An intelligent model unified description and sharing discovery method

Abstract

An embodiment of the present invention discloses a unified description and shared discovery method for intelligent models, comprising the following steps: S10: using PMML to describe the relevant information of the intelligent model to form a PMML model description document, storing the PMML model description document in an XML database, and storing the intelligent model in an intelligent model library; S30: encapsulating the intelligent model to form an intelligent simulation model; S50: responding to an execution instruction to realize the shared discovery of the intelligent model.

Description

Technical Field

[0001] This invention relates to the field of data mining. More specifically, it relates to a method for unified description and shared discovery of intelligent models. Background Art

[0002] In July 1999, the Data Mining Group (DMG) developed Predictive Model Markup Language (PMML) version 1.0, based on XML specifications. Its core idea is to leverage XML's inherent ability to hierarchically structure data and its self-descriptive nature to standardize and portability the description and management of data mining and machine learning models. Users can create predictive models in system A, describe them using the PMML standard, and then transfer this description document to another system B to apply the PMML-described model for prediction. Therefore, the PMML standard can be used not only for modeling but also for data transformation, enabling model sharing across different platform systems and maximizing the effectiveness of machine learning models.

[0003] PMML uses XML to represent data mining models. After more than 20 years of development, it has now been extended to express most machine learning models. The entire structure of PMML is described using XML Schema. A PMML document can contain one or more machine learning models. Each PMML document is an XML document with the root element being of type PMML, including six parts: Header, MiningBuildTask, DataDictionary, TransformationDictionary, MODEL-ELEMENT, and Extension. The Header and DataDictionary are mandatory, while the others are optional. PMML currently supports 18 types of machine learning model descriptions, including association rule models, Bayesian network models, clustering models, and neural networks.

[0004] There are some existing studies on data mining model management based on PMML, but no PMML specification has yet been used to describe intelligent simulation model resources, thus achieving a unified description and shared discovery of intelligent simulation models. Summary of the Invention

[0005] The purpose of this invention is to provide a unified description and sharing discovery method for intelligent models. The intelligent model resource sharing and publishing method is based on the PMML standard. Through unified encapsulation and description, a general and standardized intelligent model description template is formed, enabling the reuse and integration of intelligent models generated by different algorithms, software or platforms. This provides a unified standard registration, publishing and sharing discovery mechanism for intelligent model resource services.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] This invention provides a unified description and shared discovery method for intelligent models, comprising the following steps:

[0008] S10: Use PMML to describe the information related to the intelligent model, form a PMML model description document, store the PMML model description document in an XML database, and store the intelligent model in an intelligent model library.

[0009] S30: Encapsulate the intelligent model to form an intelligent simulation model;

[0010] S50: Enables shared discovery of intelligent models in response to execution instructions.

[0011] In a specific example, the information related to the intelligent model includes: data dictionary, mining pattern, data transformation, model definition, model output, model objective, model interpretation, and model validation.

[0012] In a specific example, before the intelligent model and the PMML model description document are stored in the database, the PMML model description document needs to undergo syntax checking and semantic validation.

[0013] The syntax check is used to check whether the PMML model description document conforms to the PMML description specification;

[0014] The semantic verification is used to verify the model definition, model output, model explanation, and model verification content of the PMML model description document through the extended stylesheet language conversion tool.

[0015] In a specific example, step S30 includes:

[0016] S301: Define the corresponding intelligent simulation model interface according to the different intelligent simulation model functions to be implemented, and hide the implementation details;

[0017] S302: Generate a simulation model framework based on the intelligent simulation model interface;

[0018] S303: When the simulation starts, the simulation engine drives the simulation framework to call the intelligent model to realize the simulation deduction.

[0019] In a specific example, the simulation model framework does not generate any autonomous behavior; its response comes from the intelligent model. Before the simulation is invoked, an intelligent model configuration file is added so that the simulation model framework can find the intelligent model.

[0020] In a specific example, the intelligent simulation model interface is a collection of intelligent simulation model attributes and operation methods, wherein,

[0021] The intelligent simulation model interface includes: intelligent simulation model metadata interface, input interface, output interface, initialization interface, intelligent model interface, and simulation scheduling and control interface.

[0022] In a specific example, the simulation model framework is an interface provided to the outside world, without containing specific functional implementations. In the actual simulation process, the generated simulation model framework interface calls the intelligent model to perform simulation deduction.

[0023] In a specific example, step S50 includes:

[0024] When a simulation system requests to execute the intelligent model, it needs to request the intelligent model from the intelligent model library, find the available intelligent model, and then encapsulate the intelligent model into the simulation platform for execution.

[0025] The beneficial effects of this invention are as follows:

[0026] This invention provides a unified description and sharing discovery method for intelligent models. The intelligent model resource sharing and publishing is based on the PMML standard. Through standardized descriptions, a universally applicable intelligent model description template is formed, enabling the reuse and integration of intelligent models generated by different algorithms, software, or platforms. This provides a unified standard description, encapsulation, and sharing discovery mechanism for intelligent model resource services. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 The flowchart illustrates a method for unified description and sharing of intelligent models according to an embodiment of the present invention. Detailed Implementation

[0029] To make the present invention, its technical solutions, and advantages clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0030] This invention provides a unified description and sharing discovery method for intelligent models. The intelligent model resource sharing and publishing is based on the PMML standard. Through unified encapsulation and description, it provides a unified standard registration, publishing and sharing discovery mechanism for intelligent model resource services.

[0031] One embodiment of the present invention provides a unified description and shared discovery method for intelligent models, such as... Figure 1 As shown, the following steps are included:

[0032] S10: In order to enable the intelligent model to be shared between different platform systems, PMML is used to describe various intelligent model-related information to meet the various needs of users to call on demand. Since the trained intelligent model does not need to be modified frequently, the PMML model description document formed after using PMML description is stored in the XML database, and the intelligent model is stored in the intelligent model library.

[0033] In one specific embodiment, the intelligent model-related information includes: data dictionary, mining mode, data transformation, model definition, model output, model objective, model interpretation, and model validation, as shown in Table 1, which contains the PMML model description document.

[0034] Table 1 PMML Model Description Document

[0035]

[0036]

[0037] In one specific embodiment, before the intelligent model and its PMML model description document are stored in the database, the PMML model description document needs to undergo syntax checking and semantic verification.

[0038] First, a syntax check is performed to verify whether the PMML model description document conforms to the description specification, i.e., a certain version of the PMML description specification. Then, semantic verification is performed by using the Extended Style Sheet Language Transformation (XSLT) tool to verify the model definition, model output, model interpretation, model verification, and other parts of the PMML model description document, thereby ensuring the correctness of the model semantics.

[0039] S30: Encapsulate the intelligent model to form an intelligent simulation model. For an intelligent model to be used as a resource service, it must be encapsulated. The encapsulation process includes: defining the intelligent simulation model interface, generating the simulation model framework, and calling the intelligent model during simulation. Specifically:

[0040] S301: Define the corresponding intelligent simulation model interface according to the different intelligent simulation model functions to be implemented, and hide the implementation details;

[0041] S302: Generate a simulation model framework based on the intelligent simulation model interface;

[0042] S303: When the simulation starts, the simulation engine drives the simulation framework to call the intelligent model to realize the simulation deduction.

[0043] In one specific embodiment, the intelligent simulation model interface is a collection of intelligent simulation model attributes and operation methods, wherein,

[0044] The intelligent simulation model interface includes: an intelligent simulation model metadata interface, an input interface, an output interface, an initialization interface, an intelligent model interface, and a simulation scheduling and control interface, as shown in Table 2.

[0045] Table 2 Intelligent Simulation Model Interface

[0046]

[0047]

[0048]

[0049] In one specific embodiment, the simulation model framework is an interface provided to the outside world, without containing specific functional implementations. In the actual simulation process, the generated simulation model framework interface calls the intelligent model to perform simulation behavior inference.

[0050] In one specific embodiment, the simulation model framework does not generate any autonomous behavior; its response comes from the intelligent model. Before the simulation is invoked, an intelligent model configuration file is added, enabling the simulation model framework to find the intelligent model.

[0051] S50: Enables shared discovery of intelligent models in response to execution instructions.

[0052] In one specific embodiment, step S50 includes: when a simulation system requests to execute the intelligent model, it needs to request the intelligent model from the intelligent model library, obtain the available intelligent model by searching, and then encapsulate the intelligent model into the simulation platform for execution.

[0053] The following description is based on another specific embodiment:

[0054] 1. Intelligent Model Description. This section uses a simple example to demonstrate the process of describing an intelligent model using a PMML model description document. Based on an existing sample dataset, it assumes that an applicable machine learning model has been trained using a machine learning system (Spark, Scikit-learn, SPSS). Here, the expression `x_scaled` represents the standardization of the input variable `x`; `y_1_1`, `y_1_2`, and `y_1_3` are hidden layers expressed using the tangent activation function, and the expression `y_scaled` represents the regression of the output variable; finally, the expression for `y` represents the restoration of the neural network model's output variable.

[0055] The first step is to determine the version of the PMML document specification, model comments, and other information. The second step is to set the header information, which can include attributes such as version, description, model version, and an application element. The third step is to define a data dictionary, where x and y are continuous variables, with x ranging from [0, 5] and y ranging from [0, 25]. The fourth step is to define a transformation dictionary, using the function NormContinuous to transform the values ​​of x and y to the range [-1, 1], while restoring the neural network output value to the range [0, 25]. The fifth step is to set up the prediction model, which contains four elements: MingSchema defines the required fields, NeuralInput defines the neural network input, NeuralLayer defines the activation function and bias of this layer, and NeuralOutput defines the output layer of the neural network, performing a restoration operation on the output.

[0056] 2. Intelligent simulation model encapsulation

[0057] Intelligent models expressed using PMML must be encapsulated before being published as resource services. The encapsulation process for this intelligent simulation model component mainly includes: defining the simulation model interface; generating the simulation model framework; and calling the simulation model resources.

[0058] The core of the simulation model interface definition describes the simulation model interface, which is broadly categorized into simulation model metadata, inputs, outputs, initialization, intelligent models, and simulation scheduling and control. After defining the simulation model interface, a simulation model framework is formed. The generation of this framework primarily addresses the issues of intelligent model integration and simulation execution control. Before simulation execution, the simulation model framework is generated, the intelligent model is loaded, and the intelligent simulation model is created. During simulation execution, an internal interface exists between the intelligent model and the simulation model framework, allowing the framework to call the intelligent model to perform behavioral logic deduction. The simulation model framework also sends control commands such as start, pause, and stop commands to the simulation engine.

[0059] 3. Intelligent simulation model sharing and discovery

[0060] The simulation model resource search is based on the open-source native XML database (NXD) Xindice[]. The lxml[] tool is used to parse the intelligent model document into a tree structure using its Parser, etree, etc., and then XQuery is performed to match the requested simulation resources and intelligent model resources.

[0061] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. For those skilled in the art, other variations or modifications can be made based on the above description. It is impossible to exhaustively list all the implementation methods here. All obvious variations or modifications derived from the technical solutions of the present invention are still within the protection scope of the present invention.

Claims

1. A unified description and shared discovery method for intelligent models, characterized in that, Includes the following steps: S10: Use PMML to describe the information related to the intelligent model, form a PMML model description document, store the PMML model description document in an XML database, and store the intelligent model in an intelligent model library. S30: Encapsulate the intelligent model to form an intelligent simulation model; S50: Enables shared discovery of intelligent models in response to execution instructions; Step S30 includes: S301: Define the corresponding intelligent simulation model interface according to the different intelligent simulation model functions to be implemented, and hide the implementation details; S302: Generate a simulation model framework based on the intelligent simulation model interface; S303: When the simulation starts, the simulation engine drives the simulation model framework to call the intelligent model to realize the simulation deduction; The intelligent simulation model interface is a collection of intelligent simulation model attributes and operation methods, wherein, The intelligent simulation model interface includes: an intelligent simulation model metadata interface, an input interface, an output interface, an initialization interface, an intelligent model interface, and a simulation scheduling and control interface; The simulation model framework does not generate any autonomous behavior; its response comes from the intelligent model. Before the simulation is invoked, an intelligent model configuration file is added so that the simulation model framework can find the intelligent model. Step S50 includes: When a simulation system requests to execute the intelligent model, it needs to request the intelligent model from the intelligent model library, find the available intelligent model, and then encapsulate the intelligent model into the simulation platform for execution.

2. The method according to claim 1, characterized in that, The information related to the intelligent model includes: Data dictionary, mining patterns, data transformation, model definition, model output, model objectives, model interpretation, and model validation.

3. The method according to claim 1, characterized in that, Before the intelligent model and the PMML model description document are stored in the database, the PMML model description document needs to undergo syntax checking and semantic validation. The syntax check is used to check whether the PMML model description document conforms to the PMML description specification; The semantic verification is used to verify the model definition, model output, model explanation, and model verification content of the PMML model description document through the extended stylesheet language conversion tool.

4. The method according to claim 1, characterized in that, The simulation model framework is an interface provided to the outside world, without containing specific functional implementations. In the actual simulation process, the generated simulation model framework interface calls the intelligent model to perform simulation deduction.

Images