XGBoost-based model calling method, system and equipment
By building a model management platform and using container orchestration tools to dynamically manage the life cycle of the XGBoost model, we solved the problems of high model application risks, weak scalability, and slow deployment, and achieved stable model operation and rapid updates.
Patent Information
- Application Number
- CN202510838436.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-20
- Publication Date
- 2025-10-03
AI Technical Summary
In existing technologies, the XGBoost model faces problems in fields such as financial technology, healthcare, and elderly care services, such as high model application risks, weak scalability, and slow deployment. In particular, model calculation anomalies can easily lead to application blockage, model files are coupled in quotation components, and are difficult to expand. Each time it goes online, it must be packaged and restarted, making it impossible to quickly update and iterate.
By building a model management platform, decoupling model training units and running units, using container orchestration tools to dynamically create and destroy models, combining API gateways and service meshes for routing and load balancing, and dynamically adjusting the weight and number of Pods, the full lifecycle management of the model can be achieved.
It improves the stability and scalability of model operation, avoids application blocking caused by model calculation anomalies, enables rapid model updates and iterations, and dynamically adjusts resource configuration without restarting the application.
Smart Images
Figure CN120743512A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of model calling technology and relates to a model calling method, system and device based on XGBoost. Background Art
[0002] In the field of fintech pricing, there are cases of model calls. Taking the current motor vehicle premium rate setting and pricing as an example, the existing technologies are mostly based on the traditional generalized linear model (GLM). This type of algorithm model has the advantages of high interpretability and a wide range of usage scenarios. However, with the rapid development of machine learning related algorithms in recent years, the gradient boosting (Graadient Boosting) algorithm represented by XGBoost has demonstrated better accuracy and excellent robustness in the processing of structured data. Compared with the traditional GLM model, XGBoost model training and prediction are more complicated. Generally, the model bin file after XGBoost training needs to be embedded in the application component to load and read when applying predictions, and the application's memory and CPU need to be used for prediction calculations. If an anomaly occurs, the impact on the application is relatively large. There is currently no good application solution to reduce the risk. There are problems such as high model application risk, weak scalability, and slow deployment and online launch. The specific defects are as follows:
[0003] (1) High model risk: The model and the core quotation component are deployed in the same application. If the model calculation anomaly occurs, it can easily spread and cause the quotation application to be blocked;
[0004] (2) Weak model scalability: The model bin file is coupled with the quotation component. As a result, there are more and more subsequent model files, the code logic is bloated and difficult to expand;
[0005] (3) Slow model deployment: The current model is a model bin file that needs to be embedded in the application file package. Each time it goes online, it must be packaged and restarted, making it impossible to quickly update the iterative model.
[0006] Similarly, in all situations where model calls are required, including not only financial technology but also healthcare, elderly care services and other fields, there are problems such as high model application risks, weak scalability and slow deployment. Summary of the Invention
[0007] The present invention provides a model calling method, system and device based on XGBoost to solve the technical problems of high risk, weak scalability and slow deployment of existing models.
[0008] In a first aspect, the present invention provides a model calling method based on XGBoost, comprising the following steps:
[0009] Obtain each model trained by the model training unit and store each bin file model obtained after training;
[0010] Receive model call requests from application units and distribute the model call requests to the corresponding model running instance Pod according to predetermined routing rules and load balancing strategies;
[0011] Output model scheduling and routing results to corresponding application units.
[0012] Furthermore, the model training unit acquires each model trained, and stores each bin file model obtained after training; specifically:
[0013] Obtain the corresponding bin file models trained by each model training sub-unit of the model training unit, and distribute each bin file model to the corresponding model running instance Pod for storage; each Pod only loads and runs one bin file model.
[0014] Furthermore, the receiving of the model call request from the application unit distributes the model call request to the corresponding model running instance Pod according to a predetermined routing rule and load balancing strategy; including:
[0015] Receive model call requests from application units, and perform routing and load balancing of model call requests according to predetermined routing rules and load balancing strategies;
[0016] Routes the request to the corresponding model running instance Pod based on the requested model identifier;
[0017] Dynamically adjust routing rules based on the user's geographic location and / or time factors;
[0018] Dynamically adjust the Pod weight based on the actual load of the Pod;
[0019] Based on historical data on model call volume, predict future load conditions and adjust the weight or number of Pods in advance.
[0020] Furthermore, the model management component manages the entire life cycle of each bin file model, including registration, deployment, update, and retirement of the corresponding bin file model;
[0021] The Pod configuration of each bin file model is modified online through the model management component, and the model management component uses a container orchestration tool to realize the dynamic creation and destruction of Pod.
[0022] Furthermore, the model call request sent by the application unit is received through the API gateway.
[0023] In a second aspect, the present invention provides a model calling system based on XGBoost, comprising:
[0024] The model training unit is used to train the input XGBoost model and output the trained corresponding model to the model management platform;
[0025] The model management platform is used to store trained models and schedule and route model call requests from application units.
[0026] The application unit is used to send model call requests to the model management platform according to needs; and output after receiving feedback from the model management platform.
[0027] Furthermore, the model training unit includes several model training subunits, each of which trains the XGBoost model input thereto, generates a corresponding bin file model, and outputs the trained bin file model to the model management platform;
[0028] The application unit includes several application sub-units, each of which sends a model call request to the model management platform according to demand; and outputs after receiving feedback from the model management platform;
[0029] The model management platform is used to store the corresponding models trained by each model training sub-unit, and to perform model scheduling and routing for the model call requests sent by each application unit.
[0030] Furthermore, the model management platform includes a model management component and several model running instance Pods;
[0031] The model management component is used to receive the corresponding bin file models trained by each model training sub-unit and distribute each bin file model to the corresponding Pod; it is also used for the full life cycle management of each bin file model, including registration, deployment, update and retirement of the corresponding bin file model;
[0032] Each Pod is independently deployed in a container, and each Pod only loads and runs one bin file model.
[0033] Furthermore, the model management component includes an API gateway, which is used to receive model call requests sent by the application unit and distribute the model call requests to corresponding Pods according to routing rules and load balancing strategies;
[0034] The model management component uses a service grid or predetermined routing rules to implement routing and load balancing of model call requests;
[0035] The model management component dynamically adjusts routing rules based on the user's geographic location and / or time factors.
[0036] Furthermore, the model management component is configured to:
[0037] The Pod configuration of each bin file model can be modified online. The model management component uses the container orchestration tool to realize the dynamic creation and destruction of Pods;
[0038] Dynamically adjust the Pod weight based on the actual load of the Pod;
[0039] Based on historical data on model call volume, predict future load conditions and adjust the weight or number of Pods in advance.
[0040] In a third aspect, the present invention provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the XGBoost-based model calling method as described above when executing the computer program.
[0041] In a fourth aspect, the present invention provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the steps of the above-mentioned XGBoost-based model calling method.
[0042] The above-mentioned model calling method, system and device based on XGBoost save the trained XGBoost model into a bin file. The application component generally needs to load the model bin file, and then use the corresponding prediction method to perform model prediction during actual prediction. Before the improvement, the number of models was relatively large, which would result in a large number of model bin files that needed to be loaded for the same application, and the actual call volume of different models was different. The call volume of different models in the application was also inconsistent, but there was no way to achieve load balancing and free configuration of the model in the application. After the improvement, the present invention decouples the model training unit and the model operation unit through the model management platform, so that in the model training and calling stages, the model operation resources can be dynamically arranged and dynamically routed for different models, which greatly improves the stability and scalability of model operation and training. BRIEF DESCRIPTION OF THE DRAWINGS
[0043] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments of the present invention. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0044] Figure 1 This is a flow chart of a model calling method based on XGBoost according to an embodiment of the present invention;
[0045] Figure 2 yes Figure 1 Flowchart of step S2 in FIG.
[0046] Figure 3 1 is a schematic diagram of a model calling system based on XGBoost according to an embodiment of the present invention;
[0047] Figure 4 It is a structural diagram of a computer device provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0048] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.
[0049] The present invention is applicable to all occasions where model calling is required, including but not limited to the fields of financial technology, medical health, and elderly care services. It should be noted that the innovation of the present invention lies in the construction of a model management platform on which model scheduling and routing are performed according to the needs of the application end. The model training unit trains the XGBoost model input thereto and generates the corresponding bin file model, which is not the key point of the present invention. Building a model management platform is not available in existing model calling technologies.
[0050] like Figure 1 As shown, an embodiment of the present invention provides a model calling method based on XGBoost, comprising the following steps:
[0051] S1 obtains each model trained by the model training unit and stores each bin file model obtained after training.
[0052] Specifically:
[0053] Obtain the corresponding bin file models trained by each model training sub-unit of the model training unit, and distribute each bin file model to the corresponding model running instance Pod for storage; each Pod only loads and runs one bin file model.
[0054] S2 receives the model call request from the application unit and distributes the model call request to the corresponding model running instance Pod according to the predetermined routing rules and load balancing strategy.
[0055] like Figure 2 As shown, step S2 specifically includes:
[0056] S21 receives the model call request sent by the application unit, and performs routing and load balancing of the model call request according to the predetermined routing rules and load balancing strategy;
[0057] S22 routes the request to the corresponding model running instance Pod according to the requested model identifier;
[0058] S23 dynamically adjusts routing rules based on the user's geographic location and / or time factors;
[0059] S24 dynamically adjusts the Pod weight based on the actual load of the Pod;
[0060] S25 predicts future load conditions based on historical data on model call volume and adjusts the weight or number of Pods in advance.
[0061] It should be noted that steps S23 to S25 are performed in no particular order.
[0062] In this embodiment, the model management component is used to manage the entire life cycle of each bin file model, including registration, deployment, update, and retirement of the corresponding bin file model.
[0063] The Pod configuration of each bin file model is modified online through the model management component, and the model management component uses a container orchestration tool to realize the dynamic creation and destruction of Pod.
[0064] The model call request sent by the application unit is received through the API gateway.
[0065] S3 outputs the model scheduling and routing results to the corresponding application units.
[0066] like Figure 3 As shown, an embodiment of the present invention provides a model calling system based on XGBoost, including a model training unit, a model management platform and an application unit.
[0067] The model training unit is used to train the input XGBoost model and output the trained corresponding model to the model management platform.
[0068] The model management platform is used to store trained models and to schedule and route model call requests from application units.
[0069] The application unit is used to send model call requests to the model management platform according to demand; and output after receiving feedback from the model management platform.
[0070] Furthermore, the model training unit includes several model training sub-units, each model training sub-unit trains the XGBoost model input thereto, generates a corresponding bin file model, and outputs the trained bin file model to the model management platform.
[0071] Furthermore, the application unit includes several application sub-units, each of which sends a model call request to the model management platform according to demand; and outputs after receiving feedback from the model management platform.
[0072] Furthermore, the model management platform is used to store the corresponding models trained by each model training sub-unit, and to perform model scheduling and routing for the model call requests sent by each application unit.
[0073] Furthermore, the model management platform includes a model management component and several model running instance Pods;
[0074] The model management component is used to receive the corresponding bin file models trained by each model training sub-unit and distribute each bin file model to the corresponding Pod; it is also used for the full life cycle management of each bin file model, including registration, deployment, update and retirement of the corresponding bin file model;
[0075] Each Pod is independently deployed in a container, and each Pod only loads and runs one bin file model.
[0076] Furthermore, the model management component receives the model call request sent by the application unit and routes the request to the corresponding Pod according to the model identifier of the request;
[0077] The model management component uses a service grid or predetermined routing rules to implement routing and load balancing of model call requests;
[0078] The model management component dynamically adjusts routing rules based on the user's geographic location and / or time factors.
[0079] Furthermore, the model management component includes an API gateway, which is used to receive the model call request sent by the application unit and distribute the model call request to the corresponding Pod according to routing rules and load balancing strategies.
[0080] Furthermore, the model management component is configured to be able to modify the Pod configuration of each bin file model online, and the model management component uses a container orchestration tool to realize dynamic creation and destruction of Pod.
[0081] Furthermore, the model management component dynamically adjusts the weight of the Pod according to the actual load of the Pod.
[0082] Furthermore, the model management component predicts future load conditions based on historical data of model call volume and adjusts the weight or number of Pods in advance.
[0083] Furthermore, the container orchestration tool is Kubernetes, and the automated management of the model Pod is achieved through the Kubernetes API. The automated management includes the lifecycle management, health check, and automatic recovery of the Pod.
[0084] As an embodiment of the present invention, after each XGBoost model is trained, a corresponding model file (bin file) is generated and stored in a unified model storage system (such as a distributed file system or cloud storage). Figure 1 In the illustrated embodiment, a unified model storage system is provided within the model management platform. The model management component receives the corresponding bin file models trained by each model training subunit and stores each bin file model in the corresponding model storage system. The model management component distributes each bin file model to a corresponding Pod; each bin file model is independently deployed into a container, forming a model execution instance (Pod). Each Pod only loads and runs one bin file model, preventing multiple bin file models from competing for resources in the same container.
[0085] The model management component uses container orchestration tools (such as Kubernetes) to dynamically create and destroy Pods. Pods are dynamically created and destroyed through the Kubernetes API. For example, when the call volume for a certain bin file model increases, a new Pod is automatically created; when the call volume for a certain bin file model decreases, the redundant Pods are automatically destroyed.
[0086] The model management component uses the Kubernetes API to automate the management of bin file model Pods, including Pod lifecycle management, health checks, and automatic recovery. The model management component manages the entire lifecycle of each bin file model, including registration, deployment, updates, and retirement. The model management component regularly checks the health of Pods. If a Pod is found to be abnormal (such as crashed or unresponsive), it automatically restarts it or creates a new Pod to replace it.
[0087] The model management component is configured to modify the configuration of bin file model pods online. Based on the actual call volume of the bin file model, the model management component dynamically adjusts the number of pods or resource allocation without restarting the service. For bin file models with high call volume, the model management component automatically increases the number of pods and memory to improve processing capacity; for bin file models with low call volume, the model management component reduces the number of pods and memory to conserve resources.
[0088] In addition, the model management platform is equipped with a configuration interface to allow administrators or the system to manually adjust the number of Pods and resource allocation (such as CPU, memory, etc.) according to business needs. For example, more resources can be allocated to high-priority models to ensure their performance.
[0089] The model management platform includes several pods for model execution, each running in a separate pod. The model management component receives model call requests from application units and routes the requests to the corresponding pod based on the requested model identifier (such as the model ID or model name). The model management platform provides flexible routing rule configuration, supporting precise routing based on model identifiers as well as dynamic routing based on business logic. For example, routing rules can be dynamically adjusted based on the user's location and / or time of day.
[0090] The model management component uses a service mesh or predefined routing rules to route and load balance model call requests. The model management component includes an API gateway, which receives model call requests from application units and distributes them to the corresponding pods based on routing rules and load balancing policies. The model management component is configured to dynamically update routing rules and load balancing policies, without restarting the service.
[0091] The present invention can use a variety of load balancing algorithms, including but not limited to the following:
[0092] (1) Random call: Randomly assign requests to available Pods.
[0093] (2) Sequential call: Requests are dispatched in the order of the Pods.
[0094] (3) Weighted call: Dynamically allocate requests based on the Pod's load or configured weight.
[0095] The model management component dynamically adjusts the weight of the Pod according to the actual load of the Pod (such as CPU usage, memory usage, response time, etc.), thereby achieving more efficient load balancing.
[0096] The model management component predicts future load conditions based on historical data on the number of model calls in the bin file and adjusts the weight or number of Pods in advance.
[0097] The embodiment of the present invention can achieve that when different applications call different bin file models, they call the specified bin file model according to the specified routing rules and load balancing strategy, and the operations of different bin file models are no longer coupled, which greatly improves the stability and scalability of the model operation.
[0098] The XGBoost-based model calling system provided in the embodiment of the present invention can realize online management and configurable orchestration of bin file models, and can also dynamically load balance the routing of bin file models during runtime, as specifically embodied as follows:
[0099] (1) It can be dynamically started directly after the model training is completed, without the need to modify the application and dynamically load and restart.
[0100] (2) Different bin file models run on different instances, which realizes the isolation of different bin file model operations and calls, reduces the coupling between bin file models, and improves the stability of bin file model operations.
[0101] (3) A routing mechanism has been added to the bin file model operation call, which can provide a unified bin file model calling platform to the outside world, and then route to the specified bin file model instance that needs to be called and run according to different calling parameters or model parameters.
[0102] (4) Dynamically configured load balancing strategy can configure different calling strategies according to actual application scenarios, such as random, weighted or sequential calling, etc., to achieve the load balancing configuration of bin file model running instance calls, and also greatly increase the scalability and stability of bin file model operation.
[0103] The above-mentioned XGBoost-based model calling method is as follows: Figure 4 As shown, an embodiment of the present invention further provides a schematic diagram of the structure of a computer device, which includes a processor and a memory coupled to the processor. The memory stores a computer program, which, when executed by the processor, causes the processor to perform the steps of the XGBoost-based model calling method in the above embodiment.
[0104] For other details about how the processor in the above-mentioned computer device implements the above-mentioned technical solution, please refer to the description of the XGBoost-based model calling method provided in the above-mentioned invention embodiment, which will not be repeated here.
[0105] Among them, the processor can also be called a CPU (Central Processing Unit), and the processor may be an integrated circuit chip with signal processing capabilities; the processor can also be a general-purpose processor, DSP (Digital Signal Processor), ASIC (Application Specific Integrated Circuit), FPGA (Field Programmable Gate Array) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, among which the general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc.
[0106] An embodiment of the present invention further provides a computer-readable storage medium having a readable computer program stored thereon; wherein the computer program may be stored in the above-mentioned storage medium in the form of a software product, including a number of instructions for causing a computer device (which may be a personal computer, server, or network device, etc.) or a processor to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: a USB flash drive, a mobile hard drive, a magnetic disk or an optical disk, a ROM (Read-Only Memory), a RAM (Random Access Memory), and other media that can store program code, or a terminal device such as a computer, server, mobile phone, or tablet.
[0107] In the several embodiments provided herein, it should be understood that the disclosed system can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the module division is merely a logical function division. In actual implementation, there may be other division methods, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not implemented. In addition, the coupling or direct coupling or communication connection shown or discussed can be through some interface, indirect coupling or communication connection of devices or modules, and can be electrical, mechanical or other forms.
[0108] The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules may be selected to achieve the purpose of this embodiment according to actual needs.
[0109] In addition, the functional modules in the various embodiments of the present invention may be integrated into a single processing module, each module may exist physically separately, or two or more modules may be integrated into a single module. The integrated modules may be implemented in the form of hardware or software functional modules. If the integrated modules are implemented in the form of software functional modules and sold or used as independent products, they may be stored in a computer-readable storage medium.
[0110] In the above embodiments, all or part of the embodiments may be implemented by software, hardware, firmware, or any combination thereof. When implemented by software, all or part of the embodiments may be implemented in the form of a computer program product.
[0111] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center via a wired (e.g., coaxial cable, optical fiber, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) method. The computer-readable storage medium can be any available medium that a computer can store or a data storage device such as a server or data center that includes one or more available media integrated. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium), or a semiconductor medium (e.g., an SSD (solid state disk)).
[0112] The technical solution provided by the present invention is introduced in detail above. Specific examples are used in the present invention to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.
[0113] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, optical storage, etc.) containing computer-usable program code.
[0114] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0115] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.
[0116] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.
[0117] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.
Claims
1. The model calling method based on XGBoost is characterized by: The following steps are involved: Obtain each model trained by the model training unit and store each bin file model obtained after training; Receive model call requests from application units and distribute the model call requests to the corresponding model running instance Pod according to predetermined routing rules and load balancing strategies; Output model scheduling and routing results to corresponding application units.
2. The model calling method based on XGBoost according to claim 1, characterized in that The acquisition of each model trained by the model training unit and storage of each bin file model obtained after training is specifically as follows: Obtain the corresponding bin file models trained by each model training sub-unit of the model training unit, and distribute each bin file model to the corresponding model running instance Pod for storage; each Pod only loads and runs one bin file model.
3. The model calling method based on XGBoost according to claim 1, characterized in that: The receiving unit receives the model call request from the application unit and distributes the model call request to the corresponding model running instance Pod according to the predetermined routing rules and load balancing strategy; including: Receive model call requests from application units, and perform routing and load balancing of model call requests according to predetermined routing rules and load balancing strategies; Routes the request to the corresponding model running instance Pod based on the requested model identifier; Dynamically adjust routing rules based on the user's geographic location and / or time factors; Dynamically adjust the Pod weight based on the actual load of the Pod; Based on historical data on model call volume, predict future load conditions and adjust the weight or number of Pods in advance.
4. The XGBoost-based model calling method according to claim 3, wherein: The model management component manages the entire life cycle of each bin file model, including registration, deployment, update, and retirement of the corresponding bin file model; The Pod configuration of each bin file model is modified online through the model management component, and the model management component uses a container orchestration tool to realize the dynamic creation and destruction of Pod.
5. The model calling system based on XGBoost is characterized by: include: The model training unit is used to train the input XGBoost model and output the trained corresponding model to the model management platform; The model management platform is used to store trained models and schedule and route model call requests from application units. The application unit is used to send model call requests to the model management platform according to needs; and output after receiving feedback from the model management platform.
6. The XGBoost-based model calling system according to claim 5, characterized in that: The model training unit includes several model training subunits, each of which trains the XGBoost model input thereto, generates a corresponding bin file model, and outputs the trained bin file model to the model management platform; The application unit includes several application sub-units, each of which sends a model call request to the model management platform according to demand; and outputs after receiving feedback from the model management platform; The model management platform is used to store the corresponding models trained by each model training sub-unit, and to perform model scheduling and routing for the model call requests sent by each application unit.
7. The XGBoost-based model calling system according to claim 6, characterized in that: The model management platform includes a model management component and several model running instance Pods; The model management component is used to receive the corresponding bin file models trained by each model training sub-unit and distribute each bin file model to the corresponding Pod; it is also used for the full life cycle management of each bin file model, including registration, deployment, update and retirement of the corresponding bin file model; Each Pod is independently deployed in a container, and each Pod only loads and runs one bin file model.
8. The XGBoost-based model calling system according to claim 7, characterized in that: The model management component includes an API gateway, which is used to receive model call requests from the application unit and distribute the model call requests to the corresponding Pod according to routing rules and load balancing strategies; The model management component uses a service grid or predetermined routing rules to implement routing and load balancing of model call requests; The model management component dynamically adjusts routing rules based on the user's geographic location and / or time factors.
9. The XGBoost-based model calling system according to claim 7, characterized in that: The model management component is configured to: The Pod configuration of each bin file model can be modified online. The model management component uses the container orchestration tool to realize the dynamic creation and destruction of Pods; Dynamically adjust the Pod weight based on the actual load of the Pod; Based on historical data on model call volume, predict future load conditions and adjust the weight or number of Pods in advance.
10. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the XGBoost-based model calling method are implemented.