Inference acceleration method and inference prediction system for predicting loan risk

By converting the decision tree model into a tensor computing model suitable for GPU, the problem of low inference efficiency of decision tree model under large-scale data is solved, efficient acceleration of lending risk prediction is achieved, and the reasoning of financial loan default risk control is significantly improved Speed ​​and efficiency.

CN119048221AInactive Publication Date: 2024-11-29WESHARE TECH SERVICES (SHENZHEN) LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411101884.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-12
Publication Date
2024-11-29
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

When using decision tree models to predict lending risk, traditional CPU operations cannot meet the needs of huge data scale.

Method used

Convert the decision tree model into a mathematical model suitable for GPU tensor operations, thereby using GPU to accelerate inference, implement batch parallel inference prediction, and improve the prediction efficiency of the decision tree model.

Benefits of technology

The efficiency of lending risk reasoning and prediction has been significantly improved, the inference results remain unchanged, and the efficiency has been increased by hundreds of times, meeting the processing needs of massive data.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119048221A_ABST
    Figure CN119048221A_ABST
Patent Text Reader

Abstract

The invention relates to an inference acceleration method and an inference prediction system for predicting a loan risk, and the method comprises the steps: obtaining a plurality of loan customer information groups, and analyzing a preset comprehensive index of each loan customer information group from the plurality of loan customer information groups; judging whether the preset comprehensive indexes of the loan customer information groups meet preset requirements or not; for the loan customer information group meeting the preset requirement, starting a decision tree model to perform reasoning prediction of default probability, and obtaining a first reasoning result; and for the loan customer information group which does not meet the preset requirement, converting the decision tree model into a tensor operation model, and starting the tensor operation model to perform reasoning prediction of default probability to obtain a second reasoning result. The method and the device have the beneficial effects that the efficiency of reasoning and predicting the debit and credit risk can be remarkably improved under the condition that the reasoning result is kept unchanged.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of financial intelligent analysis, and in particular to an inference acceleration method and an inference prediction system for predicting lending risks. Background Art

[0002] With the rapid development of financial businesses such as consumer loans and cash loans, how to reasonably avoid the default risk of individual borrowers has become an urgent problem for financial platforms to solve. In recent years, big data and machine learning technologies have become increasingly mature, and more and more big data teams have devoted themselves to financial business, applying various machine learning models to financial risk prediction; among them, data engineers use decision trees to predict whether borrowers will default in the future, which is an effective prediction method.

[0003] When using the decision tree model to predict lending risk, there is a problem that the amount of data to be inferred and predicted is particularly large, and the number of feature dimensions n is very large; if traditional CPU calculations are used for prediction, the efficiency of the inference calculation cannot meet the needs of the huge data scale, considering that the time complexity is proportional to the product of m and n. Therefore, the industry urgently needs a technical solution that can accelerate the operating efficiency of the decision tree model and thus improve the efficiency of inference prediction of lending risk. Summary of the invention

[0004] The technical problem to be solved by the present invention is: how to design an inference acceleration method for predicting loan risks, which can significantly improve the efficiency of inference prediction of loan risks.

[0005] To solve the above problems, the embodiments of the present invention propose a reasoning acceleration method and a reasoning prediction system for predicting loan risks, which aims to solve the problem that the decision tree model is used for reasoning in the field of financial loan default risk control, and the operation efficiency is low and cannot meet the growing data scale. The inventor believes that the structure of the decision tree itself is not suitable for accelerated reasoning using GPU. By converting the decision tree model into a mathematical model suitable for GPU tensor operations, that is, a tensor operation model, the GPU can be used for accelerated reasoning, and its reasoning efficiency can be improved by hundreds of times, and the reasoning result remains unchanged.

[0006] GPU tensor operation acceleration technology has been effectively applied in many deep learning models, but the structure of the decision tree model is not suitable for direct acceleration using GPU tensor operations. Therefore, it is of great significance to convert the decision tree model into a tensor operation model suitable for GPU tensor operations, thereby using the powerful computing power of the GPU to greatly improve the prediction and reasoning efficiency of the decision tree.

[0007] In the first aspect, the present application proposes a method for accelerating reasoning for predicting lending risks, which is used to accelerate the process of obtaining reasoning results, and includes: S1, obtaining multiple loan customer information groups, and parsing out preset comprehensive indicators of each loan customer information group from the multiple loan customer information groups; S2, judging whether the preset comprehensive indicators of the multiple loan customer information groups meet the preset requirements; S3, for the loan customer information groups that meet the preset requirements, starting the decision tree model to perform reasoning prediction of the probability of default, and obtaining the first reasoning result; S4, for the loan customer information groups that do not meet the preset requirements, converting the decision tree model into a tensor operation model, starting the tensor operation model to perform reasoning prediction of the probability of default, and obtaining the second reasoning result. Among them, the multiple loan customer information groups are equivalent to a large number of samples to be inferred; the decision tree model corresponds to the reasoning prediction of the conventional speed, and the tensor operation model corresponds to the faster reasoning prediction.

[0008] Furthermore, the preset sample set includes multiple training samples; the conversion of the decision tree model into a tensor operation model includes: S11, converting the original decision tree into a decision tree model; S12, performing feature processing based on the decision tree model to obtain five feature matrices; S13, based on the sample set X to be inferred and the five feature matrices, performing inference prediction of the customer's default probability to obtain the result matrix R.

[0009] Furthermore, the converting the original decision tree into a decision tree model includes: S101, training the original decision tree according to n features of a plurality of training samples and obtaining a decision tree model.

[0010] Furthermore, the feature processing is performed based on the decision tree model to obtain five feature matrices, including: S102, based on the n features of multiple training samples and the m decision nodes of the decision tree model, the first matrix E corresponding to the decision tree model is obtained; S103, according to the m decision nodes in the decision tree model and according to the feature judgment condition, the threshold matrix T is obtained; S104, the second matrix L is parsed from the decision tree model; S105, according to the number of elements with a value of 1 in each column of the second matrix L, the Count matrix is ​​obtained; S106, the third matrix M is parsed from the decision tree model. In one embodiment, the feature judgment condition is: Fn is less than Threshold, where Fn is a feature and Threshold is a threshold.

[0011] Furthermore, the inference prediction of the customer's default probability is performed based on the sample set X to be inferred and the five feature matrices to obtain the result matrix R, including: S107, obtaining the sample set X to be inferred whose feature number is consistent with the feature number and feature definition of the training set; S108, based on the sample set X to be inferred, the first matrix E, the threshold matrix T, the second matrix L, the Count matrix, and the third matrix M, batch parallel inference prediction is implemented to obtain the result matrix R.

[0012] Furthermore, the obtaining of multiple loan customer information groups and parsing out preset comprehensive indicators of each loan customer information group from the multiple loan customer information groups includes: obtaining multiple loan customer information groups and parsing out data indicators and characteristic indicators of each loan customer information group from the multiple loan customer information groups.

[0013] Further, the determination of whether the preset comprehensive indicators of multiple loan customer information groups meet the preset requirements includes: determining whether the data indicators of multiple loan customer information groups meet the preset data requirements; determining the determination result objects that do not meet the preset data requirements as loan customer information groups that do not meet the preset requirements; determining whether the remaining loan customer information groups that meet the preset data requirements meet the preset characteristic degree requirements; determining the determination result objects that do not meet the preset characteristic degree requirements as loan customer information groups that do not meet the preset requirements; determining the determination result objects that meet the preset characteristic degree requirements as loan customer information groups that meet the preset requirements. In the above scheme, the final result meets the preset requirements, and at the same time meets the data indicator, i.e., the data volume m, and meets the characteristic degree indicator, i.e., the number of characteristic dimensions n.

[0014] In a second aspect, the present application further proposes an inference prediction system, which includes a unit for executing the method described in the first aspect. The technical effect of the inference prediction system is that it can accelerate the operating efficiency of the decision tree model, thereby improving the efficiency of inference prediction of loan risks.

[0015] The following is a detailed description of the specific algorithms used in each step:

[0016] First, a decision tree model is obtained, corresponding to step S101.

[0017] As a prerequisite, there is a training sample set, Train-Set, which contains several training samples, each of which has n features. Through decision tree machine learning training, a binary decision tree has been obtained (non-binary trees can be converted into standard binary decision trees). For the sake of convenience, a training sample set is used as an example for the subsequent explanation. Assume that the sample set has r samples, each of which has 5 features. The sample set example is as follows: Figure 3 The trained decision tree is Figure 4As shown. In practical applications, the embodiments of the present invention do not limit the number of samples and the number of features, and the shape and parameters of the decision tree are also not restricted.

[0018] Second, calculate the first matrix E of feature - decision nodes, corresponding to the steps of S102.

[0019] The first matrix E is the matrix for calculating features and for decision nodes. If there are n features in the sample and m decision nodes (non - leaf nodes) in the decision tree: taking Figure 3 the sample set as an example, there are 6 features in total. Taking Figure 4 the binary decision tree shown as an example, there are 6 decision nodes (blue nodes in the figure). Therefore, the first matrix E is a 6 - row and 6 - column matrix. Each row represents a feature, and each column represents a decision node. The value of the i - th row and j - th column in the first matrix E is E ij ; where the value of E ij is defined as:

[0020]

[0021] Therefore Figure 4 the value of the first matrix E of the decision tree in

[0022]

[0023] Third, calculate the threshold matrix T, corresponding to the steps of S103.

[0024] In the decision tree, each decision node (blue nodes in the figure) corresponds to a decision condition, and the decision condition is F n <Threshold, where F n is a feature and Threshold is a threshold. The thresholds for all decision nodes 1, node 2,..., node n are arranged in sequence to form a threshold matrix T.

[0025] T = [1.2 2.5 3.6 1.8 100.2 61.8]

[0026] Taking Figure 4 the decision tree as an example, the value of the threshold matrix T is as shown in the above formula.

[0027] Fourth, calculate the second matrix L of leaf node - decision nodes, corresponding to the steps of S104.

[0028] In Figure 4 the decision tree shown, the orange nodes are leaf nodes and the blue nodes are decision nodes. The number of rows of the second matrix L is equal to the number of decision nodes, and the number of columns is equal to the number of leaf nodes. L ijrepresents the i-th row and j-th column element of the second matrix L, which is defined as:

[0029]

[0030] for Figure 4 For the decision tree shown, the value of the second matrix L is:

[0031]

[0032] In the above-mentioned second matrix L, rows 0 to 5 respectively represent decision node 0, decision node 1, decision node 2, decision node 3, decision node 4, and decision node 5; rows 0 to 6 respectively represent leaf node 6, leaf node 7, leaf node 8, leaf node 9, leaf node 10, leaf node 11, and leaf node 12.

[0033] Fifth, define the Count matrix, corresponding to step S105.

[0034] The Count matrix is ​​defined as the number of elements with a value of 1 in each column of the second matrix L. Therefore, for the second matrix L in the above example, the corresponding Count matrix is:

[0035] Count = [2 1 2 1 0 2 1]

[0036] Sixth, define a third matrix M of leaf nodes and classification mapping, corresponding to step S106.

[0037] Among them, the number of rows of M is the number of leaf nodes, and the number of columns is the number of categories. ij represents the i-th row and j-th column of the third matrix M. The third matrix M is defined as:

[0038]

[0039] The decision tree has a total of 7 leaf nodes and 2 classification categories, so the third matrix M has 7 rows and 2 columns; the first column corresponds to category 1, and the second column corresponds to category 2.

[0040] The third matrix M of the decision tree is:

[0041]

[0042] Seventh, batch parallel reasoning, corresponding to steps S107 and S108.

[0043] The sample set X to be inferred has an unlimited number of rows, and the number of features of X is consistent with the number of features and feature definitions of the training set. The result set obtained by inference is represented by the result matrix R. The inference calculation formula is:

[0044] R = (((X × E<T>) × L) == Count) × M (Inference formula)

[0045] The number of rows of the result matrix R is equal to the number of rows of the input samples, and the number of columns is equal to the number of classification categories. The value at the i-th row and j-th column in the result matrix R is R ij , where the definition of R is:

[0046]

[0047] For example: The i-th row vector of R is [1, 0], indicating that the i-th sample belongs to the classification corresponding to the 1st column, i.e., category 1.

[0048] In the inference formula, the less-than operation of matrices is defined as: Given two matrices P and Q, where P and Q have the same number of rows and columns, the result of P less than 0 operation is a matrix S with the same number of rows and columns as P and Q. The value at the i-th row and j-th column of the S matrix is S ij, The definition of matrix S is:

[0049] If S = P < Q, then

[0050] In the inference formula, the identity operation of matrices is defined as: Given two matrices P and Q, where P and Q have the same number of rows and columns, the result of P == Q operation is a matrix S with the same number of rows and columns as P and Q. The definition of matrix S is:

[0051] If S is the result of P == Q, then

[0052] Based on the above matrix linear algebra operation inference formula, a GPU programming technology is used to build an inference engine based on tensor operations, and its inference operation efficiency can be increased by hundreds of times compared with the CPU.

[0053] In summary, when using the decision tree model to predict lending risks, there are problems that the amount of data m to be inferred and predicted is extremely large, and the number of its feature dimensions n is very large; if traditional CPU operations are used for prediction, considering that the time complexity is proportional to the product of m and n, the inference operation efficiency cannot meet the needs of the huge data scale; for the above technical problems, the solution of this application can significantly improve the inference prediction efficiency of lending risks while keeping the inference results unchanged. Brief Description of the Drawings

[0054] The drawings here are incorporated into the specification and form a part of this specification, showing the embodiments in line with the present invention, and are used together with the specification to explain the principles of the present invention.

[0055] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0056] Figure 1 A flow chart of the inference acceleration method for predicting loan risks provided by the present invention.

[0057] Figure 2 Another flow chart of the inference acceleration method for predicting lending risk provided by the present invention.

[0058] Figure 3 A sample set diagram of the inference acceleration method for predicting loan risks provided by the present invention.

[0059] Figure 4 A decision tree diagram of the inference acceleration method for predicting lending risks provided by the present invention. Figure 5 A block diagram of an electronic device provided by the present invention. DETAILED DESCRIPTION

[0060] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0061] It should be understood that when used in this specification and the appended claims, the terms "include" and "comprises" indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or other features, integers, steps, operations, elements, components and / or combinations thereof.

[0062] It should also be understood that the terms used in the present specification are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the present specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0063] It should be further understood that the term "and / or" used in the present specification and the appended claims refers to one or any combination and all possible combinations of the associated listed items, and includes these combinations.

[0064] As used in this specification and the appended claims, the term "if" may be interpreted as "when" or "upon" or "in response to determining" or "in response to detecting," depending on the context. Similarly, the phrases "if it is determined" or "if [described condition or event] is detected" may be interpreted as meaning "upon determination" or "in response to determining" or "upon detection of [described condition or event]" or "in response to detecting [described condition or event]," depending on the context.

[0065] Example 1

[0066] See also Figures 1 to 4 , among which Figure 1 An inference acceleration method for predicting loan risks provided in an embodiment of the present invention is used to accelerate the process of obtaining inference results. The inference acceleration method includes:

[0067] S1, obtaining a plurality of loan customer information groups, and parsing out a preset comprehensive index of each loan customer information group from the plurality of loan customer information groups.

[0068] S2, determining whether the preset comprehensive indicators of multiple loan customer information groups meet preset requirements.

[0069] S3, for the loan customer information group that meets the preset requirements, start the decision tree model to perform inference prediction of the default probability and obtain the first inference result.

[0070] S4, for the loan customer information group that does not meet the preset requirements, the decision tree model is converted into a tensor operation model, and the tensor operation model is started to perform inference prediction of the default probability to obtain a second inference result.

[0071] In one embodiment, the preset sample set includes a plurality of training samples; the step of converting the decision tree model into a tensor operation model includes:

[0072] S11, converting the original decision tree into a decision tree model.

[0073] S12, performing feature processing based on the decision tree model to obtain five feature matrices.

[0074] S13, based on the sample set X to be inferred and the five feature matrices, perform inference prediction on the customer's default probability and obtain the result matrix R.

[0075] In one embodiment, the inference acceleration method for predicting loan risk described in this application specifically includes the following steps:

[0076] S101, training the original decision tree according to n features of multiple training samples and obtaining a decision tree model.

[0077] S102, based on n features of multiple training samples and m decision nodes of the decision tree model, obtain a first matrix E corresponding to the decision tree model.

[0078] S103, obtaining a threshold matrix T according to the m decision nodes in the decision tree model and the feature decision conditions.

[0079] S104, parsing a second matrix L from the decision tree model.

[0080] S105 , obtaining a Count matrix according to the number of elements with a value of 1 in each column of the second matrix L.

[0081] S106, parsing a third matrix M from the decision tree model.

[0082] S107, obtaining a sample set X to be inferred whose feature number and feature definition are consistent with those of the training set.

[0083] S108, based on the sample set X to be inferred, the first matrix E, the threshold matrix T, the second matrix L, the Count matrix, and the third matrix M, batch parallel inference prediction is implemented to obtain the result matrix R.

[0084] Furthermore, in the inference acceleration method for predicting lending risk, the step of obtaining multiple loan customer information groups and parsing out preset comprehensive indicators for each loan customer information group from the multiple loan customer information groups includes: obtaining multiple loan customer information groups and parsing out data indicators and characteristic indicators for each loan customer information group from the multiple loan customer information groups.

[0085] Furthermore, in the inference acceleration method for predicting lending risk, the step of judging whether the preset comprehensive indicators of the plurality of loan customer information groups meet the preset requirements includes:

[0086] S701, determining whether the data indicators of multiple loan customer information groups meet preset data requirements.

[0087] S702, determining the judgment result objects that do not meet the preset data requirements as loan customer information groups that do not meet the preset requirements.

[0088] S703, determining whether the remaining loan customer information groups that meet the preset data requirements meet the preset characteristic degree requirements.

[0089] S704, determining the judgment result objects that do not meet the preset characteristic degree requirements as loan customer information groups that do not meet the preset requirements.

[0090] S705, determining the judgment result objects that meet the preset characteristic degree requirements as the loan customer information group that meets the preset requirements.

[0091] The embodiment of the present invention also provides an inference prediction system, which includes a unit for executing the method described in any of the above embodiments. The inference prediction system can greatly improve the inference speed of the default decision tree model in financial businesses such as consumer loans and cash loans by hundreds of times; it can also greatly improve the efficiency of default prediction for massive loan customers, thereby better improving fund security and utilization efficiency, and providing important support for the digital transformation and intelligent development in the field of financial intelligent analysis.

[0092] In summary, when using a decision tree model to predict loan risks, there is a problem that the amount of data m to be inferred and predicted is particularly large, and the number of its feature dimensions n is very large; if traditional CPU operations are used for prediction, then considering that the time complexity is proportional to the product of m and n, the efficiency of its reasoning operation cannot meet the needs of a large data scale; in response to the above technical problems, the present application solution can significantly improve the efficiency of reasoning and predicting loan risks while keeping the reasoning results unchanged.

[0093] Example 2

[0094] See also Figure 5 , Figure 5 This is a block diagram of an electronic device provided by the present invention. The electronic device can be a terminal or a server, wherein the terminal can be an electronic device with communication functions such as a smart phone, a tablet computer, a laptop computer, a desktop computer, a personal digital assistant, and a wearable device. The server can be an independent server or a server cluster composed of servers.

[0095] It includes a processor 111 , a communication interface 112 , a memory 113 , and a communication bus 114 , wherein the processor 111 , the communication interface 112 , and the memory 113 communicate with each other via the communication bus 114 .

[0096] The memory 113 is used to store computer programs.

[0097] In one embodiment of the present invention, the processor 111 is used to implement the method provided by any one of the aforementioned method embodiments when executing the program stored in the memory 113 .

[0098] It should be understood that in the embodiment of the present application, the processor 111 may be a central processing unit (CPU), and the processor 502 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0099] It is understood by those skilled in the art that all or part of the processes in the method for implementing the above embodiment can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a storage medium, which is a computer-readable storage medium. The computer program is executed by at least one processor in the computer system to implement the process steps of the embodiment of the above method.

[0100] Therefore, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the method provided in any one of the aforementioned method embodiments are implemented.

[0101] The storage medium is a physical, non-transient storage medium, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a magnetic disk, or an optical disk, etc., which can store program codes. The computer-readable storage medium can be non-volatile or volatile.

[0102] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions, but such implementation should not be considered to be beyond the scope of the present invention.

[0103] In the several embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of each unit is only a logical function division, and there may be other division methods in actual implementation. For example, units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0104] The steps in the method of the embodiment of the present invention can be adjusted in order, combined and deleted according to actual needs. The units in the device of the embodiment of the present invention can be combined, divided and deleted according to actual needs. In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0105] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, terminal, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention.

[0106] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0107] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

[0108] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for accelerating reasoning for predicting lending risk, characterized in that: The inference acceleration method is used to accelerate the process of obtaining inference results, and the inference acceleration method includes: S1, obtaining multiple loan customer information groups, and parsing out preset comprehensive indicators of each loan customer information group from the multiple loan customer information groups; S2, determining whether the preset comprehensive indicators of the plurality of loan customer information groups meet the preset requirements; S3, for the loan customer information group that meets the preset requirements, start the decision tree model to perform inference prediction of the default probability and obtain the first inference result; S4, for the loan customer information group that does not meet the preset requirements, the decision tree model is converted into a tensor operation model, and the tensor operation model is started to perform inference prediction of the default probability to obtain a second inference result.

2. The inference acceleration method for predicting loan risk according to claim 1, characterized in that: The preset sample set includes a plurality of training samples; the step of converting the decision tree model into a tensor operation model, starting the tensor operation model to perform inference prediction of the probability of default, and obtaining a second inference result includes: S11, converting the original decision tree into a decision tree model; S12, performing feature processing based on the decision tree model to obtain five feature matrices; S13, based on the sample set X to be inferred and the five feature matrices, perform inference prediction on the customer's default probability and obtain the result matrix R.

3. The inference acceleration method for predicting lending risk according to claim 2, characterized in that: The converting the original decision tree into a decision tree model comprises: S101, training the original decision tree according to n features of multiple training samples and obtaining a decision tree model.

4. The inference acceleration method for predicting lending risk according to claim 3, characterized in that: The feature processing is performed based on the decision tree model to obtain five feature matrices, including: S102, obtaining a first matrix E corresponding to the decision tree model based on n features of a plurality of training samples and m decision nodes of the decision tree model; S103, obtaining a threshold matrix T according to the m decision nodes in the decision tree model and the feature decision conditions; S104, parsing a second matrix L from the decision tree model; S105, obtaining a Count matrix according to the number of elements with a value of 1 in each column of the second matrix L; S106, parsing a third matrix M from the decision tree model.

5. The reasoning acceleration method for predicting lending risk according to claim 4, characterized in that: The inference prediction of the customer's default probability is performed based on the sample set X to be inferred and the five feature matrices to obtain the result matrix R, including: S107, obtaining a sample set X to be inferred whose feature number is consistent with the feature number and feature definition of the training set; S108, based on the sample set X to be inferred, the first matrix E, the threshold matrix T, the second matrix L, the Count matrix, and the third matrix M, batch parallel inference prediction is implemented to obtain the result matrix R.

6. The inference acceleration method for predicting loan risk according to claim 1, characterized in that: The step of obtaining a plurality of loan customer information groups and parsing out a preset comprehensive index of each loan customer information group from the plurality of loan customer information groups includes: A plurality of loan customer information groups are obtained, and data indicators and characteristic indicators of each loan customer information group are parsed from the plurality of loan customer information groups.

7. The inference acceleration method for predicting lending risk according to claim 6, characterized in that: The step of judging whether the preset comprehensive indicators of the plurality of loan customer information groups meet the preset requirements includes: Determining whether the data indicators of the plurality of loan customer information groups meet the preset data requirements; Determine the judgment result objects that do not meet the preset data requirements as loan customer information groups that do not meet the preset requirements; Determine whether the remaining loan customer information groups that meet the preset data requirements meet the preset feature degree requirements; Determine the judgment result objects that do not meet the preset characteristic degree requirements as loan customer information groups that do not meet the preset requirements; The judgment result objects that meet the preset characteristic degree requirements are determined as loan customer information groups that meet the preset requirements.

8. An inference prediction system, characterized in that: The inference prediction system comprises a unit for executing the method according to any one of claims 1 to 7.