Training methods, systems, media, and programs for intelligent tile transformation recommendation models
Through the training method of intelligent tile transformation recommendation model, the problem of low efficiency of tile transformation operation in CAD design is solved. The recommended model is generated through machine learning, manual input is reduced, and the operation efficiency and model effectiveness are improved. It is suitable for complex Tiles change scene.
Patent Information
- Application Number
- CN202411297595.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-18
- Publication Date
- 2024-11-29
- Estimated Expiration
- 2044-09-18
AI Technical Summary
During the CAD design process, the tile transformation operation is inefficient, especially complex operations require users to be familiar with the cumbersome operation process, which is not friendly to beginners.
The training method of intelligent tile transformation recommendation model is adopted. By obtaining user operation sequences, filtering and machine learning, the tile transformation recommendation model is generated, reducing user manual input, and improving operation efficiency.
It improves the efficiency of tile transformation operations and the correlation between the model and drawing characteristics, reduces the difficulty of learning of machine learning, improves the effectiveness and universality of the model, and reduces the impact of users' subjective operations.
Smart Images

Figure CN118839387B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of computer-aided design (CAD), and in particular to a training method, system, medium and program for an intelligent block transformation recommendation model. Background Art
[0002] In the CAD design process, block transformation is one of the commonly used operations, including scaling, rotation, mirroring, copying, etc. Traditional block transformation operations require users to manually select blocks and then select the corresponding transformation command. The operation steps are cumbersome and inefficient.
[0003] Especially for some complex block transformation operations, whether through the UI or through commands, the operation is relatively complicated and requires users to be familiar with the corresponding operation process, which is not very friendly to beginners.
[0004] Therefore, there is an urgent need for a method that can improve the efficiency of tile change operations. Summary of the invention
[0005] The purpose of the present invention is to provide a training method, system, medium and program for an intelligent block transformation recommendation model, which can partially solve or alleviate the above-mentioned deficiencies in the prior art and improve the correlation between the model and the drawing characteristics of the drawing itself.
[0006] In order to solve the above-mentioned technical problems, the present invention specifically adopts the following technical solutions:
[0007] A first aspect of the present invention is to provide a training method for an intelligent block transformation recommendation model, comprising the steps of:
[0008] S100: Acquire multiple groups of operation sequences generated when at least one user performs transformation operations on multiple tiles, wherein the tiles include: at least one entity, and the operation sequence includes: (i) a command name corresponding to the tile; (ii) at least one branch command name associated with the command name; (iii) a parameter group for describing the branch command name, the parameter group including: (a) an entity; and (b) a characteristic number and / or a vector; and when there are multiple branch command names in the operation sequence, the branch command names are recorded and arranged in the operation sequence according to the order in which they are generated during the transformation operation;
[0009] S101: performing a first filtering operation on a plurality of groups of operation sequences; wherein S101 includes:
[0010] Identify the branch command name belonging to a first operation category in at least one of the operation sequences, wherein the first operation category includes one or more of the following: create entity, close entity, save entity, zoom in entity, zoom out entity, copy entity;
[0011] Deleting the branch command name and the parameter group associated with the branch command name in the corresponding operation sequence;
[0012] S102: Using multiple groups of the operation sequences after the filtering operation to form a training set, and using the training set to perform machine learning to learn at least one block transformation recommendation model; wherein the command name and the branch command name are used as input data for machine learning, and the parameter group is used as output data for the machine learning.
[0013] In some embodiments, before S102, the step of:
[0014] S103: performing a second filtering operation on the plurality of groups of operation sequences; wherein S103 includes:
[0015] S31: Acquire at least one intermediate instruction sequence of the operation sequence, wherein the intermediate instruction sequence is located between a first branch command name and a second branch command name; wherein the first branch command name and the second branch command name both correspond to the same entity, and the entities corresponding to the first branch command name and the second branch command name are located at different positions in the drawing file;
[0016] S32: using a set filtering model to calculate a filtering index according to the length of the intermediate instruction sequence and a set length threshold;
[0017] Wherein, the filtering model is: ;
[0018] Alternatively, the filtering model is: ;
[0019] S33: When the filtering index is greater than the set filtering threshold, the corresponding intermediate instruction sequence is deleted to obtain the new operation sequence.
[0020] In some embodiments, before S102, the step further includes: S104: performing a third filtering operation on the plurality of groups of the operation sequences; wherein S104 includes:
[0021] S41: Obtain the first priority instruction type;
[0022] S42: dividing the operation sequence including the first priority instruction type into a first data set, and dividing the remaining operation sequences into a second data set;
[0023] S43: Acquire a first number of the operation sequences from the first data set, and acquire a second number of the operation sequences from the second data set, wherein the first number is greater than the second number;
[0024] S44: forming the training set according to the operation sequence acquired in S43.
[0025] In some embodiments, S41 includes the steps of:
[0026] Prioritize the branch command names according to the frequencies of occurrence of the branch command names in the multiple groups of operation sequences;
[0027] At least one branch command name having a priority level greater than a set level is used as a first priority instruction type.
[0028] In some embodiments, the first priority instruction type is set by a user.
[0029] In some embodiments, S102 includes the steps of:
[0030] S21: Screening out a first sample combination having a plurality of the operation sequences from the training set;
[0031] S22: performing machine learning on the first sample combination to obtain the block transformation recommendation model;
[0032] S23: Selecting a plurality of the operation sequences that are not used for machine learning from the training set as a second sample combination, and using the second sample combination to verify the recommendation accuracy of the tile transformation recommendation model.
[0033] In some embodiments, when obtaining at least two recommended block transformation models, the step of:
[0034] When receiving a real-time operation sequence input by a user, outputting a recommendation result according to the real-time operation sequence through at least two of the tile transformation recommendation models;
[0035] receiving and recording the selection frequency of the recommendation results of the at least two image block transformation recommendation models by the user;
[0036] Determining the recommendation priority of the at least two image block transformation recommendation models according to the selection frequency;
[0037] The recommendation results at the next moment are displayed differently according to the recommendation priority.
[0038] The present invention also provides a training system for an intelligent block transformation recommendation model, comprising:
[0039] A sequence acquisition module, used to acquire multiple groups of operation sequences generated when at least one user performs transformation operations on multiple blocks, wherein the blocks include: at least one entity, and the operation sequence includes: (i) a command name corresponding to the block; (ii) at least one branch command name associated with the command name; (iii) a parameter group for describing the branch command name, wherein the parameter group includes: (a) an entity; and (b) a characteristic number and / or a vector; and when there are multiple branch command names in the operation sequence, the branch command names are recorded and arranged in the operation sequence according to the order in which they are generated during the transformation operation;
[0040] A first filtering module is configured to perform a first filtering operation on the plurality of operation sequences; wherein the first filtering module comprises:
[0041] an identification unit, configured to identify the branch command name belonging to a first operation category in at least one of the operation sequences, wherein the first operation category includes one or more of the following: creating an entity, closing an entity, saving an entity, enlarging an entity, reducing an entity, and copying an entity;
[0042] a deleting unit, used to delete the branch command name and the parameter group associated with the branch command name in the corresponding operation sequence;
[0043] A training module is used to form a training set using multiple groups of operation sequences after filtering operations, and use the training set to perform machine learning to learn at least one block transformation recommendation model; wherein the command name and the branch command name are used as input data for machine learning, and the parameter group is used as output data for the machine learning.
[0044] The present invention also provides a computer storage medium storing one or more instructions, which, when executed by one or more computers, enable the one or more computers to execute the training method of the intelligent tile transformation recommendation model as described in any one of the embodiments.
[0045] The present invention also provides a computer program product, comprising computer-readable instructions, which, when executed on a computer device, enable the computer device to execute the training method for the intelligent tile transformation recommendation model as described in any one of the embodiments.
[0046] Beneficial technical effects:
[0047] For application scenarios with extremely high repeatability of block transformation, the present invention preferably provides a method for targeted prediction of parameter groups for block transformation. Specifically, for the massive operating data of the CAD background, the present invention focuses on selecting operation sequences that can reflect the user's operation logic (and the sequence can record the operation order between each branch command) as learning samples, and selects to focus on predicting parameter groups through operation sequences, thereby reducing the learning difficulty of the machine learning process (reducing sample processing and learning workload). At the same time, for application scenarios with extremely high repeatability of parameter groups, the strong correlation between such operation sequences with a certain length and parameter groups is also conducive to improving the effectiveness of model training.
[0048] In other words, the present invention provides a method for obtaining a model based on training of operation sequences with strong correlation. It is worth noting that the "strong correlation" here has at least two meanings: 1. The logical correlation of the internal data of the operation sequence is strong, and 2. The logical correlation of the operation sequence and the drawing characteristics of the drawing itself is strong.
[0049] Furthermore, before machine learning is performed, subjective operations such as closing, saving, and zooming in will be further deleted to reduce the impact of the user's personalized subjective operations on the machine learning process, so that the inherent association of the trained model is more in line with the drawing characteristics of the drawings themselves (such as architectural design specifications, etc.). Moreover, the elimination of such personalized data can actually improve the versatility of the model among different users.
[0050] Furthermore, by deleting the tentative operations, the influence of the subjective operations of the drafting engineer on the machine learning can be further reduced, such as improving the logical association between the command sequence (command name and multiple branch commands recorded in sequence) and the parameter group in the operation sequence (avoiding the problem of overfitting). In other words, filtering the tentative operations can make the reflected logical association strongly associated with the drawing program itself, reducing the influence of the user's subjective operation on the training process to a certain extent. BRIEF DESCRIPTION OF THE DRAWINGS
[0051] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, each element or part is not necessarily drawn according to the actual scale. Obviously, the drawings described below are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings without paying creative labor.
[0052] Figure 1a is a schematic diagram of a method flow in an exemplary embodiment of the present invention;
[0053] Figure 1b is a schematic diagram of a method flow in another exemplary embodiment of the present invention;
[0054] Figure 2 is a schematic diagram of an operation interface in an exemplary embodiment of the present invention;
[0055] Figure 3a is a schematic diagram of system modules in an exemplary embodiment of the present invention;
[0056] Figure 3b It is a schematic diagram of system modules in an exemplary embodiment of the present invention. DETAILED DESCRIPTION
[0057] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.
[0058] Herein, suffixes such as "module", "component" or "unit" used to represent elements are only used to facilitate the description of the present invention, and have no specific meanings by themselves. Therefore, "module", "component" or "unit" can be used mixedly.
[0059] In this document, the terms "upper", "lower", "inner", "outer", "front", "back", "one end", "the other end" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance.
[0060] In this document, unless otherwise clearly specified and limited, the terms "installed", "provided with", "connected", etc. should be understood in a broad sense. For example, "connected" can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection, a direct connection, or an indirect connection through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.
[0061] Herein "and / or" includes any and all combinations of one or more of the associated listed items.
[0062] Herein, "plurality" means two or more than two, ie, it includes two, three, four, five, etc.
[0063] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.
[0064] As used in this specification, the term "about" typically means + / - 5% of the stated value, more typically + / - 4% of the stated value, more typically + / - 3% of the stated value, more typically + / - 2% of the stated value, even more typically + / - 1% of the stated value, and even more typically + / - 0.5% of the stated value.
[0065] In this specification, some embodiments may be disclosed in a format of being in a range. It should be understood that this description of "being in a range" is only for convenience and brevity, and should not be interpreted as a rigid limitation on the disclosed range. Therefore, the description of the range should be considered to have specifically disclosed all possible sub-ranges and independent numerical values within this range. For example, the description of the range 1-6 should be considered to have specifically disclosed sub-ranges such as from 1 to 3, from 1 to 4, from 1 to 5, from 2 to 4, from 2 to 6, from 3 to 6, etc., as well as individual numbers within this range, such as 1, 2, 3, 4, 5 and 6. Regardless of the breadth of the range, the above rules apply.
[0066] Definition of noun:
[0067] "Entity": In this article, an entity refers to the graphic data on a CAD file that is displayed on the CAD drawing interface. An entity has attributes, which are data values that control specific visual characteristics of an entity or element (such as visibility, color, and line style). In different embodiments, an entity may also be referred to as a "pixel" or "element".
[0068] "Block": In this article, a block is a general term for a group of entities, each of which can have its own layer, line type, color and other characteristics. For example, in the application process, CAD operates a block as an independent and complete object. Users can insert blocks into any specified position at a certain scale and angle as needed. For example, blocks can include one or more of the following types: nested blocks (a nested block is a block containing other blocks, also known as an internal block), external blocks, dynamic blocks, etc.
[0069] "File" is also called "drawing file" or "drawing": In this article, file refers to various files that can be run in the CAD system and are used to draw, edit, modify, store, and view CAD drawings. Common CAD file formats include, but are not limited to, DWG, DXF, DWT, DWF, DWL, DWS, DWX, MNU, MNC, MNL, MNS, CUI, CUIX, SHX, PAT, LIN, CTB, STB, PLT, PC3, etc.
[0070] In existing CAD operations, when a user performs a block transformation operation, the user often needs to manually select the block and manually input the corresponding transformation command. However, the operation efficiency of this manual input method is very low.
[0071] In recent years, with the continuous development of the construction industry, manufacturing industry and other industries involving the use of drawing software, the amount of information and graphics drawn using various types of drawing software, especially CAD software, are larger than before. For example, a CAD drawing file drawn using CAD is usually very large, containing several or even dozens of design drawings, and also includes multiple layouts, which will significantly increase the drawing workload of the drawing engineer.
[0072] Correspondingly, the present invention proposes a method for predicting parameter groups according to operation sequences for mapping application scenarios with extremely high drawing repetitiveness, so as to reduce the repetitive workload of mapping engineers and improve mapping efficiency.
[0073] Embodiment 1
[0074] See also Figure 1a As shown, the present invention provides a training method for an intelligent block transformation recommendation model, comprising the steps of:
[0075] S100: Acquire multiple groups of operation sequences generated when at least one user (e.g., a graphics engineer) performs transformation operations on multiple blocks, wherein the blocks include: at least one entity, and the operation sequences include: (i) a command name corresponding to the block; (ii) at least one branch command name associated with the command name; (iii) a parameter group for describing the branch command name, and the parameter group includes: (a) an entity; and (b) a characteristic number and / or vector;
[0076] For example, for different tile transformation operations received by a computer system that executes the transformation operation, the branch command name corresponding to the tile transformation is recorded.
[0077] Preferably, when there are multiple branch command names in the operation sequence, the branch command names are recorded and arranged in the operation sequence in the order in which they are generated during the transformation operation process;
[0078] S102: using a plurality of groups of the operation sequences (preferably operation sequences obtained after performing at least one filtering operation) to form a training set, and using the training set to perform machine learning to learn and obtain at least one block transformation recommendation model (also referred to as: model);
[0079] For example, the command name and the branch command name are used as input data for machine learning, and the parameter group is used as output data for the machine learning.
[0080] In some embodiments, one or more of the following machine learning methods may be used for model training: support vector machine (SVM), random forest (RF), deep learning, etc.
[0081] In actual application, when the block transformation recommendation model receives a sequence of operations, it can output a corresponding predicted parameter group. When the user believes that the parameter group meets the current drawing requirements, the parameter group setting can be completed by simply clicking to select it without continuing manual input.
[0082] For application scenarios with extremely high repeatability of block transformation, the present invention preferably provides a method for targeted prediction of parameter groups for block transformation. Specifically, for the massive operating data of the CAD background, the present invention focuses on selecting operation sequences that can reflect the user's operation logic (and the sequence can record the operation order between each branch command) as learning samples, and selects to focus on predicting parameter groups through operation sequences, thereby reducing the learning difficulty of the machine learning process (reducing sample processing and learning workload). At the same time, for application scenarios with extremely high repeatability of parameter groups, the strong correlation between such operation sequences with a certain length and parameter groups is also conducive to improving the effectiveness of model training.
[0083] It can be understood that the command name refers to an instruction for creating a block or an entity in a block. For example, when the computer system receives an instruction to create a rectangle (equivalent to a block or entity), the command name recorded is a rectangle. Correspondingly, when the computer system receives further operations on the rectangle (such as size editing, rotation, movement, etc. of the rectangle), the branch command name will be recorded accordingly (such as the branch command name is size editing, rotation, movement, etc.). Specifically, when the branch command name is size editing, a parameter group is recorded accordingly, which includes: feature numbers (i.e., the length and width data used to describe the rectangle), entity (i.e., the name, ID, etc. of the rectangle, used to bind the parameter group to the branch command name). For another example, when the branch command name is movement, a parameter group is recorded accordingly, which includes: feature numbers (used to describe the distance of movement), vectors (used to describe the direction of movement), and entities. Among them, feature numbers and vectors are necessary data for implementing transformation operations such as branch commands. For another example, when the computer system receives an instruction to create a straight line (equivalent to a block or entity), the command name recorded accordingly is a straight line. Correspondingly, when the user edits the dimension of a line, the computer system will continue to receive and record the branch command name - dimension editing, and the parameter group: characteristic number (ie, the length of the line), entity (such as the name of the line).
[0084] In other words, the command name corresponds to a drawing instruction in CAD. For example, CIRCLE is a circle drawing instruction; in the process of creating and editing a circle, multiple branch commands such as move, rotate, and size edit may also be generated (i.e., instructions for transforming the circle). It can be understood that branch commands are multiple branch steps recorded in the process from the creation of a block to the end of the operation. The command is the entry point for creation, and the branch command is the conditional judgment and branch flow formed by the process from the entry point to the final result.
[0085] In other words, in some embodiments, the characteristic numbers / vectors are data (also referred to as geometric characteristic data) that are specific to the corresponding geometric figure (i.e., entity, such as a circle or a line). For example, when the entity is a circle, the geometric characteristic data may include one or more of the following: (1) characteristic numbers: coordinate position of the center of the circle (center X, Y, Z coordinates), radius, diameter, circumference, or area; (2) vectors: normal coordinates (normal X, Y, Z coordinates). For another example, when the entity is an arc, the geometric characteristic data includes one or more of the following: (1) characteristic numbers: starting point coordinates (starting point X, Y, Z coordinates), center coordinates (center X, Y, Z coordinates), endpoint coordinates (endpoint X, Y, Z coordinates), starting point angle, endpoint angle, total angle, arc length, or area; (2) vectors: normal coordinates (normal X, Y, Z coordinates).
[0086] See also Figure 2 As shown, taking the process of drawing a circle as an example, the application of the block transformation recommendation model is exemplarily described:
[0087] 1) The user selects the circle drawing function on the panel (the computer system receives the command name - circle at this time);
[0088] 2) CAD prompts the user to select the center of the circle or enter different branch commands;
[0089] 3) If the user enters the button corresponding to the branch command, the CAD prompt will switch to two-point circle drawing or three-point circle drawing (it will be different depending on the user's input. For example, when entering three-point circle drawing, the computer system will receive the branch command name - three-point circle drawing);
[0090] 4) The user continues to click the screen (or enter coordinates), and the clicked point is the center of the circle;
[0091] 5) After CAD obtains the center of the circle input by the user, it further prompts the user to enter the radius;
[0092] 6) The user drags the mouse on the screen or directly enters numbers;
[0093] 7) CAD recognizes the user's input and interprets it as a certain radius;
[0094] 8) The CAD software draws a circle on the screen.
[0095] Specifically, in this process, when the user's instruction operation is completed, the following symbol sequence (that is, operation sequence) will be generated:
[0096] Circle b_center (0,0) b_radius 100
[0097] Here Circle represents the command name, b_center and b_radius are branch command names, and (0,0) and 100 are parameters.
[0098] For another example, in some embodiments, during the machine learning process, the command name and at least a portion of the branch command names in a set of operation sequences are used as input data for the machine learning, and another portion of the branch command names and the parameter group are used as output data for the machine learning. Thus, the tile transformation recommendation model trained in the present invention can also predict the next tile transformation operation and the parameter group required for the tile transformation based on the user's current operation records.
[0099] Furthermore, for highly repetitive operation scenarios, in order to strengthen the correlation between samples, the present invention can also filter operations with weak logic through filtering operations. For example, in some embodiments, before S102, the steps are also included:
[0100] S101: performing a first filtering operation on a plurality of groups of operation sequences; wherein S101 includes:
[0101] Identify the branch command name belonging to a first operation category in at least one of the operation sequences, wherein the first operation category includes one or more of the following: create entity, close entity, save entity, zoom in entity, zoom out entity, copy entity;
[0102] Deleting the corresponding branch command name and the parameter group associated with the branch command name in the corresponding operation sequence;
[0103] In this embodiment, before machine learning is performed, subjective operations such as closing, saving, and zooming in will be further deleted to reduce the impact of the user's personalized subjective operations on the machine learning process, so that the inherent association of the trained model is more in line with the drawing characteristics of the drawings themselves (such as architectural design specifications, etc.). Moreover, the elimination of such personalized data can improve the versatility of the model among different users.
[0104] For example, in some embodiments, a second operation category may be set, and the second operation category includes one or more of the following: rotation, mirroring, size editing, movement, etc. Preferably, the branch command names belonging to the second operation category are all retained as training samples for machine learning.
[0105] In some embodiments, before S102, the step of:
[0106] S103: performing a second filtering operation on the plurality of groups of operation sequences; wherein S103 includes:
[0107] S31: Acquire at least one intermediate instruction sequence of the operation sequence, wherein the intermediate instruction sequence is located between a first branch command name and a second branch command name; wherein the first branch command name and the second branch command name both correspond to the same entity, and the entities corresponding to the first branch command name and the second branch command name are located at different positions in the drawing file;
[0108] For example, in some embodiments, when the distance between the coordinates of the entities corresponding to the first branch command name and the second branch command name is greater than a set distance, the two are considered to be in different positions.
[0109] For example, when the user sets the entity at position A at the first moment T1, the first branch command name is recorded accordingly; when the user moves the entity from position A to position B at the second moment T2 and performs at least one tile transformation operation at position B, multiple branch command instructions are also recorded; then, the user moves the entity from position B to position A at the third moment T3; wherein, in the process from the second moment to before the third moment, multiple branch command names are recorded in sequence according to the order of generation, and the last branch command name is identified as the second branch command name, and the remaining branch command names are identified as intermediate instruction sequences;
[0110] S32: using a set filtering model to calculate a filtering index according to the length of the intermediate instruction sequence and a set length threshold;
[0111] Wherein, the filtering model is: ;
[0112] Alternatively, the filtering model is: ;
[0113] S33: When the filtering index is greater than the set filtering threshold, the corresponding intermediate instruction sequence is deleted (or the intermediate instruction sequence and the second branch command name are deleted) to obtain the new operation sequence.
[0114] For example, in some embodiments, when the filtering index is greater than 30%, it is considered necessary to filter the relevant data.
[0115] For example, in the actual mapping process, the mapping engineer may generate some tentative map block transformation operations, which will be identified as repetitive samples in machine learning. For example, when the mapping engineer sets a window on the west wall of the room (the first branch command name will be recorded at this time), the mapping engineer may then try to move the window to the east wall, and successively perform a series of edits on the window (the multiple branch command names recorded thereby form an intermediate instruction sequence, as well as a second branch command name), and finally restore the window to the west wall. At this point, when the intermediate instruction sequence is too long, it is considered that there are too many tentative operations, and in this embodiment, the sequences generated by these tentative operations are preferably partially deleted.
[0116] The second filtering operation can also delete operation sequences with too high similarity. For example, during the drawing process, the user may modify the same entity repeatedly and undo / redo multiple times. These samples repeated multiple times in a short period of time will affect the training and easily cause overfitting of the training on specific samples.
[0117] In this embodiment, by deleting the trial operations, the influence of the subjective operation of the drawing engineer on the machine learning can be further reduced, such as improving the logical association between the command sequence (command name and multiple branch commands recorded in sequence) and the parameter group in the operation sequence. In other words, filtering the trial operations can make the reflected logical association strongly associated with the drawing program itself, and to a certain extent, reduce the influence of the user's personal operation on the training process.
[0118] In some embodiments, before S102, the step further includes: S104: performing a third filtering operation on the plurality of groups of the operation sequences; wherein S104 includes:
[0119] S41: Obtain the first priority instruction type;
[0120] S42: dividing the operation sequence including the first priority instruction type into a first data set, and dividing the remaining operation sequences into a second data set;
[0121] S43: Acquire a first number of the operation sequences from the first data set, and acquire a second number of the operation sequences from the second data set, wherein the first number is greater than the second number;
[0122] S44: forming the training set according to the operation sequence acquired in S43.
[0123] In this embodiment, by extracting the data set differently based on the instruction type, the model can be further bound to a specific drawing scene. For example, for drawing different types of 3D models such as high-rise buildings, large mechanical equipment, and large bridge buildings, the correlation between the model and the drawing specifications of the specific model can be enhanced by collecting the instruction type by classification.
[0124] In some embodiments, the steps include:
[0125] Prioritize the branch command names according to the frequencies of occurrence of the branch command names in the multiple groups of operation sequences;
[0126] At least one branch command name having a priority level greater than a first set level is used as a first priority instruction type.
[0127] For another example, in some embodiments, the priority instruction type may also be set by the user in combination with different scenario requirements.
[0128] Furthermore, in some embodiments, the second data set may be further divided into two or more data sets according to the priority instruction type.
[0129] For example, in some embodiments, a second priority instruction type is further provided, and the second priority instruction type is a branch command name whose priority is less than or equal to the first set level and greater than the second set level. The second data set is split in such a way that the operation sequence containing the branch command name belonging to the second priority instruction type is divided into the first subset, and the remaining operation sequences are divided into the second subset.
[0130] In some embodiments, the first priority instruction type is set by a user.
[0131] In some embodiments, S102 includes the steps of:
[0132] S21: Screening out a first sample combination having a plurality of the operation sequences from the training set;
[0133] S22: performing machine learning on the first sample combination to obtain a block transformation recommendation model;
[0134] S23: Selecting a plurality of the operation sequences that are not used for machine learning from the training set as a second sample combination, and using the second sample combination to verify the recommendation accuracy of the block transformation recommendation model. When the difference between the recommendation result (such as the output parameter group) and the actual parameter group recorded in the second sample combination is small, it is considered that the currently trained model is relatively reliable.
[0135] Moreover, through the cross-validation process, it can be proved that the model provided by the present invention, which is trained based on the operation sequence with strong correlation, has high reliability. It is worth noting that the "strong correlation" here has at least two meanings: 1. The logical correlation between the internal data of the operation sequence is strong, and 2. The logical correlation between the operation sequence and the drawing characteristics of the drawing itself is strong.
[0136] To further illustrate the advantages of this application, an exemplary application scenario is used as an example below:
[0137] Application scenario: Drawing of high-rise buildings
[0138] The inventors noticed that in the process of drawing high-rise buildings, drafting engineers usually need to draw the areas of each floor of the high-rise building in sequence. In addition, the internal layout of each floor is often highly repetitive (for example, the location layout of each room is similar on different floors, and for example, the water supply and drainage pipe routing, circuit routing, door or window distribution inside the room are similar, which makes the drawing operations of drafting engineers highly repetitive, such as rotation, copying, mirroring, etc. The frequency of operations will be very high), but not completely consistent. At this time, drafting engineers are required to manually draw the internal house layout of each floor one by one. However, this drawing method is very inefficient; and when the drawing time is too long, drafting engineers are also very likely to make erroneous operations due to fatigue, such as causing errors in the location of doors and windows, and these internal detail differences are also difficult to detect during the entire drawing and review process.
[0139] That is to say, in this kind of high-repetition drawing scenario, not only do drawing engineers need to spend a very long time working, but the drawing accuracy is also easily reduced.
[0140] On the contrary, when the recommendation model trained by the present application is used, for drawing scenarios with highly repetitive operations, when the operation sequence currently input by the draftsman into the CAD (such as command names and multiple branch command names) is collected, the recommended transformation operations can be predicted for the draftsman.
[0141] Therefore, for highly repetitive transformation operations, the drafting engineer can directly select the recommended transformation operation with one click, eliminating the cumbersome manual operation process. Therefore, the present invention can greatly reduce the drawing workload of the drafting engineer for the drawing of large buildings. At the same time, by reducing this repetitive workload, it is also beneficial for the drafting engineer to focus more on the core area of the building, reasonably allocate the drawing energy in different areas, and thus improve the overall drawing efficiency and accuracy.
[0142] Embodiment 2
[0143] The present invention also provides a model updating method, which can optimize the recommendation degree of multiple models in real time. For example, in some embodiments, see Figure 1b As shown, when obtaining at least two of the block transformation recommendation models, the steps include:
[0144] S201: when receiving a real-time operation sequence input by a user, outputting a recommendation result according to the real-time operation sequence through at least two of the block transformation recommendation models;
[0145] S202: receiving and recording the selection frequency of the recommendation results of the at least two image block transformation recommendation models by the user in a first time period;
[0146] S203: Determine the recommendation priority of the at least two image block transformation recommendation models according to the selection frequency;
[0147] For example, different tile transformation recommendation models are set to the first recommendation priority and the second recommendation priority in sequence according to the magnitude of the selection frequency.
[0148] S204: Display the recommendation result (such as parameter group and / or branch command name) at the next moment differently according to the recommendation priority.
[0149] For example, in some embodiments, the recommendation results may be displayed in order according to the recommendation priority. For example, recommendation results of different recommendation priorities are displayed and recommended in different display modes.
[0150] For example, in some embodiments, the recommendation results are provided in the form of boxes (such as selection boxes, pictures, etc.), and the boxes of the recommendation results with the first recommendation priority are displayed in a first size, and the boxes of the remaining recommendation results are displayed in a second size, and the first size is larger than the second size. Furthermore, when there are multiple recommendation priorities, more different sizes can be used for differential display.
[0151] For another example, in some embodiments, different display modes may also be: displaying through different colors, displaying through different brightness, and so on.
[0152] For example, in some embodiments, different tile transformation recommendation models may be trained using different machine learning methods.
[0153] For another example, in some embodiments, different tile transformation recommendation models may be trained using different sample data sets (eg, the data sets may be taken from different types of drawings).
[0154] For example, in some embodiments, at least two recommendation results may be displayed in sequence on the interface, and the currently selected recommendation result may be determined based on a selection instruction (such as a click operation) input by the user, and when the corresponding recommendation result is selected, the selection frequency of the corresponding model is also increased by 1.
[0155] In some embodiments, the user may also select a recommended result and modify the recommended result, thereby completing the input of the parameter group only by adjusting local data.
[0156] The model updating method provided by the present invention can further display the recommendation results in a differentiated manner in combination with the actual drawing scene (such as the user's personal operating habits, the drawing specifications referenced by the current drawing), thereby performing personalized screening and display of the recommendation results. In addition, the special sample-strong correlation operation sequence adopted by the present invention can reduce the difficulty of machine learning and reduce the complexity of the output model to a certain extent, which also makes the multi-model recommendation form adopted by the present invention easier to implement (i.e., reducing the pressure of the computer system's input data collection and model calculation process).
[0157] The model in this embodiment can be trained using any of the methods described above.
[0158] Furthermore, after running for a period of time, if the recommendation accuracy of the block transformation recommendation model in the second period is less than the preset first accuracy, the operation sequence collected in the second period may be used to update the block transformation recommendation model.
[0159] Similarly, since the update adopted by the present invention only needs to use a strongly correlated operation sequence (and preferably can be pre-processed through the above-mentioned filtering operation) for local update, this can further reduce the real-time update pressure of multiple models and reduce the pressure on users.
[0160] Embodiment 3
[0161] The present invention also provides a training system for an intelligent block transformation recommendation model, see Figure 3a As shown, including:
[0162] The sequence acquisition module 10 is used to acquire multiple groups of operation sequences generated when at least one user performs transformation operations on multiple blocks, wherein the blocks include: at least one entity, and the operation sequence includes: (i) a command name corresponding to the block; (ii) at least one branch command name associated with the command name; (iii) a parameter group for describing the branch command name, and the parameter group includes: (a) an entity; and (b) a characteristic number and / or a vector; and when there are multiple branch command names in the operation sequence, the branch command names are recorded and arranged in the operation sequence according to the order in which they are generated during the transformation operation process;
[0163] The training module 20 is used to form a training set using multiple groups of the operation sequences (preferably the operation sequences after the filtering operation), and use the training set to perform machine learning to learn at least one block transformation recommendation model; wherein the command name and the branch command name are used as input data for machine learning, and the parameter group is used as output data for the machine learning.
[0164] Furthermore, it further comprises: a first filtering module 30, configured to perform a first filtering operation on the plurality of groups of the operation sequences; wherein the first filtering module 30 comprises:
[0165] an identification unit, configured to identify the branch command name belonging to a first operation category in at least one of the operation sequences, wherein the first operation category includes one or more of the following: creating an entity, closing an entity, saving an entity, enlarging an entity, reducing an entity, and copying an entity;
[0166] a deleting unit, used to delete the branch command name and the parameter group associated with the branch command name in the corresponding operation sequence;
[0167] In some embodiments, the system further includes: a second filtering module 40, configured to perform a second filtering operation on the plurality of groups of operation sequences; wherein the second filtering module 40 includes:
[0168] an acquisition unit, configured to acquire at least one intermediate instruction sequence of the operation sequence, wherein the intermediate instruction sequence is located between a first branch command name and a second branch command name; wherein the first branch command name and the second branch command name both correspond to the same entity, and the entities corresponding to the first branch command name and the second branch command name are located at different positions in the drawing file;
[0169] A filtering index calculation unit, used to calculate the filtering index by using a set filtering model through the length of the intermediate instruction sequence and a set length threshold;
[0170] Wherein, the filtering model is: ;
[0171] Alternatively, the filtering model is: ;
[0172] The determination unit is used to delete the corresponding intermediate instruction sequence when the filtering index is greater than a set filtering threshold, so as to obtain the new operation sequence.
[0173] In some embodiments, the system includes: a third filtering module 50, configured to perform a third filtering operation on a plurality of groups of the operation sequences; wherein the third filtering module 50 includes:
[0174] An instruction type acquisition unit, used for acquiring a first priority instruction type;
[0175] a data set division unit, used for dividing the operation sequence including the first priority instruction type into a first data set, and dividing the remaining operation sequences into a second data set;
[0176] a data set acquisition unit, configured to acquire a first number of the operation sequences from the first data set, and acquire a second number of the operation sequences from the second data set, wherein the first number is greater than the second number;
[0177] The training set forming unit is used to form the training set according to the operation sequence acquired in the data set acquiring unit.
[0178] Furthermore, the instruction type acquisition unit is used to prioritize the branch command names according to the frequency of occurrence of the branch command names in the multiple groups of operation sequences; and to take at least one branch command name with a priority greater than a set level as the first priority instruction type.
[0179] In some embodiments, it also includes: a verification module 60, which is used to screen out a first sample combination having multiple operation sequences from the training set; perform machine learning on the first sample combination to obtain the tile transformation recommendation model; select multiple operation sequences that are not used for machine learning from the training set as second sample combinations, and use the second sample combination to verify the recommendation accuracy of the tile transformation recommendation model.
[0180] Correspondingly, see Figure 3b As shown, the present invention also provides a model updating system, the system can obtain at least two block transformation recommendation models, and correspondingly the system includes:
[0181] The output module 100 is configured to, when receiving a real-time operation sequence input by a user, output a recommendation result according to the real-time operation sequence through at least two of the block transformation recommendation models;
[0182] A recording module 200, configured to receive and record the selection frequency of the recommendation results of the at least two image block transformation recommendation models by the user;
[0183] A recommendation module 300, configured to determine the recommendation priority of the at least two image block transformation recommendation models according to the selection frequency;
[0184] The display module 400 is used to display the recommendation results at the next moment differently according to the recommendation priority.
[0185] The present invention can be widely used in mechanical design, architectural design, industrial design and other fields, and can improve the operational efficiency of block transformation and optimize user experience.
[0186] It is understandable that the model updating system of the present invention may also include one or more of the above modules, and the included modules can implement any of the above methods or steps, which will not be repeated here.
[0187] In some embodiments, the execution order of each filtering operation in the present invention can be flexibly set by the user in combination with the actual learning scenario. Another aspect of the present invention also provides a computer storage medium, which stores one or more instructions, and when the instructions are executed by one or more computers, the one or more computers execute the training method or update method of the intelligent tile transformation recommendation model in any of the above embodiments.
[0188] The present invention also provides a computer program product, including computer-readable instructions, which, when executed on a computer device, enable the computer device to execute a training method or an updating method for the intelligent tile transformation recommendation model in any of the above embodiments.
[0189] In some embodiments, the user may refer to a drafting engineer, a software designer, or a computer system.
[0190] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for a computer terminal (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present invention.
[0191] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the enlightenment of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present invention and the claims, which all fall within the protection of the present invention.
Claims
1. A training method for an intelligent block transformation recommendation model, characterized in that: Includes steps: S100: Acquire multiple groups of operation sequences generated when at least one user performs transformation operations on multiple tiles, wherein the tiles include: at least one entity, and the operation sequence includes: (i) a command name corresponding to the tile; (ii) at least one branch command name associated with the command name; (iii) a parameter group for describing the branch command name, the parameter group including: (a) an entity; and (b) a characteristic number and / or a vector; and when there are multiple branch command names in the operation sequence, the branch command names are recorded and arranged in the operation sequence according to the order in which they are generated during the transformation operation; S101: performing a first filtering operation on a plurality of groups of operation sequences; wherein S101 includes: Identify the branch command name belonging to a first operation category in at least one of the operation sequences, wherein the first operation category includes one or more of the following: create entity, close entity, save entity, zoom in entity, zoom out entity, copy entity; Deleting the corresponding branch command name and the parameter group associated with the branch command name in the corresponding operation sequence; S102: Using multiple groups of the operation sequences after corresponding filtering operations to form a training set, and using the training set to perform machine learning to learn at least one block transformation recommendation model; wherein the command name and the branch command name are used as input data for machine learning, and the parameter group is used as output data for the machine learning.
2. The training method of the intelligent block transformation recommendation model according to claim 1, characterized in that: Before S102, the step further includes: S103: performing a second filtering operation on the plurality of groups of operation sequences; wherein S103 includes: S31: Acquire at least one intermediate instruction sequence of the operation sequence, wherein the intermediate instruction sequence is located between a first branch command name and a second branch command name; wherein the first branch command name and the second branch command name both correspond to the same entity, and the entities corresponding to the first branch command name and the second branch command name are located at different positions in the drawing file; S32: using a set filtering model to calculate a filtering index according to the length of the intermediate instruction sequence and a set length threshold; Wherein, the filtering model is: ; Alternatively, the filtering model is: ; S33: When the filtering index is greater than the set filtering threshold, the corresponding intermediate instruction sequence is deleted to obtain the new operation sequence.
3. The training method of the intelligent block transformation recommendation model according to claim 1, characterized in that: Before S102, the method further includes the following step: S104: performing a third filtering operation on the plurality of groups of the operation sequences; wherein S104 includes: S41: Obtain the first priority instruction type; S42: dividing the operation sequence including the first priority instruction type into a first data set, and dividing the remaining operation sequences into a second data set; S43: Acquire a first number of the operation sequences from the first data set, and acquire a second number of the operation sequences from the second data set, wherein the first number is greater than the second number; S44: forming the training set according to the operation sequence acquired in S43.
4. The training method of the intelligent block transformation recommendation model according to claim 3, characterized in that: S41 comprises the steps of: Prioritize the branch command names according to the frequencies of occurrence of the branch command names in the multiple groups of operation sequences; At least one branch command name having a priority level greater than a set level is used as a first priority instruction type.
5. The training method of the intelligent block transformation recommendation model according to claim 3, characterized in that: The first priority instruction type is set by a user.
6. A training method for an intelligent block transformation recommendation model according to any one of claims 1 to 5, characterized in that: S102 includes the steps of: S21: Screening out a first sample combination having a plurality of the operation sequences from the training set; S22: performing machine learning on the first sample combination to obtain the block transformation recommendation model; S23: Selecting a plurality of the operation sequences that are not used for machine learning from the training set as a second sample combination, and using the second sample combination to verify the recommendation accuracy of the tile transformation recommendation model.
7. The training method of the intelligent block transformation recommendation model according to claim 1, characterized in that: When obtaining at least two of the block transformation recommendation models, the step further includes: When receiving the real-time operation sequence input by the user, outputting a recommendation result according to the real-time operation sequence through at least two of the tile transformation recommendation models; receiving and recording the selection frequency of the recommendation results of at least two image block transformation recommendation models by the user; Determining the recommendation priority of at least two of the block transformation recommendation models according to the selection frequency; The recommendation results at the next moment are displayed differently according to the recommendation priority.
8. A training system for an intelligent block transformation recommendation model, characterized in that: include: A sequence acquisition module, used to acquire multiple groups of operation sequences generated when at least one user performs transformation operations on multiple blocks, wherein the blocks include: at least one entity, and the operation sequence includes: (i) a command name corresponding to the block; (ii) at least one branch command name associated with the command name; (iii) a parameter group for describing the branch command name, wherein the parameter group includes: (a) an entity; and (b) a characteristic number and / or a vector; and when there are multiple branch command names in the operation sequence, the branch command names are recorded and arranged in the operation sequence according to the order in which they are generated during the transformation operation; A first filtering module is configured to perform a first filtering operation on the plurality of operation sequences; wherein the first filtering module comprises: an identification unit, configured to identify the branch command name belonging to a first operation category in at least one of the operation sequences, wherein the first operation category includes one or more of the following: creating an entity, closing an entity, saving an entity, enlarging an entity, reducing an entity, and copying an entity; A deleting unit, used to delete the corresponding branch command name and the parameter group associated with the branch command name in the corresponding operation sequence; A training module is used to form a training set using multiple groups of operation sequences after corresponding filtering operations, and use the training set to perform machine learning to learn at least one block transformation recommendation model; wherein the command name and the branch command name are used as input data for machine learning, and the parameter group is used as output data for the machine learning.
9. A computer storage medium, characterized in that The computer storage medium stores one or more instructions, which, when executed by one or more computers, enable the one or more computers to perform the training method of the intelligent tile transformation recommendation model as described in any one of claims 1 to 7.
10. A computer program product, characterized in that The method comprises computer-readable instructions, which, when executed on a computer device, enable the computer device to execute the training method of the intelligent tile transformation recommendation model as claimed in any one of claims 1 to 7.
Citation Information
Patent Citations
Product recommendation method and device based on federated learning, computer equipment and medium
CN112287244A
Graphics Processing
GB202204012D0
