0080] In addition, the computing device can also include a video display (such as a liquid crystal display) and a user interaction interface (such as a keyboard, a mouse, a touch input device, etc.). All components of the computing device can be connected to each other via a bus and / or a network.

[0081] The service binding method of a machine learning task according to the exemplary embodiments of the present disclosure can be described as various interconnected or coupled functional blocks or functional diagrams. However, these functional blocks or functional diagrams can be equally integrated as a single logic device or operate with non-exact boundaries.

[0082] Therefore, with reference to Figure 1 The described service binding method of a machine learning task can be implemented by a system including at least one computing device and at least one storage device storing instructions.

[0083] According to an exemplary embodiment of the present disclosure, the at least one computing device is a computing device for performing a service binding method for machine learning tasks according to an exemplary embodiment of the present disclosure, and a set of computer executable instructions is stored in the storage device, and when the set of computer executable instructions is executed by the at least one computing device, the service binding method for machine learning tasks described with reference to Figure 1 is performed.

[0084] The above describes various exemplary embodiments of the present disclosure, and it should be understood that the above description is only exemplary and is not exhaustive, and the present disclosure is not limited to the disclosed exemplary embodiments. Many modifications and changes are obvious to those skilled in the art without departing from the scope and spirit of the present disclosure. Therefore, the protection scope of the present disclosure should be subject to the scope of claims.

Claims

1. A service binding method for a machine learning task, wherein, The service binding method includes: The service information data generated by each service capability operator of the machine learning task during the operation process is obtained. The machine learning task includes multiple service stages, each service stage includes multiple service capability operators, the multiple service stages include a model service stage and a model monitoring stage, the service capability operator of the model monitoring stage includes a traffic replication operator, and the traffic replication operator is used to capture the traffic characteristics of the online real-time traffic data of the model service stage. For each service capability operator, based on the service information data generated by the current service capability operator, the first health status of the current service capability operator is determined according to the pre-set first state check rules; For each service phase, based on the first health status of each service capability operator in the current service phase, the second health status of the current service phase is determined according to the pre-set second status check rules. Based on the second health status of at least one of the multiple service stages, an alarm operation and at least one of the corresponding service operations are triggered according to a pre-set operation triggering rule.

2. The service binding method as described in claim 1, wherein, The service information data includes at least one of standard indicator data and business indicator data, wherein the standard indicator data represents indicator data that is common to various machine learning tasks, and the business indicator data represents indicator data that is used in specific machine learning tasks.

3. The service binding method as described in claim 1, wherein, The second status check rule includes a default check rule and a backup check rule, wherein determining the second health status of the current service phase according to the preset second status check rule includes: The second health status of the current service phase is determined according to the default inspection rules. When the default check rule fails, the second status check rule is switched from the default check rule to the backup check rule, and the second health status of the current service stage is determined according to the backup check rule.

4. The service binding method as described in claim 1, wherein, The first health state and the second health state include at least one of the following: running state, success state, failure state, and warning state. The running state indicates that the current node is running, the success state indicates that the current node has completed its operation, the failure state indicates that the current node has failed to run, and the warning state indicates that the current node is running abnormally. The current node corresponding to the first health state is the current service capability operator, and the current node corresponding to the second health state is the current service stage.

5. The service binding method as described in claim 4, wherein, The service binding method also includes: The user interface represents the running status as a first symbol, the success status as a second symbol, the failure status as a third symbol, and the warning status as a fourth symbol, thereby displaying the first health status of each service capability operator and the second health status of each service stage.

6. The service binding method as described in claim 4, wherein, Based on the second health status of at least one of the multiple service stages, trigger an alarm operation and at least one of the corresponding service operations according to pre-set operation triggering rules, including: If the second health status of at least one of the multiple service stages is a warning state or a failure state, an alarm operation and at least one of the corresponding service operations of the service stage are triggered according to the pre-set operation triggering rules.

7. The service binding method as described in claim 4, wherein, The service phase of the machine learning task also includes at least one of the data workflow phase and the model training phase.

8. The service binding method as described in claim 7, wherein, The service capability operator in the model monitoring phase further includes at least one of a data drift operator and a model drift operator, wherein the data drift operator is used to determine whether data drift has occurred based on the traffic characteristics, and the model drift operator is used to determine whether model drift has occurred based on the traffic characteristics.

9. The service binding method as described in claim 8, wherein, Based on the second health status of at least one of the multiple service stages, trigger an alarm operation and at least one of the corresponding service operations according to pre-set operation triggering rules, including: When the second health state of the model monitoring phase is a failure state, and the data drift operator determines that data drift has occurred, at least one of the following is triggered: an alarm operation and an A / B traffic allocation operation of the model service phase.

10. The service binding method as described in claim 8, wherein, Based on the second health status of at least one of the multiple service stages, trigger an alarm operation and at least one of the corresponding service operations according to pre-set operation triggering rules, including: When the second health state of the model monitoring phase is a failure state, and the model drift operator determines that model drift has occurred, at least one of the following is triggered: an alarm operation and a model retraining operation in the model training phase.

11. A service binding device for a machine learning task, wherein, The service binding device includes: The data acquisition unit is configured to: acquire service information data generated by each service capability operator of the machine learning task during the operation process, wherein the machine learning task includes multiple service stages, each service stage includes multiple service capability operators, the multiple service stages include a model service stage and a model monitoring stage, the service capability operator of the model monitoring stage includes a traffic replication operator, and the traffic replication operator is used to capture the traffic characteristics of the online real-time traffic data of the model service stage; The first inspection unit is configured to: for each service capability operator, based on the service information data generated by the current service capability operator, determine the first health status of the current service capability operator according to the pre-set first state inspection rules; The second inspection unit is configured to: for each service stage, based on the first health status of each service capability operator in the current service stage, determine the second health status of the current service stage according to the pre-set second status inspection rules; The operation triggering unit is configured to: trigger at least one of the alarm operation and the service operation of the corresponding service stage according to the second health status of at least one of the plurality of service stages, based on the pre-set operation triggering rules.

12. The service binding device as claimed in claim 11, wherein, The service information data includes at least one of standard indicator data and business indicator data, wherein the standard indicator data represents indicator data that is common to various machine learning tasks, and the business indicator data represents indicator data that is used in specific machine learning tasks.

13. The service binding apparatus as claimed in claim 11, wherein, The second status check rule includes a default check rule and a backup check rule, wherein the second check unit is configured as follows: The second health status of the current service phase is determined according to the default inspection rules. When the default check rule fails, the second status check rule is switched from the default check rule to the backup check rule, and the second health status of the current service stage is determined according to the backup check rule.

14. The service binding apparatus as claimed in claim 11, wherein, The first health state and the second health state include at least one of the following: running state, success state, failure state, and warning state. The running state indicates that the current node is running, the success state indicates that the current node has completed its operation, the failure state indicates that the current node has failed to run, and the warning state indicates that the current node is running abnormally. The current node corresponding to the first health state is the current service capability operator, and the current node corresponding to the second health state is the current service stage.

15. The service binding apparatus as claimed in claim 14, wherein, The service binding device also includes: The status display unit is configured to: represent the running status as a first symbol, the success status as a second symbol, the failure status as a third symbol, and the warning status as a fourth symbol through a user interface, thereby displaying the first health status of each service capability operator and the second health status of each service stage.

16. The service binding apparatus of claim 14, wherein, The operation triggering unit is configured as follows: If the second health status of at least one of the multiple service stages is a warning state or a failure state, an alarm operation and at least one of the corresponding service operations of the service stage are triggered according to the pre-set operation triggering rules.

17. The service binding apparatus of claim 14, wherein, The service phase of the machine learning task also includes at least one of the data workflow phase and the model training phase.

18. The service binding apparatus of claim 17, wherein, The service capability operator in the model monitoring phase further includes at least one of a data drift operator and a model drift operator, wherein the data drift operator is used to determine whether data drift has occurred based on the traffic characteristics, and the model drift operator is used to determine whether model drift has occurred based on the traffic characteristics.

19. The service binding apparatus as claimed in claim 18, wherein, The operation triggering unit is configured as follows: When the second health state of the model monitoring phase is a failure state, and the data drift operator determines that data drift has occurred, at least one of the following is triggered: an alarm operation and an A / B traffic allocation operation of the model service phase.

20. The service binding device as claimed in claim 18, wherein, The operation triggering unit is configured as follows: When the second health state of the model monitoring phase is a failure state, and the model drift operator determines that model drift has occurred, at least one of the following is triggered: an alarm operation and a model retraining operation in the model training phase.

21. A computer-readable storage medium for storing instructions, wherein, When the instruction is executed by at least one computing device, it causes the at least one computing device to perform the service binding method for the machine learning task as described in any one of claims 1 to 10.

22. A system comprising at least one computing device and at least one storage device for storing instructions, wherein, When the instruction is executed by the at least one computing device, it causes the at least one computing device to perform the service binding method for the machine learning task as described in any one of claims 1 to 10.

Images