CN111275203A - Decision tree construction method, device, equipment and storage medium based on column storage - Google Patents

Abstract

The invention discloses a decision tree construction method, a device, equipment and a storage medium based on column storage, relating to the technical field of machine learning, wherein the method comprises the following steps: reading corresponding column data according to the first splitting characteristic of each node in the current layer, and dividing each sample to each node in the current layer according to the column data; obtaining grouping samples in each node, and counting the characteristic distribution condition of each characteristic in the grouping samples; and generating a next-layer node according to the characteristic distribution condition, splitting the next-layer node until convergence, and obtaining a decision tree. Therefore, the decision tree is constructed based on the column storage, the data reading time is shortened, and the working efficiency is improved.

Description

Column-based decision tree construction method, device, device and storage medium

technical field

The present invention relates to the technical field of machine learning, and in particular, to a method, device, device and storage medium for constructing a decision tree based on column storage.

Background technique

With the development of computer technology, more and more technologies (big data, distributed, blockchain, artificial intelligence, etc.) are applied in the financial field, and the traditional financial industry is gradually transforming into financial technology (Fintech). The security and real-time requirements of the industry also put forward higher requirements for technology.

Decision trees are widely used in the financial field, but the current decision tree construction is mainly based on row storage for data reading. Constructing a decision tree in the case of a large amount of data will take a lot of time in the process of data reading, and thus the work efficiency is not high.

SUMMARY OF THE INVENTION

The invention provides a decision tree construction method, device, equipment and storage medium based on column storage, aiming at saving the training time of the decision tree and improving the work efficiency.

To achieve the above object, the present invention provides a method for constructing a decision tree based on column storage, the method comprising:

Read the corresponding column data according to the first splitting feature of each node in the current layer, and divide each sample into each node in the current layer according to the column data;

Obtain the grouped samples in the respective nodes, and count the feature distribution of each feature in the grouped samples;

According to the feature distribution, the next layer of nodes is generated;

The next layer nodes are split until convergence, and a decision tree is obtained.

Preferably, each node corresponds to a different first splitting feature, and the step of dividing each sample into each node in the current layer according to the column data includes:

Obtain the first features of the respective samples from the column data, where the first features correspond to the first split features;

Sending the first splitting feature of each node to one or more execution machines, and the one or more execution machines divides each sample into each node in the current layer according to the first feature .

Preferably, the step of generating the next layer of nodes according to the feature distribution includes:

Sending the feature distribution to one or more execution machines, and the one or more execution machines perform parallel solutions to obtain an optimal value;

The optimal value is used as the second split feature to generate the next layer of nodes.

Preferably, the step of reading the corresponding column data according to the first splitting feature of each node in the current layer includes:

The first splitting feature of each node in the current layer is acquired, and column data corresponding to the first splitting feature is read from the data table.

Preferably, before the step of reading the corresponding column data according to the first splitting feature of each node in the current layer, the step further includes:

Scan the sample data, perform feature consistency processing on the sample data, and store the sample data in columns according to the features.

Preferably, the step of acquiring the grouped samples in the respective nodes and counting the feature distribution of each feature in the grouped samples includes:

Denote the samples divided into the same node as the grouped samples, and obtain the grouped samples corresponding to the respective nodes;

Statistics are performed on the grouped samples according to preset feature statistics rules to obtain the feature distribution of each feature.

Preferably, the first splitting feature of each node in the current layer is education, deposit and age, respectively, and the corresponding column data is read according to the first splitting feature of each node in the current layer, and according to the The step of dividing each sample into each node in the current layer by column data includes:

Read the education column data, deposit column data and age column data corresponding to education, deposit and age from the data table;

Sending the educational background column data, deposit column data, and age column data to three execution machines respectively, and the three execution machines divide the samples into each node in the current layer according to the corresponding column data. .

In addition, in order to achieve the above object, the present invention also provides a decision tree construction device based on column storage, and the decision tree construction device based on column storage includes:

a dividing module, configured to read the corresponding column data according to the first splitting feature of each node in the current layer, and divide each sample into each node in the current layer according to the column data;

A statistics module, used to obtain the grouped samples in the respective nodes, and count the feature distribution of each feature in the grouped samples;

a generation module, configured to generate the next layer of nodes according to the feature distribution;

The obtaining module is used for splitting the nodes of the next layer until convergence, and obtaining a decision tree.

In addition, in order to achieve the above object, the present invention also provides a decision tree construction device based on column storage, and the decision tree construction device based on column storage includes a processor, a memory, and a decision tree based on column storage stored in the memory. A tree construction program, when the column storage-based decision tree construction program is run by the processor, implements the steps of the column storage-based decision tree construction method as described above.

In addition, in order to achieve the above object, the present invention also provides a computer storage medium on which a decision tree construction program based on column storage is stored, and the decision tree construction program based on column storage is implemented when a processor runs The steps of the column store-based decision tree construction method are described above.

Compared with the prior art, the present invention provides a decision tree construction method, device, device and storage medium based on column storage. The corresponding column data is read according to the first splitting feature of each node in the current layer, and the corresponding column data is read according to the The column data divides each sample into each node in the current layer; obtains the grouped samples in the each node, and counts the feature distribution of each feature in the grouped sample; generates the following according to the feature distribution. One layer of nodes; split the next layer of nodes until convergence to obtain a decision tree. As a result, the decision tree is constructed based on column storage, which reduces the time for data reading and improves work efficiency.

Description of drawings

1 is a schematic diagram of the hardware structure of a column storage-based decision tree construction device involved in various embodiments of the present invention;

2 is a schematic flowchart of the first embodiment of the method for constructing a decision tree based on column storage of the present invention;

FIG. 3 is a schematic diagram of the functional modules of the first embodiment of the device for constructing a decision tree based on column storage according to the present invention.

The realization, functional characteristics and advantages of the present invention will be further described with reference to the accompanying drawings in conjunction with the embodiments.

Detailed ways

It should be understood that the specific embodiments described herein are only used to explain the present invention, but not to limit the present invention.

The column storage-based decision tree construction device mainly involved in the embodiments of the present invention refers to a network connection device capable of realizing network connection, and the column storage-based decision tree construction device may be a server, a cloud platform, or the like.

Referring to FIG. 1 , FIG. 1 is a schematic diagram of a hardware structure of a decision tree construction device based on column storage according to various embodiments of the present invention. In this embodiment of the present invention, the column storage-based decision tree construction device may include a processor 1001 (eg, Central Processing Unit, CPU), a communication bus 1002 , an input port 1003 , an output port 1004 , and a memory 1005 . Among them, the communication bus 1002 is used to realize the connection communication between these components; the input port 1003 is used for data input; the output port 1004 is used for data output, and the memory 1005 can be a high-speed RAM memory or a non-volatile memory (non-volatile memory). memory), such as a disk memory, the memory 1005 may optionally also be a storage device independent of the aforementioned processor 1001 . Those skilled in the art can understand that the hardware structure shown in FIG. 1 does not constitute a limitation of the present invention, and may include more or less components than those shown in the drawings, or combine some components, or arrange different components.

Continuing to refer to FIG. 1 , the memory 1005 as a readable storage medium in FIG. 1 may include an operating system, a network communication module, an application program module, and a decision tree construction program based on column storage. In FIG. 1, the network communication module is mainly used to connect to the server and perform data communication with the server; and the processor 1001 is used to call the column storage-based decision tree construction program stored in the memory 1005, and perform the following operations:

Read the corresponding column data according to the first splitting feature of each node in the current layer, and divide each sample into each node in the current layer according to the column data;

Obtain the grouped samples in the respective nodes, and count the feature distribution of each feature in the grouped samples;

According to the feature distribution, the next layer of nodes is generated;

The next layer nodes are split until convergence, and a decision tree is obtained.

The processor 1001 is further configured to call the column storage-based decision tree construction program stored in the memory 1005, and perform the following operations:

Obtain the first features of the respective samples from the column data, where the first features correspond to the first split features;

Sending the first splitting feature of each node to one or more execution machines, and the one or more execution machines divides each sample into each node in the current layer according to the first feature

The processor 1001 is further configured to call the column storage-based decision tree construction program stored in the memory 1005, and perform the following operations:

Sending the feature distribution to one or more execution machines, and the one or more execution machines perform parallel solutions to obtain an optimal value;

The optimal value is used as the second split feature to generate the next layer of nodes.

The processor 1001 is further configured to call the column storage-based decision tree construction program stored in the memory 1005, and perform the following operations:

The first splitting feature of each node in the current layer is acquired, and column data corresponding to the first splitting feature is read from the data table.

The processor 1001 is further configured to call the column storage-based decision tree construction program stored in the memory 1005, and perform the following operations:

Scan the sample data, perform feature consistency processing on the sample data, and store the sample data in columns according to the features.

The processor 1001 is further configured to call the column storage-based decision tree construction program stored in the memory 1005, and perform the following operations:

Denote the samples divided into the same node as the grouped samples, and obtain the grouped samples corresponding to the respective nodes;

Statistics are performed on the grouped samples according to preset feature statistics rules to obtain the feature distribution of each feature.

The processor 1001 is further configured to call the column storage-based decision tree construction program stored in the memory 1005, and perform the following operations:

Read the education column data, deposit column data and age column data corresponding to education, deposit and age from the data table;

Sending the educational background column data, deposit column data, and age column data to three execution machines respectively, and the three execution machines divide the samples into each node in the current layer according to the corresponding column data. .

Decision Tree (Decision Tree) is a decision analysis method for evaluating the project risk and judging its feasibility by forming a decision tree to obtain the probability that the expected value of the net present value is greater than or equal to zero on the basis of the known probability of occurrence of various situations. A graphical method that uses probability analysis intuitively. In machine learning, a decision tree is a predictive model that represents a mapping relationship between object attributes and object values.

The embodiment of the present invention provides a decision tree construction method based on column storage. In SQL (Structured Query Language) Server (database), Page (page) is the basic unit of data storage, and data row is the storage unit of actual data, and data starts after Page Header (page header). In turn, they are stored on the Page. This way of storing records on a Page by row is called row storage. When data is stored contiguously in a single column rather than multiple rows, it is called columnar storage. Column storage is inferior to row storage in terms of write efficiency and data integrity. The advantage of column storage is that redundant data will not be generated during the reading process, which is still very important in the field of big data processing where data integrity is not high.

Referring to FIG. 2 , FIG. 2 is a schematic flowchart of a first embodiment of a method for constructing a decision tree based on column storage according to the present invention.

In this embodiment, the column storage-based decision tree construction method is applied to a column storage-based decision tree construction device, and the method includes:

Step S101, reading the corresponding column data according to the first splitting feature of each node in the current layer, and dividing each sample into each node in the current layer according to the column data;

Each node of each layer of the decision tree has its own splitting feature, and the node is split according to the splitting feature to obtain the next node. In this embodiment, the splitting feature of the current layer is marked as the first splitting feature. After the first splitting feature is obtained, the corresponding column data is read according to the first splitting feature. The first splitting feature is acquired, and column data corresponding to the first splitting feature is read from the data table.

In this embodiment, each node corresponds to a different first splitting feature, and the step of dividing each sample into each node in the current layer according to the column data includes:

Obtain the first feature of each sample from the column data, the first feature corresponds to the first split feature; mark the value of the first split feature as each sample according to the column data the first feature. For example, if the first split feature is age, and the age of a certain sample in the read column data is 20 years old, the first feature of the sample is 20 years old.

Sending the first splitting feature of each node to one or more execution machines, and the one or more execution machines divides each sample into each node in the current layer according to the first feature . Since the first splitting feature corresponding to each node is different, the first splitting feature of each node may be sent to multiple execution machines, and the multiple execution machines perform parallel processing. For example, if there are 4 nodes in the current layer, and the corresponding first split features are feature A, feature B, feature C, and feature D, the first split features of these four nodes can be sent to No. 1 for execution. machine, executor No. 2, executor No. 3, and executor No. 4 are processed separately. Thereby, multi-thread processing is realized, and processing efficiency is improved.

Further, before the step of reading the corresponding column data according to the first splitting feature of each node in the current layer, the step further includes:

Scan the sample data, perform feature consistency processing on the sample data, and store the sample data in columns according to the features.

Column storage needs to split a complete piece of data into multiple single columns for storage, and also read the corresponding columns during the reading process. The data stored in the column will be split before being stored, so the sample data that needs to be stored in the column can be processed for feature consistency. Specifically, each piece of sample data is scanned, a target feature value corresponding to the target feature is obtained, the target feature value is stored as target sample data according to a preset logic, and other values in the sample data are ignored. If one or more target eigenvalues are missing in a certain sample data, the sample data is discarded or the missing target eigenvalues are filled in a preset manner. After consistency processing, unified sample data can be obtained.

Further, the sample data is stored in columns according to the target features, and each column contains one target feature.

Step S102, acquiring the grouping samples in the respective nodes, and making statistics on the feature distribution of each feature in the grouping samples;

Specifically, the step of acquiring the grouped samples in the respective nodes and counting the feature distribution of each feature in the grouped samples includes:

Step S102a, denoting the samples divided into the same node as the grouping samples, and obtaining the grouping samples corresponding to the respective nodes;

After each sample is divided into each node of the current layer, each node includes part of the samples in the sample, and the sample divided into the same node is recorded as the grouped sample, so that each node There are corresponding grouping samples, and multiple grouping samples corresponding to each node are obtained respectively.

In step S102b, statistics are performed on the grouped samples according to a preset feature statistics rule to obtain the feature distribution of each feature.

The multiple grouped samples are grouped according to the first splitting feature in the current layer. It is understandable that if other features of the multiple grouped samples are not grouped at the layers above the current layer, they need to be Group in a layer below the current layer.

In this embodiment, a preset feature statistical rule is set, and the preset feature statistical rule may be to use a normal distribution to perform statistics on each feature, or to perform statistics by means of an average value, a maximum value, a minimum value, a median value, etc. . Then, statistics are performed on the grouped samples according to the preset feature statistics rule to obtain the respective characteristics of each feature.

Step S103 , generating a next layer of nodes according to the feature distribution.

After the feature distribution is obtained, the optimal value can be obtained, and the next node is generated by splitting according to the optimal value, so that the decision tree is continuously split until convergence.

Specifically, the step of generating the next layer of nodes according to the feature distribution includes:

Step S103a, sending the feature distribution to one or more execution machines, and the one or more execution machines perform parallel solutions to obtain the optimal value;

Specifically, the feature distribution of each node is respectively sent to one or more execution machines, and the one or more execution machines solve the feature distribution according to a preset algorithm. For example, the optimal value of the feature distribution can be obtained by calculating the entropy and the GINI coefficient (impurity) through an algorithm. For example, the feature with the highest information gain can be selected as the test feature, and the node samples can be divided into subsets by using this feature, so that the mixing degree of samples of different categories in each subset will be the lowest, and the information required to divide the samples in each subset ( entropy) is minimum, and thus the optimal value can be obtained.

In this embodiment, the solution can be solved by a plurality of the execution machines at the same time, which realizes parallel computing and greatly improves the work efficiency.

Step S103b, using the optimal value as the second split feature to generate the next layer of nodes.

The optimal value is the second splitting feature of the node, and the next node is generated based on the second splitting feature.

Step S104: Split the next layer node until convergence, and obtain a decision tree.

Specifically, the node samples in the nodes of the next layer are counted to obtain a second feature distribution, and the execution machine calculates an optimal value based on the second feature distribution, and continues to generate nodes. The above steps are repeated continuously until the convergence condition of the decision tree is reached, and the final decision tree is obtained after pruning. After the decision tree has grown sufficiently, prune off excess branches.

In this embodiment, branch pruning can be performed by post-pruning. Specifically, according to the classification error rate of each branch and the weight of each branch, calculate the expected classification error rate when the node is not pruned; for each non-leaf node, calculate the classification error rate after the node is pruned. When the classification error rate becomes larger, pruning is abandoned; otherwise, the node is forced to be a leaf node and the category is marked. After a series of pruned decision tree candidates are generated, test data (data not involved in modeling) are used to evaluate the classification accuracy of each candidate decision tree, and the decision tree with the smallest classification error rate is retained.

The decision tree model is often used in the financial field, for example, in the bank's credit business, a credit risk decision tree can be constructed. When constructing the credit risk decision tree, a large amount of user data is obtained as a sample, and the sample includes multiple credit features, and the credit features include education, deposit, occupation, age, historical loan records, family situation, fixed assets, and the like.

In the process of constructing the credit risk decision tree, it is assumed that the first splitting features of each node in the current layer are education, deposit and age, respectively, and the corresponding data is read according to the first splitting feature of each node in the current layer. The step of dividing each sample into each node in the current layer according to the column data includes: reading the education column data, deposit column data and age corresponding to education, deposit and age from the data table column data; send the educational background column data, deposit column data and age column data to three execution machines respectively, and the three execution machines divide the samples into the current layer according to the corresponding column data respectively Each node of , the each node is: education node, deposit node and age node.

Further, the samples in the education node, the deposit node and the age node are marked as grouped samples: sample A, sample B and sample C, and the distribution of each feature in each grouped sample is counted respectively. For sample A, obtain the separate conditions of other features except the educational level feature; for sample B, obtain the separate conditions of other features except the deposit feature; for sample C, obtain other features except the age feature It can be understood that the other features can be specifically selected as required, for example, one or more features of the upper-level node of the current-level node can also be eliminated. After obtaining the characteristic distribution of the samples in the education node, the deposit node and the age node, the next node is generated according to the characteristic distribution. For example, for education nodes, it can be split according to occupational characteristics to obtain the next layer of nodes. The above steps are repeated continuously until the convergence conditions are met, and the credit risk decision tree is obtained.

In this embodiment, through the above solution, the corresponding column data is read according to the first splitting feature of each node in the current layer, and each sample is divided into each node in the current layer according to the column data; The grouped samples in the nodes are collected, and the feature distribution of each feature in the grouped samples is counted; according to the feature distribution, the next layer of nodes is generated, and the next layer of nodes is split until convergence to obtain a decision tree. As a result, the decision tree is constructed based on column storage, which reduces the time for data reading and improves work efficiency.

In addition, this embodiment also provides an apparatus for constructing a decision tree based on column storage. Referring to FIG. 3 , FIG. 3 is a schematic diagram of the functional modules of the first embodiment of the apparatus for constructing a decision tree based on column storage according to the present invention.

In this embodiment, the column-stored decision tree construction device is a virtual device, which is stored in the memory 1005 of the column-stored decision tree construction device shown in FIG. 1 , so as to realize all the functions of the column-stored decision tree construction program: It is used to read the corresponding column data according to the first splitting feature of each node in the current layer, and divide each sample into each node in the current layer according to the column data; used to obtain the data in each node. Group the samples, and count the feature distribution of each feature in the grouped samples; for generating the next layer of nodes according to the feature distribution; for splitting the next layer of nodes until convergence, to obtain a decision tree .

Specifically, the column storage-based decision tree construction device includes:

A division module 10, configured to read the corresponding column data according to the first splitting feature of each node in the current layer, and divide each sample into each node in the current layer according to the column data;

A statistical module 20, configured to obtain the grouped samples in the respective nodes, and to count the feature distribution of each feature in the grouped samples;

A generating module 30, configured to generate the next layer of nodes according to the feature distribution;

The obtaining module 40 is used for splitting the nodes of the next layer until convergence, and obtaining a decision tree.

Further, the generation module includes:

a sending unit, configured to send the feature distribution to one or more executive machines, and the one or more executive machines perform parallel solutions to obtain an optimal value;

a generating unit, configured to use the optimal value as the second splitting feature to generate a node of the next layer.

Further, the dividing module includes:

The reading unit is configured to acquire the first splitting feature, and read column data corresponding to the first splitting feature from the data table.

Further, the generation module also includes:

The scanning unit is configured to scan the sample data, perform feature consistency processing on the sample data, and store the sample data in columns according to the features.

Further, the statistical module includes:

an obtaining unit, configured to record the samples divided into the same node as the grouped samples, and obtain the grouped samples corresponding to the respective nodes;

A statistical unit, configured to perform statistics on the grouped samples according to a preset feature statistical rule to obtain the feature distribution of each feature.

In addition, an embodiment of the present invention also provides a computer storage medium, where a column storage-based decision tree construction program is stored, and the column storage-based decision tree construction program is implemented by the processor as described above when running The steps of the column store-based decision tree construction method will not be repeated here.

Compared with the prior art, a method, device, device and storage medium for constructing a decision tree based on column storage proposed by the present invention relate to the technical field of machine learning. The method includes: according to the first splitting feature of each node in the current layer Read the corresponding column data, and divide each sample into each node in the current layer according to the column data; obtain the grouped samples in each node, and count the feature distribution of each feature in the grouped samples ; According to the feature distribution, generate the next layer of nodes; split the next layer of nodes until convergence, and obtain a decision tree. As a result, the decision tree is constructed based on column storage, which reduces the time for data reading and improves work efficiency.

It should be noted that, herein, the terms "comprising", "comprising" or any other variation thereof are intended to encompass non-exclusive inclusion, such that a process, method, article or system comprising a series of elements includes not only those elements, It also includes other elements not expressly listed or inherent to such a process, method, article or system. Without further limitation, an element qualified by the phrase "comprising a..." does not preclude the presence of additional identical elements in the process, method, article or system that includes the element.

The above-mentioned serial numbers of the embodiments of the present invention are only for description, and do not represent the advantages or disadvantages of the embodiments.

From the description of the above embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus a necessary general hardware platform, and of course hardware can also be used, but in many cases the former is better implementation. Based on this understanding, the technical solutions of the present invention can be embodied in the form of software products in essence or the parts that make contributions to the prior art. The computer software products are stored in a storage medium (such as ROM/RAM) as described above. , magnetic disk, optical disk), including several instructions to make a terminal device execute the method described in each embodiment of the present invention.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the scope of the present invention. Any equivalent structure or process transformation made by using the contents of the description and drawings of the present invention, or directly or indirectly applied in other related technical fields , are similarly included in the scope of patent protection of the present invention.

Claims (10)

1. a decision tree construction method based on column storage, is characterized in that, the method comprises: Read the corresponding column data according to the first splitting feature of each node in the current layer, and divide each sample into each node in the current layer according to the column data; Obtain the grouped samples in the respective nodes, and count the feature distribution of each feature in the grouped samples; According to the feature distribution, the next layer of nodes is generated; The next layer nodes are split until convergence, and a decision tree is obtained. 2. The method according to claim 1, wherein each node corresponds to a different first splitting feature, The step of dividing each sample into each node in the current layer according to the column data includes: Obtain the first features of the respective samples from the column data, where the first features correspond to the first split features; Sending the first splitting feature of each node to one or more execution machines, and the one or more execution machines divides each sample into each node in the current layer according to the first feature . 3. The method according to claim 1, wherein the step of generating a next layer of nodes according to the feature distribution comprises: Sending the feature distribution to one or more execution machines, and the one or more execution machines perform parallel solutions to obtain an optimal value; The optimal value is used as the second split feature to generate the next layer of nodes. 4. The method according to claim 1, wherein the step of reading the corresponding column data according to the first splitting feature of each node in the current layer comprises: The first splitting feature of each node in the current layer is acquired, and column data corresponding to the first splitting feature is read from the data table. 5. The method according to claim 1, wherein before the step of reading the corresponding column data according to the first splitting feature of each node in the current layer, the step further comprises: Scan the sample data, perform feature consistency processing on the sample data, and store the sample data in columns according to the features. 6. The method according to claim 1, wherein the step of acquiring the grouped samples in the respective nodes and counting the feature distribution of each feature in the grouped samples comprises: Denote the samples divided into the same node as the grouped samples, and obtain the grouped samples corresponding to the respective nodes; Statistics are performed on the grouped samples according to preset feature statistics rules to obtain the feature distribution of each feature. 7. The method according to claim 1, wherein the first splitting feature of each node in the current layer is education, deposit and age respectively, and the first splitting feature of each node in the current layer is based on the first splitting feature of each node in the current layer. The feature reads the corresponding column data, and divides each sample into each node in the current layer according to the column data, including: Read the education column data, deposit column data and age column data corresponding to education, deposit and age from the data table; Sending the educational background column data, deposit column data, and age column data to three execution machines respectively, and the three execution machines divide the samples into each node in the current layer according to the corresponding column data. . 8. A column storage-based decision tree construction device, wherein the column storage-based decision tree construction device comprises: a dividing module, configured to read the corresponding column data according to the first splitting feature of each node in the current layer, and divide each sample into each node in the current layer according to the column data; A statistics module, used to obtain the grouped samples in the respective nodes, and count the feature distribution of each feature in the grouped samples; a generation module, configured to generate the next layer of nodes according to the feature distribution; The obtaining module is used for splitting the nodes of the next layer until convergence, and obtaining a decision tree. 9. A column storage-based decision tree construction device, wherein the column storage-based decision tree construction device comprises a processor, a memory, and a column storage-based decision tree construction program stored in the memory, wherein When the column storage-based decision tree construction program is run by the processor, the steps of implementing the column storage-based decision tree construction method according to any one of claims 1-7 are implemented. 10. A computer storage medium, characterized in that, a column storage-based decision tree construction program is stored on the computer storage medium, and the column storage-based decision tree construction program is implemented by a processor when it is run as claimed in claim 1- Steps of the column storage-based decision tree construction method described in any one of 7.

Images