CN116187524B - Supply chain analysis model comparison method and device based on machine learning - Google Patents
Supply chain analysis model comparison method and device based on machine learning Download PDFInfo
- Publication number
- CN116187524B CN116187524B CN202211633885.XA CN202211633885A CN116187524B CN 116187524 B CN116187524 B CN 116187524B CN 202211633885 A CN202211633885 A CN 202211633885A CN 116187524 B CN116187524 B CN 116187524B
- Authority
- CN
- China
- Prior art keywords
- supply chain
- rule
- data
- sets
- chain data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0201—Market modelling; Market analysis; Collecting market data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0201—Market modelling; Market analysis; Collecting market data
- G06Q30/0203—Market surveys; Market polls
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/06—Buying, selling or leasing transactions
- G06Q30/0601—Electronic shopping [e-shopping]
- G06Q30/0605—Supply or demand aggregation
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Strategic Management (AREA)
- Finance (AREA)
- Development Economics (AREA)
- Accounting & Taxation (AREA)
- Economics (AREA)
- Entrepreneurship & Innovation (AREA)
- Marketing (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- Game Theory and Decision Science (AREA)
- Human Resources & Organizations (AREA)
- Data Mining & Analysis (AREA)
- Tourism & Hospitality (AREA)
- Quality & Reliability (AREA)
- Operations Research (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
In the comparison method, when two supply chain analysis models are compared, the two supply chain analysis models are converted into two corresponding rule sets, wherein the rule in each rule set can be regarded as the basis for prediction of the corresponding supply chain analysis model. Thereafter, the comparison results of the two supply chain analysis models are determined by calculating the similarity of the two rule sets. It should be noted that, in the process of comparing different supply chain analysis models, the scheme obtains the prediction basis of the models at the same time, which is convenient for the user to intuitively understand the differences between the different supply chain analysis models, that is, to provide explanation information for the differences between the supply chain analysis models.
Description
Technical Field
One or more embodiments of the present disclosure relate to the field of machine learning, and more particularly, to a method and apparatus for comparing supply chain analysis models based on machine learning.
Background
The ultra-large supply chain data of the large supply chain integrated service group company needs to construct a plurality of data analysis models in a plurality of business application scenes to mine data value and enable business management. Data analysis models (abbreviated as supply chain analysis models) applied to supply chain business application scenarios require periodic retraining in business applications, and an updated (updated) model may deviate from the model trained in the previous period in some links without significant changes in overall accuracy. For example, in various online learning scenarios of a supply chain enterprise, the supply chain analysis model may be updated at intervals. If a large number of automated machine learning applications (AutoML) were considered, a large number of supply chain analysis model updates were automatically performed by the machine according to human preset goals, which may be comparable in terms of conventional model metrics such as accuracy, but with some substantial differences between the models. Therefore, how to automatically identify the difference between two models and give an interpretable result is the first problem to be solved.
In existing solutions, people often focus attention on the model accuracy and other metrics. However, when the model is updated, only the accuracy is considered and it is not fully and deeply known which changes have occurred. In the intelligent warehouse analysis scene of the integrated service business in the supply chain, besides knowing the change of the accuracy of the model, the model difference is needed to be known to be exactly where, so that a better supporting effect is achieved on business operation analysis decisions of enterprises, and the decision of business experts on the model is more reasonable. Therefore, providing an interpretable variant abstracted method before redeploying a new supply chain analysis model is critical to establishing understanding and trust. In this case, the interpretable model comparison may serve to highlight model differences and aid in model selection.
For a given two supply chain analysis models, the present approach can identify changes between models and the cause of the changes, thereby supporting model developers and end users to learn about the differences in the models.
Disclosure of Invention
One or more embodiments of the present disclosure describe a method and apparatus for comparing supply chain analysis models based on machine learning, which can facilitate a user to intuitively understand differences between supply chain analysis models.
In a first aspect, a method for comparing supply chain analysis models based on machine learning is provided, including:
acquiring two supply chain analysis models and two corresponding supply chain data sets trained in advance for an intelligent warehouse analysis task; the two supply chain analysis models comprise an already online supply chain analysis model and an upcoming supply chain analysis model; each supply chain data in the supply chain dataset includes inventory features, production features, and sales features;
inputting the two supply chain data sets into the two supply chain analysis models correspondingly to obtain the predicted stock-to-stock ratio of each supply chain data in the two supply chain data sets;
determining two rule sets corresponding to the two supply chain analysis models, wherein the two rule sets comprise training a decision tree for any first supply chain analysis model in the two supply chain analysis models based on the corresponding first supply chain data set and the predicted stock-to-stock ratio of each supply chain data in the corresponding first supply chain data set; the trained decision tree comprises branch nodes and leaf nodes, each branch node corresponds to a splitting characteristic and a characteristic threshold, and each leaf node has a corresponding score which is a stock-to-sell ratio predicted by the decision tree; wherein the split feature comprises an inventory feature, a production feature, or a sales feature; determining a first rule set corresponding to the first supply chain analysis model according to each splitting condition corresponding to a path from a root node to a leaf node in the decision tree and the score of the leaf node; any splitting condition is determined based on splitting characteristics and characteristic thresholds corresponding to branch nodes through which the path passes; obtaining two rule sets corresponding to the two supply chain analysis models in this way;
And calculating the similarity of the two rule sets, and determining the comparison result of the two supply chain analysis models according to the similarity.
In a second aspect, a supply chain analysis model comparison device based on machine learning is provided, including:
the acquisition unit is used for acquiring two supply chain analysis models trained in advance for the intelligent warehouse analysis task and two corresponding supply chain data sets; the two supply chain analysis models comprise an already online supply chain analysis model and an upcoming supply chain analysis model; each supply chain data in the supply chain dataset includes inventory features, production features, and sales features;
the input unit is used for inputting the two supply chain data sets into the two supply chain analysis models correspondingly to obtain the predicted stock-to-stock ratio of each supply chain data in the two supply chain data sets;
a determining unit, configured to determine two rule sets corresponding to the two supply chain analysis models, where the determining unit includes training, for any first supply chain analysis model of the two supply chain analysis models, a decision tree based on the corresponding first supply chain data set and a predicted expense ratio of each supply chain data therein; the trained decision tree comprises branch nodes and leaf nodes, each branch node corresponds to a splitting characteristic and a characteristic threshold, and each leaf node has a corresponding score which is a stock-to-sell ratio predicted by the decision tree; determining a first rule set corresponding to the first supply chain analysis model according to each splitting condition corresponding to a path from a root node to a leaf node in the decision tree and the score of the leaf node; any splitting condition is determined based on splitting characteristics and characteristic thresholds corresponding to branch nodes through which the path passes; obtaining two rule sets corresponding to the two supply chain analysis models in this way;
And the computing unit is used for computing the similarity of the two rule sets and determining the comparison results of the two supply chain analysis models according to the similarity.
In one or more embodiments of the present disclosure, when comparing two supply chain analysis models, the two supply chain analysis models are converted into two rule sets respectively, wherein the rule in each rule set can be regarded as a basis for prediction of the corresponding supply chain analysis model. Thereafter, the comparison results of the two supply chain analysis models are determined by calculating the similarity of the two rule sets. It should be noted that, in the process of comparing the supply chain analysis models, the prediction basis of the models is obtained at the same time, so that the user can conveniently and intuitively know the differences between the supply chain analysis models, that is, explanation information is provided for the differences between the supply chain analysis models.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present description, the drawings that are needed in the description of the embodiments will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present description, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic illustration of an implementation scenario of an embodiment disclosed herein;
FIG. 2 illustrates a machine learning based supply chain analysis model comparison method in accordance with one embodiment;
FIG. 3 shows a schematic diagram of a decision tree in one example;
FIG. 4 illustrates a machine learning based supply chain analysis model comparison device in accordance with one embodiment.
Detailed Description
The following describes the scheme provided in the present specification with reference to the drawings.
Fig. 1 is a schematic diagram of an implementation scenario of an embodiment disclosed in the present specification. In fig. 1, the rule extraction device and the rule comparison device form a model comparison system, and the comparison system is used for obtaining a comparison result of the model.
In particular, a first supply chain analysis model may be trained with a first supply chain dataset and a second supply chain analysis model may be trained with a second supply chain dataset, respectively, for an intelligent warehouse analysis task. The first supply chain data set and the second supply chain data set herein may be supply chain data sets acquired at two periods, respectively, which may have the same calibration label.
The two supply chain analysis models may be an already online supply chain analysis model and an upcoming online supply chain analysis model, respectively.
In addition, each of the first or second supply chain data sets described above includes inventory features, production features, and sales features. In addition, information such as a member company name, a primary member company name to which the member company belongs, and the like can be included.
Wherein the inventory feature comprises at least one of: the warehouse is located in urban areas, warehouse transportation resources, warehouses, warehouse positions, material types, material names, specification models, measurement units, current warehouse numbers, historical highest warehouse numbers, historical lowest warehouse numbers, average warehouse time, warehouse same ratios, warehouse ring ratios and the like.
The production characteristics include at least one of the following: manufacturer, manufacturer's place province district, material class, material name, specification model, unit of measure, productivity, daily output, production cycle, the cumulative output of this month, the cumulative output of this year, the same ratio of output and the annular ratio of output etc..
The sales feature includes at least one of: contract number, contract time, customer name, customer location urban area, customer credit rating, customer credit line, order number, order time, bill number, bill time, material category, material name, specification model, metering unit, accumulated number of warehouse-out per month, accumulated number of ticket-in per year, accumulated number of fund returns per month, accumulated number of fund returns per year, accounts receivable per month, accounts receivable per year, sales homonymy, and sales cycle ratio.
After the two supply chain analysis models are acquired, the two supply chain analysis models can be used for predicting the supply chain data in the corresponding supply chain data sets respectively, so as to obtain the predicted stock-out ratio of each supply chain data in the two supply chain data sets. Thereafter, each supply chain analysis model and the corresponding supply chain data set and the predicted stock-to-stock ratios of the respective supply chain data therein may be input into a rule extraction device and each supply chain analysis model converted to a corresponding rule set using the device. After two rule sets corresponding to the two supply chain analysis models are obtained, the two rule sets can be input into a rule comparison device to calculate the similarity of the two rule sets, and then the comparison result of the two supply chain analysis models is determined.
The method for extracting the rule by the rule extracting device comprises the following steps: the supply chain analysis model is approximated by a directly interpretable model, i.e., a proxy model for training the supply chain analysis model that provides a compact rule set to describe model behavior globally. And then acquiring a corresponding rule set based on the agent model, and acquiring a comparison result of the supply chain analysis model through comparing the rule set. The following is a detailed description.
FIG. 2 illustrates a machine learning based supply chain analysis model comparison method in accordance with one embodiment. The method may be performed by any apparatus, device, platform, cluster of devices having computing, processing capabilities. As shown in fig. 2, the method may include the following steps.
Step 202, two supply chain analysis models and corresponding two supply chain data sets are obtained, which are pre-trained for the intelligent warehouse analysis task.
The two supply chain analysis models may be an already online supply chain analysis model and an immediately online supply chain analysis model, respectively.
The two supply chain data sets may include a first supply chain data set and a second supply chain data set, wherein the first supply chain data set may be a supply chain data set acquired during a first period of time and the second supply chain data set may be a supply chain data set acquired during a second period of time. The partial moments in the first and second time periods may be repeated here, so that the partial supply chain data of the two supply chain data sets are repeated, i.e. the common data set of the two supply chain data sets.
In addition, the two training supply chain data sets have the same calibration label (i.e., a label that is manually pre-calibrated), so that two supply chain analysis models trained based on the two supply chain data sets can be predicted for the same prediction task (e.g., a smart warehouse analysis task).
The two supply chain analysis models include: a first supply chain analysis model trained based on the first supply chain dataset, and a second supply chain analysis model trained based on the second supply chain dataset. That is, a first supply chain analysis model corresponds to a first supply chain dataset and a second supply chain analysis model corresponds to a second supply chain dataset.
Step 204, inputting the two supply chain data sets into the two supply chain analysis models respectively to obtain the predicted stock-to-stock ratio of each supply chain data in the two supply chain data sets.
Namely, inputting the first supply chain data set into the first supply chain analysis model to obtain the predicted expense ratio (i.e. the predicted result) of each supply chain data in the first supply chain data set, and inputting the second supply chain data set into the second supply chain analysis model to obtain the predicted expense ratio (i.e. the predicted result) of each supply chain data in the second supply chain data set.
It should be noted that the above-mentioned predicted stock-out ratio may be a binary value, for example, may be 0 or 1, where 0 indicates a low stock-out ratio and 1 indicates a high stock-out ratio. Multiple classification values, such as high, medium, and low, are also possible, and are not limited in this specification.
At step 206, two rule sets corresponding to the two supply chain analysis models are determined.
It should be understood that the method for determining the corresponding rule set is similar for two supply chain analysis models, and thus, the method for determining the corresponding first rule set will be described below by taking an arbitrary first supply chain analysis model as an example.
In one example, determining the first rule set for the first supply chain analysis model may include training a decision tree based on the corresponding first supply chain data set and the predicted stock-to-stock ratios of the respective supply chain data therein. The trained decision tree includes branch nodes, each of which corresponds to a split feature (including inventory feature, production feature, or sales feature) and a feature threshold, and leaf nodes, each of which has a corresponding score, which is a stock-to-sell ratio predicted by the decision tree. And determining a first rule set corresponding to the first supply chain analysis model according to a plurality of splitting conditions corresponding to paths from the root node to the leaf nodes in the decision tree and the scores of the leaf nodes. Any splitting condition is determined based on splitting characteristics (including inventory characteristics, production characteristics or sales characteristics) corresponding to branch nodes through which the path passes and characteristic thresholds; thus two rule sets corresponding to the two supply chain analysis models are obtained.
The training process of the decision tree is described first.
It is assumed that the first supply chain data set described above can be expressed as: d1 = { F (i) ,y (i) } N i＝1 Where N is the number of supply chain data. Wherein F is (i) Is the eigenvector of the ith supply chain data, which is, for example, an n-dimensional vector, i.e., f= (F) 1 ,f 2 ,…，f n ) Wherein f i For stock, production, or sales features, y (i) A predicted stock-out ratio for the ith supply chain data. Then, dividing N supply chain data by a decision tree, setting splitting characteristics and characteristic thresholds at each branch node of the decision tree by dividing the N supply chain data at the branchesThe corresponding features of the supply chain data are compared to the feature threshold at the dry node to segment the supply chain data into respective child nodes, through which process the N supply chain data are finally segmented into individual leaf nodes. Thus, the score of each leaf node, i.e., the predicted stock-to-stock ratio (i.e., y) of each supply chain data in that leaf node (i) ) Is a mean value of (c).
On this basis, further decision trees can be trained further in the direction of the residual reduction. That is, after the decision tree is obtained, the residual r of each supply chain data is obtained by subtracting the predicted stock-out ratio of each supply chain data from the score of the leaf node of that supply chain data in the decision tree (i) In the form of d2= { F (i) ,r (i) } N i＝1 Is a new first supply chain dataset that corresponds to the same supply chain dataset as D1. In the same way as described above, a further decision tree can be obtained in which the N supply chain data are also partitioned into respective leaf nodes, and the score of each leaf node is the average of the residual values of the respective supply chain data. Similarly, multiple decision trees may be acquired sequentially, each decision tree being obtained based on the residuals of the previous decision tree.
FIG. 3 shows a schematic diagram of a decision tree in one example. In fig. 3, the trained decision tree includes branch nodes (node 0 and node 2) and leaf nodes (node 1, node 3 and node 4), each of which sets a split feature and a feature threshold, and the respective supply chain data is entered into the next branch node by comparing the corresponding feature to the feature threshold at the branch node, and finally divided into the leaf nodes. The split feature herein includes an inventory feature, a production feature, or a sales feature.
In fig. 3, assuming that the splitting characteristic corresponding to the node 0 is f1 and the characteristic threshold is 0.5, the "f1 ∈0.5" marked on the connection line of the node 0 to the node 1 can be understood as the splitting condition from the node 0 to the node 1. Similarly, the "f1>0.5" marked on the connection of node 0 to node 2 can be understood as a split condition from node 0 to node 2. The split condition is obtained by adding a split feature of the parent node and a feature threshold value and a comparison operator. That is, the splitting conditions described in this specification include splitting characteristics and characteristic thresholds. In addition, a comparison operator is included.
It can be seen that in the decision tree obtained by training, the path from the root node to the leaf node passes through a plurality of splitting conditions, in other words, one path in the decision tree corresponds to a set of splitting conditions, and one supply chain data is divided into the division basis of the corresponding leaf node, or can be understood as the prediction basis of the model.
In addition, leaf nodes in the decision tree may be trained to obtain a corresponding score, such as a mean of predicted stock-to-stock ratios of the individual supply chain data in the leaf node, or a mean of residuals. It will be appreciated that when the predicted stock ratio is either 0 (indicating a stock ratio is low) or 1 (indicating a stock ratio is high), then the leaf node score is either 0 or 1.
It should be noted that, a rule may be formed based on a set of splitting conditions corresponding to each path in the decision tree, and taking the score of the leaf node as a conclusion. I.e. the number of paths in the decision is the same as the number of rules formed.
Taking fig. 3 as an example, three rules can be formed as follows:
rule 1: splitting conditions: f1 is less than or equal to 0.5; conclusion: 0;
rule 2: splitting conditions: f1 is more than 0.5, and f2 is less than or equal to 0.6; conclusion: 0;
rule 3: splitting conditions: f1>0.5, f2>0.6; conclusion: 1.
The three rules form a first rule set.
It should be appreciated that fig. 3 is merely an exemplary illustration, and that in actual practice, the comparison operator in the split condition may be equal or unequal, and so on. In addition, the decision tree may include more or fewer paths, which is not limited in this specification.
In the decision tree, the first supply chain analysis model may be used as a teacher model, the decision tree may be used as a student model, and the decision tree may be trained based on a knowledge distillation method.
Wherein, training the decision tree based on the knowledge distillation method may include: and inputting each supply chain data in the first supply chain data set into a decision tree to obtain the output of each supply chain data. A first predicted loss is calculated based on the output of each supply chain data and the predicted stock-out ratio, and a second predicted loss is calculated based on the output of each supply chain data and the calibration label. And adjusting parameters of the decision tree according to the comprehensive loss of the first prediction loss and the second prediction loss.
In another example, determining a first set of rules corresponding to a first supply chain analysis model may include: and determining a preset barrel of the discrete feature in the first supply chain data according to any first supply chain data in the first supply chain data set, and coding a barrel division result of the discrete feature by utilizing a preset coding mode. And combining the coding results corresponding to the discrete features to form first discrete data corresponding to the first supply chain data. Thus, each discrete data corresponding to each supply chain data in the first supply chain data set is obtained. The predicted stock-out ratio of each supply chain data is used as a predictive label of each discrete data. Based on each piece of discrete data and the corresponding predictive label, training a linear model, and selecting the discrete data, for which the predictive result of the linear model is consistent with the predictive label, from the discrete data as target data. And determining a first rule set based on the discrete features in the target data, the preset sub-buckets to which the discrete features belong, and the predictive labels of the target data.
It should be understood that the discrete feature is a feature in which the pointers are pre-divided into a plurality of sub-buckets for the corresponding values. In one example, the discrete feature may be an age, and the value interval of the age is assumed to be: [10,20], then the sub-buckets divided in advance for them can be as follows: 10 years old, 11 years old, 12 years old, … years old, 20 years old.
In addition, the preset encoding mode may be a one-hot encoding mode. For example, assuming that the age of a user in a certain supply chain data is 20 years old, the corresponding encoding result may be: 00100000000.
for the first supply chain data, after each encoding result corresponding to each discrete feature is determined, each encoding result is combined to obtain the first discrete data. In addition, the predicted stock-out ratio of the first supply chain data may also be determined as a predictive tag of the first discrete data.
In addition, the scheme selects the discrete data, for which the prediction result of the linear model is consistent with the prediction label, from the discrete data as target data, so as to improve the accuracy of the formed rule set. Because, consistent with predictive labels, the predictive results herein can be understood as consistent with the output of two models (including a supply chain analysis model and a linear model) for discrete data, and thus the discrete data herein can be understood as a typical data, the accuracy of the rules formed can be improved when forming rules based on the typical data.
It should be noted that, in the above another example, one discrete feature in the target data and the preset sub-bucket to which it belongs correspond to one splitting condition in the above one example, except that the comparison operator in the splitting condition determined based on the other example only includes: yes and no.
Taking the above age as an example, assuming that the age of the user in a certain target data is 20 years, that is, the bucket to which the age of the user in the target data belongs is 20 years, the corresponding split condition can be determined: the age was 20 years.
It should be appreciated that a plurality of split conditions may be derived based on each discrete feature in the target data and the preset bins to which it belongs. A rule in the other example described above may be derived based on the plurality of split conditions and the predictive label of the target data. That is, in the other example described above, the number of rules in the first rule set and the number of selected target data are the same.
Further, in the above-described other example, the predictive label of the target data corresponds to the conclusion in the above-described one example.
In summary, the rules in the rule set (including the first rule set and the second rule set) determined in the embodiment of the present disclosure include the following two aspects: split conditions and conclusions.
Step 208, calculating the similarity of the two rule sets, and determining the comparison result for the two supply chain analysis models according to the similarity.
In one example, the same number of target rules may be sampled from the two rule sets, respectively, to form two target rule sets, and then the similarity of the two target rule sets may be calculated. Thus, the similarity calculation efficiency can be improved.
It should be understood that the similarity of the two rule sets is similar to the calculation method of the similarity of the two target rule sets, and thus, the description will be given below taking the calculation of the similarity of the two rule sets as an example.
Specifically, for any first rule in the first rule set, the similarity of each rule in the first rule and the second rule set is calculated. And fusing the similarities calculated for all the rules in the first rule set, and taking the obtained fusion result as the similarity of the two rule sets.
Since the method of calculating the similarity between the first rule and each rule in the second rule set is similar, the method of calculating the similarity between the first rule and any second rule in the second rule set will be described below as an example.
In one example, calculating the similarity of the first rule and the second rule may include: a common data set of the first supply chain data set and the second supply chain data set is first acquired. A first sub-data set is determined from the common data set that matches the first rule, and a second sub-data set is determined from the common data set that matches the second rule. Determining the number of coincident data of the first sub-data set and the second sub-data set, and the total number of data in the combined sub-data set obtained after combining the first sub-data set and the second sub-data set. And calculating the quotient of the number of the coincident data and the total number of the data, and taking the obtained quotient as the similarity of the first rule and the second rule.
In one example, the similarity of the first rule to the second rule may be calculated according to the following formula.
Wherein r is 1 R is the first rule 2 For the second rule, S (r 1 ,r 2 ) For the similarity of the first rule and the second rule, X (r 1 ) For a first set of sub-data matching a first rule, determined from a common dataset, X (r 2 ) A second set of sub-data that matches the second rule is determined from the common dataset. |. | represents the number of samples taken.
The matching with the first rule may be understood that the features in the supply chain data in the first sub-data set satisfy the corresponding split condition in the first rule (i.e., the split feature in the split condition is a feature in the supply chain data), and the corresponding predicted expense ratio is the same as the conclusion in the first rule. Similarly, matching with the second rule may be understood as the features in the supply chain data in the second sub-data set satisfying the corresponding split conditions in the second rule and the corresponding predicted stock-out ratios being the same as the conclusions in the second rule.
Similarly, the similarity of each rule in the first rule set to each rule in the second rule set may be calculated, and the similarity of each rule in the first rule set to each rule in the second rule set may be calculated.
Then, the similarity calculated for each rule in the first rule set can be fused, and the obtained fusion result is used as the similarity of the two rule sets.
In one example, the fusing may include: and averaging or weighted averaging the calculated similarity of each rule in the first rule set, and taking the obtained average value or weighted average value as the similarity of the two rule sets.
It should be noted that, the similarity of the two rule sets calculated based on the above formula 1 is also referred to as a statistical similarity, that is, the difference between the two rule sets is described from a statistical perspective. The statistical similarity takes a value between 0 and 1.
In another example, calculating the similarity of the first rule and the second rule may include: converting the first rule and the second rule into two element sets respectively, wherein each element set comprises a plurality of three element groups, and each three element group comprises: features, comparison operators, and feature values. And calculating the scoring of the two three-element group sets based on the number and the type of the same elements in the two three-element groups by using a scoring function. The scoring for the two element sets is determined based on the scoring for each two of the three element groups in the two element sets. And taking the calculated scoring of the two element sets as the similarity of the first rule and the second rule.
Taking the first rule as an example, converting the first rule into the element set may include: splitting each splitting condition in the first rule into a three-element group, the one three-element group comprising: features, comparison operators, and feature values. Each three-element group corresponding to each splitting condition in the first rule may constitute an element set corresponding to the first rule.
Taking fig. 3 as an example, the conversion of rule 2 into the element set may include:
splitting condition f1>0.5 into three element groups: f1, >, 0.5;
splitting the splitting condition f2 less than or equal to 0.6 into three element groups: f2 is less than or equal to 0.6.
That is, the element set corresponding to rule 2 is composed of two three element groups.
Furthermore, in one specific implementation, the scoring function described above may score two three-element groups based on the following rules: 1) If the features in the two three-element groups are different, the score of the two three-element groups is 0; 2) If the features in the two three-element groups are the same but the comparison operators are different, the scoring of the two three-element groups is 0.25; 3) If the features and comparison operators in the two three-element groups are the same, but the feature values are different, the scoring of the two three-element groups is 0.5; 4) If the feature, comparison operator, and feature value are the same in both three-element groups, the score for both three-element groups is 1.0.
It should be appreciated that, after calculating the score of each three-element group in one element set with respect to each three-element group in another element set using a scoring function, the obtained scores may be averaged, and the obtained average value may be used as the score of the two element sets, thereby obtaining the similarity of the first rule and the second rule.
Similarly, the similarity of each rule in the first rule set to each rule in the second rule set may be calculated, and the similarity of each rule in the first rule set to each rule in the second rule set may be calculated.
Then, the similarity calculated for each rule in the first rule set can be fused, and the obtained fusion result is used as the similarity of the two rule sets.
It should be noted that the similarity of two rule sets calculated by using the scoring function is also called semantic similarity, that is, the difference between the two rule sets is described from the semantic point of view.
In the present specification, the similarity of the two rule sets calculated by any one of the above methods is directly used as the similarity of the two supply chain analysis models, that is, as the comparison result of the two supply chain analysis models.
Of course, in practical applications, the similarity calculated by the two methods may be fused (for example, averaged), and the fusion result may be used as the similarity of the two supply chain analysis models, which is not limited in this specification.
By combining the above, the machine learning-based supply chain analysis model comparison method provided in the embodiments of the present disclosure converts the two supply chain analysis models into two corresponding rule sets, and then analyzes the conditions in the rules to obtain the semantic similarity of the two rule sets, or counts the number of data covered by the rules to obtain the statistical similarity of the two rule sets, so as to obtain the comparison result of the two supply chain analysis models. That is, by comparing two rule sets to obtain the variability of the two supply chain analysis models, it may be convenient for the user to intuitively learn about the variability between the supply chain analysis models, thereby enabling accurate localization and analysis of possible model problems, which helps to maintain model stability.
Corresponding to the above-mentioned supply chain analysis model comparison method based on machine learning, an embodiment of the present disclosure further provides a supply chain analysis model comparison device based on machine learning, as shown in fig. 4, the device may include:
an acquisition unit 402, configured to acquire two supply chain analysis models and corresponding two supply chain data sets trained in advance for an intelligent warehouse analysis task.
Wherein the two supply chain analysis models include an already online supply chain analysis model and an upcoming supply chain analysis model, and each of the supply chain data sets includes inventory characteristics, production characteristics, and sales characteristics.
The inventory feature includes at least one of: the warehouse is located in urban areas, warehouse transportation resources, warehouses, warehouse positions, material types, material names, specification models, measurement units, current warehouse numbers, historical highest warehouse numbers, historical lowest warehouse numbers, average warehouse time, warehouse same ratios, warehouse ring ratios and the like.
The production characteristics include at least one of the following: manufacturer, manufacturer's place province district, material class, material name, specification model, unit of measure, productivity, daily output, production cycle, the cumulative output of this month, the cumulative output of this year, the same ratio of output and the annular ratio of output etc..
The sales feature includes at least one of: contract number, contract time, customer name, customer location urban area, customer credit rating, customer credit line, order number, order time, bill number, bill time, material category, material name, specification model, metering unit, accumulated number of warehouse-out per month, accumulated number of ticket-in per year, accumulated number of fund returns per month, accumulated number of fund returns per year, accounts receivable per month, accounts receivable per year, sales homonymy, and sales cycle ratio.
And an input unit 404, configured to input the two supply chain data sets into the two supply chain analysis models respectively, so as to obtain a predicted stock-to-sell ratio of each supply chain data in the two supply chain data sets.
The determining unit 406 is configured to determine two rule sets corresponding to the two supply chain analysis models, where the determining unit includes training, for any first supply chain analysis model of the two supply chain analysis models, a decision tree based on the corresponding first supply chain set and a predicted stock-out ratio of each supply chain data therein. The trained decision tree comprises branch nodes and leaf nodes, each branch node corresponds to a splitting characteristic and a characteristic threshold, and each leaf node has a corresponding score which is a stock-to-sell ratio predicted by the decision tree. Where the split feature includes an inventory feature, a production feature, or a sales feature. And determining a first rule set corresponding to the first supply chain analysis model according to each splitting condition corresponding to the path from the root node to the leaf node and the score of the leaf node in the decision tree. Any splitting condition is determined based on splitting characteristics and characteristic thresholds corresponding to branch nodes through which the path passes; thus two rule sets corresponding to the two supply chain analysis models are obtained.
A calculating unit 408, configured to calculate the similarity between the two rule sets, and determine a comparison result for the two supply chain analysis models according to the calculated similarity.
In some embodiments, the determining unit 406 is further configured to:
determining a preset sub-bucket to which the discrete feature in the first supply chain data belongs by determining any first supply chain data in the first supply chain data, coding the sub-bucket result of the discrete feature by utilizing a preset coding mode, and combining all coding results to form first discrete data; obtaining each piece of discrete data corresponding to each piece of supply chain data in the first supply chain data set;
taking the predicted stock-out ratio of each supply chain data as a predicted label of each discrete data;
training a linear model based on each piece of discrete data and a corresponding prediction tag, and selecting discrete data, for which a prediction result of the linear model is consistent with the prediction tag, from the discrete data as target data;
and determining a first rule set based on the discrete features in the target data, the preset sub-buckets to which the discrete features belong, and the predictive labels of the target data.
In some embodiments, the determining unit 406 is further specifically configured to:
And training the decision tree based on a knowledge distillation method by taking the first supply chain analysis model as a teacher model and taking the decision tree as a student model.
In some embodiments, the supply chain data in the first supply chain data set includes calibration tags;
training the decision tree based on the knowledge distillation method comprises the following steps:
inputting each supply chain data in the first supply chain data set into a decision tree to obtain the output of each supply chain data;
calculating a first predicted loss according to the output of each supply chain data and the predicted stock-out ratio, and calculating a second predicted loss according to the output of each supply chain data and the calibration label;
and adjusting parameters of the decision tree according to the comprehensive loss of the first prediction loss and the second prediction loss.
In some embodiments, the two rule sets include a first rule set and a second rule set; the computing unit 408 is specifically configured to:
for any first rule in the first rule set, calculating the similarity of each rule in the first rule and the second rule set;
and fusing the similarities calculated for all the rules in the first rule set, and taking the obtained fusion result as the similarity of the two rule sets.
In some embodiments, the second set of rules includes a second rule; the computing unit 408 is also specifically configured to:
obtaining a common data set of the first supply chain data set and the second supply chain data set;
determining a first sub-data set matched with the first rule from the common data set, and determining a second sub-data set matched with the second rule from the common data set;
determining the number of coincident data of the first sub-data set and the second sub-data set, and the total data number in the combined sub-data set obtained after combining the first sub-data set and the second sub-data set;
and calculating the quotient of the number of the coincident data and the total number of the data, and taking the obtained quotient as the similarity of the first rule and the second rule.
In some embodiments, the second set of rules includes a second rule; the computing unit 408 is also specifically configured to:
converting the first rule and the second rule into two element sets respectively, wherein each element set comprises a plurality of three element groups, and each three element group comprises: features, comparison operators, and feature values;
calculating scoring of the two three-element groups based on the number and types of the same elements in the two three-element groups by using a scoring function;
Determining a score for the two element sets based on the scores for each two of the three element groups in the two element sets;
and taking the scoring of the two element sets as the similarity of the first rule and the second rule.
In some embodiments, the computing unit 408 is further specifically configured to:
and averaging or weighted averaging the calculated similarity of each rule in the first rule set, and taking the obtained average value or weighted average value as the similarity of the two rule sets.
In some embodiments, the apparatus further comprises: a sampling unit 410;
a sampling unit 410, configured to sample the same number of target rules from the two rule sets respectively, to form two target rule sets;
the computing unit 408 is also for:
and calculating the similarity of the two target rule sets.
The functions of the functional modules of the apparatus in the foregoing embodiments of the present disclosure may be implemented by the steps of the foregoing method embodiments, so that the specific working process of the apparatus provided in one embodiment of the present disclosure is not repeated herein.
The machine learning-based supply chain analysis model comparison device provided by the embodiment of the specification can be convenient for a user to intuitively know the difference between the supply chain analysis models.
In this specification, each embodiment is described in a progressive manner, and identical and similar parts of each embodiment are all referred to each other, and each embodiment mainly describes differences from other embodiments. In particular, for the device embodiments, since they are substantially similar to the method embodiments, the description is relatively simple, and reference is made to the description of the method embodiments in part.
The steps of a method or algorithm described in connection with the disclosure herein may be embodied in hardware, or may be embodied in software instructions executed by a processor. The software instructions may be comprised of corresponding software modules that may be stored in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor such the processor can read information from, and write information to, the storage medium. In the alternative, the storage medium may be integral to the processor. The processor and the storage medium may reside in an ASIC. In addition, the AS IC may be located in a server. The processor and the storage medium may reside as discrete components in a server.
Those skilled in the art will appreciate that in one or more of the examples described above, the functions described in the present invention may be implemented in hardware, software, firmware, or any combination thereof. When implemented in software, these functions may be stored on or transmitted over as one or more instructions or code on a computer-readable medium. Computer-readable media includes both computer storage media and communication media including any medium that facilitates transfer of a computer program from one place to another. A storage media may be any available media that can be accessed by a general purpose or special purpose computer.
The foregoing describes specific embodiments of the present disclosure. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims can be performed in a different order than in the embodiments and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
The foregoing detailed description of the embodiments has further described the objects, technical solutions and advantages of the present specification, and it should be understood that the foregoing description is only a detailed description of the embodiments of the present specification, and is not intended to limit the scope of the present specification, but any modifications, equivalents, improvements, etc. made on the basis of the technical solutions of the present specification should be included in the scope of the present specification.
Claims (8)
1. A machine learning based supply chain analysis model comparison method, comprising:
acquiring two supply chain analysis models and two corresponding supply chain data sets trained in advance for an intelligent warehouse analysis task; the two supply chain analysis models comprise an already online supply chain analysis model and an upcoming supply chain analysis model; each supply chain data in the supply chain dataset includes inventory features, production features, and sales features;
inputting the two supply chain data sets into the two supply chain analysis models correspondingly to obtain the predicted stock-to-stock ratio of each supply chain data in the two supply chain data sets;
determining two rule sets corresponding to the two supply chain analysis models, wherein the two rule sets comprise training a decision tree for any first supply chain analysis model in the two supply chain analysis models based on the corresponding first supply chain data set and the predicted stock-to-stock ratio of each supply chain data in the corresponding first supply chain data set; the trained decision tree comprises branch nodes and leaf nodes, each branch node corresponds to a splitting characteristic and a characteristic threshold, and each leaf node has a corresponding score which is a stock-to-sell ratio predicted by the decision tree; wherein the split feature comprises an inventory feature, a production feature, or a sales feature; determining a first rule set corresponding to the first supply chain analysis model according to each splitting condition corresponding to a path from a root node to a leaf node in the decision tree and the score of the leaf node; any splitting condition is determined based on splitting characteristics and characteristic thresholds corresponding to branch nodes through which the path passes; obtaining two rule sets corresponding to the two supply chain analysis models in this way; the supply chain data in the first supply chain data set includes calibration tags;
Calculating the similarity of the two rule sets, and determining the comparison result of the two supply chain analysis models according to the similarity;
the training decision tree comprises:
training a decision tree by taking the first supply chain analysis model as a teacher model and taking the decision tree as a student model based on a knowledge distillation method;
the training of the decision tree based on the knowledge distillation method comprises:
inputting each supply chain data in the first supply chain data set into a decision tree to obtain the output of each supply chain data;
calculating a first predicted loss based on the output of each supply chain data and the predicted stock-out ratio, and calculating a second predicted loss based on the output of each supply chain data and the calibration label;
and adjusting parameters of the decision tree according to the comprehensive loss of the first predicted loss and the second predicted loss.
2. The method of claim 1, wherein,
the inventory feature includes at least one of: the warehouse is located in urban areas, warehouse transportation resources, warehouses, warehouse positions, material categories, material names, specification models, measurement units, current warehouse numbers, historical highest warehouse numbers, historical lowest warehouse numbers, average warehouse time, inventory same ratios and inventory ring ratios;
The production characteristics include at least one of: the manufacturer, the province area where the manufacturer is located, the material category, the material name, the specification model, the measurement unit, the productivity, the daily output, the production period, the accumulated output of the month, the accumulated output of the year, the same ratio of the output and the annular ratio of the output;
the sales feature includes at least one of: contract number, contract time, customer name, customer location urban area, customer credit rating, customer credit line, order number, order time, bill number, bill time, material category, material name, specification model, metering unit, accumulated number of warehouse-out per month, accumulated number of ticket-in per year, accumulated number of fund returns per month, accumulated number of fund returns per year, accounts receivable per month, accounts receivable per year, sales homonymy and sales cycle.
3. The method of claim 1, wherein the first rule set is further determinable by:
determining a preset sub-bucket to which the discrete feature in the first supply chain data belongs by determining any first supply chain data in the first supply chain data, encoding the sub-bucket result of the discrete feature by using a preset encoding mode, and combining all encoding results to form first discrete data; obtaining each piece of discrete data corresponding to each piece of supply chain data in the first supply chain data set;
Taking the predicted stock-out ratio of each piece of supply chain data as a predicted label of each piece of discrete data;
training a linear model based on each piece of discrete data and the corresponding predictive label, and selecting discrete data, for which the predictive result of the linear model is consistent with the predictive label, from the discrete data as target data;
and determining the first rule set based on the discrete features in the target data and the preset sub-buckets to which the discrete features belong and the predictive labels of the target data.
4. The method of claim 1, wherein the two rule sets comprise a first rule set and a second rule set; the calculating the similarity of the two rule sets includes:
for any first rule in the first rule set, calculating the similarity of each rule in the first rule and the second rule set;
and fusing the similarities calculated for all the rules in the first rule set, and taking the obtained fusion result as the similarity of the two rule sets.
5. The method of claim 4, wherein the second set of rules comprises a second rule; the calculating the similarity between the first rule and each rule in the second rule set includes:
Obtaining a common dataset of the two supply chain datasets;
determining a first sub-data set from the common dataset that matches the first rule, and determining a second sub-data set from the common dataset that matches the second rule;
determining the number of coincident data of the first sub-data set and the second sub-data set, and the total number of data in the combined sub-data set obtained after the first sub-data set and the second sub-data set are combined;
and calculating the quotient of the number of the coincident data and the total number of the data, and taking the obtained quotient as the similarity of the first rule and the second rule.
6. The method of claim 4, wherein the second set of rules comprises a second rule; the calculating the similarity between the first rule and each rule in the second rule set includes:
converting the first rule and the second rule into two element sets respectively, wherein each element set comprises a plurality of three element groups, and each three element group comprises: features, comparison operators, and feature values;
calculating a scoring of the two three-element groups based on the number and type of the same elements in the two three-element groups by using a scoring function;
Determining a score for the two element sets based on the scores for each two three element groups in the two element sets;
and taking the scoring of the two element sets as the similarity of the first rule and the second rule.
7. The method of claim 1, wherein prior to said calculating the similarity of the two rule sets, further comprising:
sampling the same number of target rules from the two rule sets respectively to form two target rule sets;
the calculating the similarity of the two rule sets includes:
and calculating the similarity of the two target rule sets.
8. A machine learning based supply chain analysis model comparison apparatus comprising:
the acquisition unit is used for acquiring two supply chain analysis models trained in advance for the intelligent warehouse analysis task and two corresponding supply chain data sets; the two supply chain analysis models comprise an already online supply chain analysis model and an upcoming supply chain analysis model; each supply chain data in the supply chain dataset includes inventory features, production features, and sales features;
the input unit is used for inputting the two supply chain data sets into the two supply chain analysis models correspondingly to obtain the predicted stock-to-stock ratio of each supply chain data in the two supply chain data sets;
A determining unit, configured to determine two rule sets corresponding to the two supply chain analysis models, where the determining unit includes training, for any first supply chain analysis model of the two supply chain analysis models, a decision tree based on the corresponding first supply chain data set and a predicted expense ratio of each supply chain data therein; the trained decision tree comprises branch nodes and leaf nodes, each branch node corresponds to a splitting characteristic and a characteristic threshold, and each leaf node has a corresponding score which is a stock-to-sell ratio predicted by the decision tree; determining a first rule set corresponding to the first supply chain analysis model according to each splitting condition corresponding to a path from a root node to a leaf node in the decision tree and the score of the leaf node; any splitting condition is determined based on splitting characteristics and characteristic thresholds corresponding to branch nodes through which the path passes; obtaining two rule sets corresponding to the two supply chain analysis models in this way; the supply chain data in the first supply chain data set includes calibration tags;
the computing unit is used for computing the similarity of the two rule sets and determining the comparison result of the two supply chain analysis models according to the similarity;
The determining unit is specifically configured to:
training a decision tree by taking the first supply chain analysis model as a teacher model and taking the decision tree as a student model based on a knowledge distillation method;
the determining unit is further specifically configured to:
inputting each supply chain data in the first supply chain data set into a decision tree to obtain the output of each supply chain data;
calculating a first predicted loss based on the output of each supply chain data and the predicted stock-out ratio, and calculating a second predicted loss based on the output of each supply chain data and the calibration label;
and adjusting parameters of the decision tree according to the comprehensive loss of the first predicted loss and the second predicted loss.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211633885.XA CN116187524B (en) | 2022-12-19 | 2022-12-19 | Supply chain analysis model comparison method and device based on machine learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211633885.XA CN116187524B (en) | 2022-12-19 | 2022-12-19 | Supply chain analysis model comparison method and device based on machine learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116187524A CN116187524A (en) | 2023-05-30 |
| CN116187524B true CN116187524B (en) | 2023-08-25 |
Family
ID=86445230
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211633885.XA Active CN116187524B (en) | 2022-12-19 | 2022-12-19 | Supply chain analysis model comparison method and device based on machine learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116187524B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116402241B (en) * | 2023-06-08 | 2023-08-18 | 浙江大学 | Multi-model-based supply chain data prediction method and device |
| CN116975626B (en) * | 2023-06-09 | 2024-04-19 | 浙江大学 | Automatic updating method and device for supply chain data model |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111476274A (en) * | 2020-03-16 | 2020-07-31 | 宜通世纪科技股份有限公司 | Big data prediction analysis method, system, device and storage medium |
| CN113038118A (en) * | 2020-12-08 | 2021-06-25 | 重庆邮电大学 | Video QoE prediction system and method based on big data environment |
| CN113489751A (en) * | 2021-09-07 | 2021-10-08 | 浙江大学 | Network traffic filtering rule conversion method based on deep learning |
| CN113516297A (en) * | 2021-05-26 | 2021-10-19 | 平安国际智慧城市科技股份有限公司 | Prediction method and device based on decision tree model and computer equipment |
| CN114443783A (en) * | 2022-04-11 | 2022-05-06 | 浙江大学 | Supply chain data analysis and enhancement processing method and device |
| CN114648118A (en) * | 2021-12-02 | 2022-06-21 | 天翼电子商务有限公司 | Rule derivation method based on decision tree model and iterative elimination |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160300169A1 (en) * | 2015-04-10 | 2016-10-13 | Caterpillar Inc. | Supply chain network sensitivity analysis using adjusted network parameters |
| US11531915B2 (en) * | 2019-03-20 | 2022-12-20 | Oracle International Corporation | Method for generating rulesets using tree-based models for black-box machine learning explainability |
| US20210142169A1 (en) * | 2019-11-08 | 2021-05-13 | Accenture Global Solutions Limited | Prediction interpretation |
-
2022
- 2022-12-19 CN CN202211633885.XA patent/CN116187524B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111476274A (en) * | 2020-03-16 | 2020-07-31 | 宜通世纪科技股份有限公司 | Big data prediction analysis method, system, device and storage medium |
| CN113038118A (en) * | 2020-12-08 | 2021-06-25 | 重庆邮电大学 | Video QoE prediction system and method based on big data environment |
| CN113516297A (en) * | 2021-05-26 | 2021-10-19 | 平安国际智慧城市科技股份有限公司 | Prediction method and device based on decision tree model and computer equipment |
| CN113489751A (en) * | 2021-09-07 | 2021-10-08 | 浙江大学 | Network traffic filtering rule conversion method based on deep learning |
| CN114648118A (en) * | 2021-12-02 | 2022-06-21 | 天翼电子商务有限公司 | Rule derivation method based on decision tree model and iterative elimination |
| CN114443783A (en) * | 2022-04-11 | 2022-05-06 | 浙江大学 | Supply chain data analysis and enhancement processing method and device |
Non-Patent Citations (1)
| Title |
|---|
| 基于数据挖掘的供应链风险预测研究;刘智鹏;中国优秀硕士学位论文全文数据库 经济与管理科学辑;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116187524A (en) | 2023-05-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN116187524B (en) | Supply chain analysis model comparison method and device based on machine learning | |
| CN108320171B (en) | Hot-sold commodity prediction method, system and device | |
| US20040181526A1 (en) | Robust system for interactively learning a record similarity measurement | |
| CN113535963B (en) | Long text event extraction method and device, computer equipment and storage medium | |
| CN111738534B (en) | Training of multi-task prediction model, and prediction method and device of event type | |
| Chitra et al. | Customer retention in banking sector using predictive data mining technique | |
| CN110991474A (en) | Machine learning modeling platform | |
| CN104573130A (en) | Entity resolution method based on group calculation and entity resolution device based on group calculation | |
| CN110310012B (en) | Data analysis method, device, equipment and computer readable storage medium | |
| CN113254507B (en) | Intelligent construction and inventory method for data asset directory | |
| CN113807900A (en) | RF order demand prediction method based on Bayesian optimization | |
| CN114358657B (en) | Post recommendation method and device based on model fusion | |
| CN116561134B (en) | Business rule processing method, device, equipment and storage medium | |
| CN112148986A (en) | Crowdsourcing-based top-N service re-recommendation method and system | |
| CN108305174B (en) | Resource processing method, device, storage medium and computer equipment | |
| CN115526315A (en) | Generation method and device of rating card model | |
| CN115203206A (en) | Data content searching method and device, computer equipment and readable storage medium | |
| CN114357175A (en) | Data mining system based on semantic network | |
| CN114372148A (en) | Data processing method based on knowledge graph technology and terminal equipment | |
| CN116402241B (en) | Multi-model-based supply chain data prediction method and device | |
| Karpukhin et al. | HoTPP Benchmark: Are We Good at the Long Horizon Events Forecasting? | |
| KR20200129132A (en) | Data preparation method and data utilization system for data utilization | |
| CN118093769B (en) | Product data management method and system based on ERP service | |
| CN118036920A (en) | Supplier competition type matching method and system based on photovoltaic demand | |
| CN117312439A (en) | Client relation mining system, method and medium based on AI inheritance model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |