CN114385876B - Model search space generation method, device and system - Google Patents
Model search space generation method, device and system Download PDFInfo
- Publication number
- CN114385876B CN114385876B CN202210037063.9A CN202210037063A CN114385876B CN 114385876 B CN114385876 B CN 114385876B CN 202210037063 A CN202210037063 A CN 202210037063A CN 114385876 B CN114385876 B CN 114385876B
- Authority
- CN
- China
- Prior art keywords
- model
- training
- search space
- information
- recommendation
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 82
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 41
- 238000012216 screening Methods 0.000 claims description 15
- 238000004590 computer program Methods 0.000 claims description 13
- 230000004044 response Effects 0.000 claims 2
- 238000004364 calculation method Methods 0.000 abstract description 3
- 238000013145 classification model Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/903—Querying
- G06F16/9035—Filtering based on additional data, e.g. user or group profiles
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/903—Querying
- G06F16/9038—Presentation of query results
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Databases & Information Systems (AREA)
- Computational Linguistics (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Evolutionary Biology (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The embodiment of the invention provides a method, a device and a system for generating a model search space. The method comprises the following steps: acquiring modeling task information, wherein the modeling task information comprises target data set information; processing the characteristic information of the data set by using a pre-trained recommendation model for recommending a model training method to obtain a recommendation result; generating a model search space according to the recommendation result; the recommendation result comprises a feature engineering strategy and a model training algorithm, and the model search space comprises: operator module space and operator parameter space. According to the embodiment of the invention, the model search space generated according to the recommendation result has convergence, and mass calculation on the model search space is not needed, so that the resource consumption of a computer for model training is reduced. In addition, because the model search space has convergence, the time for the modeling tool to select the characteristic processing strategy, the model algorithm and the corresponding hyperparameter from the model search space is greatly reduced, so that the parameter search efficiency in the model training process is improved.
Description
Technical Field
The invention relates to the technical field of artificial intelligence, in particular to a method, a device and a system for generating a model search space.
Background
With the improvement of social informatization and intelligence level, the method for training the business model by using the modeling tool and realizing intelligent processing of big data business by using the trained business model also gradually becomes a general means of big data industry.
In the existing model training process, a huge search space is usually defined, and the search space usually comprises a plurality of processing feature strategies, model algorithms and corresponding hyper-parameters. The modeling tool performs model training from the search space search feature processing strategy, the model algorithm, and the corresponding hyper-parameters. However, the search space is very large, which results in time-consuming process of selecting the feature processing strategy, the model algorithm and the corresponding hyper-parameters from the search space by the modeling tool, and thus a computer for model training needs to consume large resources.
Disclosure of Invention
The embodiment of the invention provides a method, a device and a system for generating a model search space, which can solve the problem that the process of selecting a characteristic processing strategy, a model algorithm and a corresponding hyper-parameter from the search space by the existing modeling tool is time-consuming, so that a computer for model training needs to consume larger resources.
In order to solve the above technical problem, the present invention provides a method for generating a model search space, the method comprising:
acquiring modeling task information, wherein the modeling task information comprises target data set information;
processing the target data set information by using a pre-trained recommendation model for recommending a model training method to obtain a recommendation result;
generating a model search space according to the recommendation result;
the recommendation result comprises a feature engineering strategy and a model training algorithm, and the model search space comprises: operator module space and operator parameter space.
Optionally, in the above method, before the step of processing the target data set information by using a pre-trained recommendation model for a recommendation model training method to obtain a recommendation result, the method further includes:
obtaining a modeling data wide table to obtain sample data for training the recommendation model;
and performing model training according to the sample data to generate the recommendation model.
Optionally, in the foregoing method, the obtaining of the wide table of modeling data to obtain sample data for training the recommendation model includes:
responding to a first model training completion message, and acquiring data set information for training the first model and training information of the first model;
and obtaining a modeling data wide table according to the data set information and the training information corresponding to the first model.
Optionally, in the foregoing method, the training information includes: the system comprises operator module information and operator parameters corresponding to operators, wherein the operator module comprises a characteristic engineering strategy module and a model training module;
the step of obtaining a modeling data wide table according to the data set information and the training information corresponding to the first model comprises the following steps:
merging the data set information, the operator module information and operator parameters corresponding to operators into a training record;
and generating the modeling data wide table according to the training record.
Optionally, in the above method, after the step of generating the model search space according to the recommendation result, the method further includes:
searching in the model search space based on the target data set information to obtain a first search result;
a business model is created based on the first search result.
Optionally, in the above method, the modeling task information further includes: searching parameters and/or modeling scenarios, after the step of generating a model search space according to the recommendation result, the method further comprising:
searching in the search space based on the target data set information and the search parameter and/or the modeling scene to obtain a second search result;
and creating a business model based on the second search result.
Optionally, the method further includes:
after the business model is completed, target model training information obtained by the business model is obtained;
updating the modeling data wide table according to the target data set information and the target model training information;
and adjusting the recommendation model according to the updated modeling data wide table.
Optionally, in the foregoing method, the step of generating a model search space according to the recommendation result includes:
screening the recommendation result according to the data type of the operator parameter in the recommendation result to obtain a screening result;
generating the model search space based on the screening results.
The embodiment of the invention also provides a device for generating the model search space, which comprises:
the modeling system comprises a first acquisition module, a second acquisition module and a third acquisition module, wherein the first acquisition module is used for acquiring modeling task information which comprises target data set information;
the processing module is used for processing the target data set information by using a pre-trained recommendation model for recommending a model training method to obtain a recommendation result;
the first generation module is used for generating a model search space according to the recommendation result;
the recommendation result comprises a feature engineering strategy and a model training algorithm, and the model search space comprises: operator module space and operator parameter space.
Optionally, the apparatus further comprises:
the second acquisition module is used for acquiring a modeling data wide table to obtain sample data for training the recommendation model;
and the second generation module is used for carrying out model training according to the sample data to generate the recommendation model.
Optionally, in the above apparatus, the second obtaining module includes:
the obtaining submodule is used for responding to a first model training completion message and obtaining data set information used for training the first model and training information of the first model;
and the processing submodule is used for obtaining a modeling data wide table according to the data set information and the training information corresponding to the first model.
Optionally, in the above apparatus, the training information includes: the system comprises operator module information and operator parameters corresponding to operators, wherein the operator module comprises a characteristic engineering strategy module and a model training module;
the processing submodule comprises:
the merging unit is used for merging the data set information, the operator module information and the operator parameters corresponding to the operators into a training record;
and the generating unit is used for generating the modeling data wide table according to the training record.
Optionally, the apparatus further comprises:
the first searching module is used for searching in the model searching space based on the target data set information to obtain a first searching result;
and the first creating module is used for creating a business model based on the first search result.
Optionally, in the foregoing apparatus, the modeling task information further includes: searching for parameters and/or modeling scenarios, the apparatus further comprising:
the second searching module is used for searching in the searching space based on the target data set information and the searching parameters and/or the modeling scenes to obtain a second searching result;
and the second creating module is used for creating a business model based on the second search result.
Optionally, the apparatus further comprises:
the third acquisition module is used for acquiring target model training information obtained by the business model task after the business model is completed;
the updating module is used for updating the modeling data wide table according to the target data set information and the target model training information;
and the adjusting module is used for adjusting the recommendation model according to the updated modeling data wide table.
Optionally, in the above apparatus, the first generating module includes:
the screening submodule is used for screening the recommendation result according to the data type of the operator parameter in the recommendation result to obtain a screening result;
and the generation submodule is used for generating the model search space based on the screening result.
The embodiment of the present invention further provides a system for generating a model search space, which includes a processor, a memory, and a computer program stored in the memory and capable of running on the processor, where the computer program, when executed by the processor, implements the steps of the method for generating a model search space as described above.
An embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored, where the computer program, when executed by a processor, implements the steps of the model search space generation method as described above.
The method comprises the steps of processing target data set information by using a pre-trained recommendation model for a recommendation model training method to obtain a recommendation result; and generating a model search space according to the recommendation result, wherein the model search space generated according to the recommendation result has convergence, and mass calculation of the model search space is not needed, so that resource consumption of a computer for model training is reduced. In addition, because the model search space has convergence, the time for the modeling tool to select the characteristic processing strategy, the model algorithm and the corresponding hyper-parameters from the model search space is greatly reduced, so that the parameter search efficiency in the model training process is improved.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required to be used in the description of the embodiments of the present invention will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without inventive exercise.
FIG. 1 is a flow chart of a method for generating a model search space according to an embodiment of the present invention;
fig. 2 is a block diagram of a model search space generation apparatus according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Referring to fig. 1, fig. 1 is a flowchart of a model search space generation method according to an embodiment of the present invention, and as shown in fig. 1, the model search space generation method includes the following steps:
Here, the modeling task information may include a target data set for which a user inputs to recommend a model for prediction, and target data set information is acquired by analyzing the target data set. Wherein the target data set information includes, but is not limited to, at least one of: characteristic information of the data set and a scene corresponding to the data set. The characteristic information of the target data set includes at least one of: the method comprises the following steps of task type, total row number, total column number, category characteristic column number, continuous characteristic column number, date characteristic column number, text characteristic column number, category characteristic column average missing value rate, continuous characteristic column average missing value rate, date characteristic column average missing value rate, text characteristic column average missing value rate, category characteristic different value total number, maximum category characteristic different value number, data set file size and other characteristic information. Additionally, the target data set includes, but is not limited to, an image data set, a text data set, and a speech data set.
And 102, processing the target data set information by using a pre-trained recommendation model for recommending a model training method to obtain a recommendation result.
The method comprises the following steps of carrying out predictive analysis on target data set information by using a pre-trained recommendation model for a recommendation model training method to obtain a recommendation result of the recommendation model, wherein the recommendation result comprises the following steps: characteristic engineering strategies, model training algorithms and parameters corresponding to the algorithms. Optionally, before the step 102 of processing the target data set information by using a pre-trained recommendation model for a recommendation model training method to obtain a recommendation result, the method further includes:
obtaining a modeling data wide table to obtain sample data for training the recommendation model;
and performing model training according to the sample data to generate the recommendation model.
The modeling data wide table contains historical modeling data, and specifically comprises the following steps: data set information used by the training model, operator modules used by the training model, operator parameters corresponding to the operator models, and the like. And carrying out model training based on the modeling data to obtain a recommendation model for recommending a model training method. The recommendation model may include a multi-classification model, a regression model, or a neural network model, which is not particularly limited in the present invention.
The training process for an exemplary recommendation model is as follows: one data record in the wide table comprises two parts, namely, data set information, operator modules used by the training models, operator parameters corresponding to the operator models and other model training information, wherein the operator modules comprise but are not limited to: the device comprises a preprocessing module, a characteristic engineering strategy module and a model training module. And taking the data set information as a characteristic column X of a recommended model training sample, taking model training information such as operator modules used by the training model and operator parameters corresponding to each operator module as a target column Y of the recommended model training sample, and inputting the training sample into a machine learning model for training to obtain a final recommended model.
Optionally, the step of obtaining a wide table of modeling data to obtain sample data for training the recommended model includes:
responding to a first model training completion message, and acquiring data set information for training the first model and training information of the first model;
and obtaining a modeling data wide table according to the data set information and the training information corresponding to the first model.
Wherein the training information comprises: operator module information and operator parameters corresponding to each operator,
the step of obtaining a modeling data wide table according to the data set information and the training information corresponding to the first model comprises the following steps:
combining the data set information, the operator module information and operator parameters corresponding to the operators into a training record;
and generating the modeling data wide table according to the training record.
Before the first model training, analyzing the characteristics of the data set used for training the first model, for example, after clearing invalid columns such as id column and constant column for a table data set, at least one of the following information is counted: the method comprises the steps of obtaining feature information of a data set by using the features of task type, total row number, total column number, category feature column number, continuous feature column number, date feature column number, text feature column number, category feature column average missing value rate, continuous feature column average missing value rate, date feature column average missing value rate, text feature column average missing value rate, category feature different value total number, maximum category feature different value number, data set file size and the like, and obtaining the feature information of the data set so as to obtain the data set information. In addition, the data set can be analyzed to determine the scene corresponding to the first model, so as to obtain scene information and obtain the data set information.
After a data set and/or a user scene for model training are obtained, searching is carried out in a preset search space according to feature information of the data set and/or the user scene, and a feature processing strategy, a model training algorithm and algorithm parameters are obtained. Combining the feature processing strategy, the model training algorithm and the algorithm parameters, then operating the data set by using a first model training task based on the first model training task obtained by combining, and obtaining training information (namely model training information) corresponding to the first model training task after the model training is finished, wherein the model training information comprises but is not limited to: the system comprises a feature processing strategy, a model training algorithm and corresponding algorithm parameters. Combining the characteristic processing strategy, the model training algorithm, corresponding algorithm parameters and data set information (including characteristic information and/or scene information of the data set) records for training into a training record; and generating the modeling data wide table according to the training record.
For example, the feature information of the data set is referred to as a "data set feature column", and these data can be obtained and stored by analyzing the data set before training the model. And after the modeling is finished, counting the characteristic engineering strategies and algorithms finally selected by the modeling task and the parameters of the strategies and the algorithms. Let it be assumed that the following feature engineering strategies, algorithms and corresponding hyper-parameters are defined in the default search space (which may be understood as the pre-set feature engineering strategy modules, operators and corresponding parameter configurations within the data processing system) as follows:
missing value filling
■ strand (filling strategy)
● standardization
● LightGBM algorithm
■ boosting _ type (lifting algorithm type)
■ n _ estimators (number of sub-classifiers)
XGboost algorithm
■ boost (lifting algorithm type)
■ max _ depth (maximum tree depth)
The modules used in the model training process and the corresponding hyper-parameters are recorded as the following fields:
whether to use missing value padding
Whether or not to use normalization
Whether the LightGBM algorithm is used or not
Whether XGboost algorithm is used
Stream of missing value fill Module
Boosting _ type of LightGBM algorithm
N _ estimators of LightGBM Algorithm
Boost of XGboost algorithm
Max _ depth of XGboost algorithm
The above data columns are referred to as "search space data columns" and these data can be obtained and stored after the modeling task is completed. The "search space data column" and "data set feature column" are grouped into one record (data is a sample):
TABLE 1
When the model training task is sufficient, the data in the table can be used to train the model to generate the search space. Components of the search space: operator module space and operator parameter space. The operator module space comprises modules in the search space, such as a feature engineering strategy module, an operator module and the like, and the operator parameter space comprises parameters corresponding to the operator module.
And 103, generating a model search space according to the recommendation result.
Wherein the model search space comprises: operator module space and operator parameter space.
Optionally, the step 103 of generating a model search space according to the recommendation result includes:
screening the recommendation result according to the data type of the parameter in the recommendation result;
and generating the model search space based on the screened recommendation result.
Specifically, in order to further improve the convergence of the generated model search space and ensure the accuracy of the search while reducing the search time, the invention provides a feasible method, specifically, if the data type of the operator parameter is an integer type, the parameter of the preset bit before the sorting is selected according to a preset method; and if the data type of the operator parameters is Boolean type or floating point type, selecting a result with operator parameter values floating up and down to a specified preset range as the screened recommended result. The preset bit or the preset range may be a preset numerical value or may be set by the user.
For example, for a model training task, whether a module is added to the search space is determined, and a model may be trained by using the "feature column of the data set" as a feature and using the "use of the module" in the "data column of the search space" as a target column. It should be noted that training is not required for each model training task, and training may be performed periodically. The 'data set characteristic column' obtained by calculating the data set trained by the new model is used as X input to the model for prediction, whether the module is used or not is determined according to the prediction result, and optionally, the confidence rate of the model can be properly adjusted to be low so as to search for enough modules in the space.
Taking whether a 'standardization' module needs to be added into a search space or not as an example, taking a 'data set characteristic column' as a characteristic, taking 'whether standardization is used' in the 'search space data column' as a target column, training a model by using the LigthGBM, extracting data of the 'data set characteristic column' of the current data set, sending the data into a recommendation model for prediction, and determining whether the 'standardization' module needs to be added according to a prediction result.
After the modules in the search space are determined, a parameter space is selected for the modules, and a process similar to the selection of the modules is characterized by a data set characteristic column and a model is trained by taking the parameter in the search space data column as a target column. It should be noted that training is not required for each modeling task, and training may be performed periodically. By way of example: assuming that the "LightGBM algorithm" has been selected in the search space, the parameter space of boosting _ type is now selected, which has 4 values: gbdt, rf, dart, goss. And training a multi-classification model by taking the 'data set feature column' as a feature and the 'boosting _ type' of the LightGBM algorithm module in the 'search space data column' as a target column. Then, the data of the "data set feature column" extracted from the data set of the current task is used as X for model prediction, for example, the probabilities of the prediction results of gbdt, rf, dart and goss are 0.1, 0.2, 0.5 and 0.3 respectively, 2 values with the highest probability in the prediction results, namely dart and goss, can be taken to be added into the search space as the candidate parameter of the "boosting _ type of the LightGBM algorithm module", and the number of the generated parameters can be determined according to the running time and the resources accepted by the user. For example, if the user can accept more time, 75% of the parameters can be selected as the candidate parameters in the order from high to low in probability, that is, 3 parameters of dart, goss, and rf are selected to be added to the search space, and if it is desired to use less resources and time, the top 25% or 50% of the result of the probability ordering can be selected as the candidate parameters.
Similarly, if the parameters to be selected are of the continuous type, a regression model may be trained. Assuming that the "LightGBM algorithm" is already selected in the search space, the parameter space of "n _ estimators" is now selected, and a regression model is trained with the "data set feature column" as the feature and the "n _ estimators of the LightGBM algorithm module" in the "search space data column" as the target column. The result predicted by the model is a continuous type value, and the result can be shifted upwards by 50% and downwards by 50%, where 50% can be adjusted according to the size of the search space acceptable to the user, for example, if the predicted result is 100, then 3 parameters can be set for n _ estimators in the search space: [50,100,150].
Optionally, after the step of generating a model search space according to the recommendation result, the method further includes:
searching in the model search space based on the information of the target data set to obtain a first search result;
and creating a business model based on the search result.
Specifically, after generating the model search space, a search is performed in the search space based on the information of the target data set input by the user in step 101, so as to obtain a search result, where the search result includes an operator module and corresponding parameters. And establishing a business model based on the operator modules and the corresponding parameters obtained in the search space.
Optionally, after the step 101 of generating a model search space according to the recommendation result, the modeling task information further includes: searching parameters and/or modeling scenarios, after the step of generating a model search space according to the recommendation result, the method further comprising:
searching in the search space based on the target data set information and the search parameters and/or the modeling scene to obtain a second search result;
and creating a business model based on the second search result.
Specifically, the search parameter includes information such as search times, scene, search time, and the like. And searching in a search space according to the target data set information and the search parameters to obtain a search result, wherein the search result comprises an operator module and corresponding parameters. And establishing a business model based on the operator module obtained in the search space and the corresponding parameters.
Optionally, the method further includes:
after the business model is completed, target model training information obtained by the business model is obtained;
updating the modeling data wide table according to the target data set information and the target model training information;
and adjusting the recommendation model according to the updated modeling data wide table.
Specifically, when modeling training is performed by using a business model, target model training information obtained by the business model is obtained. And updating the modeling data wide table based on the target data set information and the target model training information corresponding to the target model training task. And after the modeling data wide table is updated, the sample data of the training recommendation model is updated, and then the recommendation model is retrained based on the updated sample data.
According to the embodiment of the invention, the model search space generated according to the recommendation result has convergence, and mass calculation on the model search space is not needed, so that the resource consumption of a computer for model training is reduced. In addition, because the model search space has convergence, the time for the modeling tool to select the characteristic processing strategy, the model algorithm and the corresponding hyperparameter from the model search space is greatly reduced, so that the parameter search efficiency in the model training process is improved.
Based on the method for generating a model search space provided in the above embodiment, an embodiment of the present invention further provides a device for generating a model search space for implementing the method, and referring to fig. 2, the device 200 for generating a model search space provided in an embodiment of the present invention includes:
a first obtaining module 201, configured to obtain modeling task information, where the modeling task information includes target data set information;
the processing module 202 is configured to process the target data set information by using a pre-trained recommendation model for recommending a model training method to obtain a recommendation result;
a first generating module 203, configured to generate a model search space according to the recommendation result;
the recommendation result comprises a feature engineering strategy and a model training algorithm, and the model search space comprises: operator module space and operator parameter space.
Optionally, the apparatus further comprises:
the second acquisition module is used for acquiring a modeling data wide table to obtain sample data for training the recommendation model;
and the second generation module is used for carrying out model training according to the sample data to generate the recommendation model.
Optionally, in the above apparatus, the second obtaining module includes:
the acquisition sub-module is used for responding to a first model training completion message and acquiring data set information used for training the first model and training information of the first model;
and the processing submodule is used for obtaining a modeling data wide table according to the data set information and the training information corresponding to the first model.
Optionally, in the above apparatus, the training information includes: the system comprises operator module information and operator parameters corresponding to operators, wherein the operator module comprises a characteristic engineering strategy module and a model training module;
the processing submodule comprises:
the merging unit is used for merging the data set information, the operator module information and the operator parameters corresponding to the operators into a training record;
and the generating unit is used for generating the modeling data wide table according to the training record.
Optionally, the apparatus further comprises:
the first searching module is used for searching in the model searching space based on the information of the target data set to obtain a first searching result;
and the first creating module is used for creating a business model based on the first search result.
Optionally, in the above apparatus, the modeling task information further includes: search parameters and/or model scenarios, the apparatus further comprising:
the second searching module is used for searching in the searching space based on the target data set information and the searching parameters and/or the modeling scenes to obtain a second searching result;
and the second creating module is used for creating a business model based on the second search result.
Optionally, the apparatus further comprises:
the third acquisition module is used for acquiring target model training information obtained by the business model after the business model is completed;
the updating module is used for updating the modeling data wide table according to the target data set information and the target model training information;
and the adjusting module is used for adjusting the recommendation model according to the updated modeling data wide table.
Optionally, in the above apparatus, the first generating module includes:
the screening submodule is used for screening the recommendation result according to the data type of the operator parameter in the recommendation result to obtain a screening result;
and the generation submodule is used for generating the model search space based on the screening result.
The embodiment of the present invention further provides a model search space generation system, which includes a processor, a memory, and a computer program stored in the memory and capable of running on the processor, and when the computer program is executed by the processor, the steps of the model search space generation method are implemented.
An embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored, where the computer program, when executed by a processor, implements the steps of the model search space generation method as described above.
The embodiment of the present invention further provides a readable storage medium, where a computer program is stored, and when the computer program is executed by a processor, the computer program implements each process of the above embodiment of the model search space generation method, and can achieve the same technical effect, and in order to avoid repetition, the details are not repeated here. The computer-readable storage medium may be a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk.
It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.
Through the above description of the embodiments, those skilled in the art will clearly understand that the method of the above embodiments can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware, but in many cases, the former is a better implementation manner. Based on such understanding, the technical solutions of the present invention or portions thereof contributing to the prior art may be embodied in the form of a software product, which is stored in a storage medium (such as ROM/RAM, magnetic disk, optical disk) and includes instructions for enabling a terminal (such as a mobile phone, a computer, a server, an air conditioner, or a network device) to execute the methods according to the embodiments of the present invention.
The above description is only for the specific embodiments of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art can easily conceive of the changes or substitutions within the technical scope of the present invention, and all the changes or substitutions should be covered within the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.
Claims (8)
1. A method for model search space generation, the method comprising:
acquiring modeling task information, wherein the modeling task information comprises target data set information;
processing the target data set information by using a pre-trained recommendation model for recommending a model training method to obtain a recommendation result;
generating a model search space according to the recommendation result;
the recommendation result comprises a feature engineering strategy and a model training algorithm, and the model search space comprises: an operator module space and an operator parameter space;
before the step of processing the target data set information by using the pre-trained recommendation model for the recommendation model training method to obtain the recommendation result, the method further comprises:
obtaining a modeling data wide table to obtain sample data for training the recommendation model; the modeling data wide table comprises historical modeling data;
performing model training according to the sample data to generate the recommendation model;
the step of obtaining the wide table of modeling data and obtaining the sample data for training the recommendation model comprises:
in response to a first model training completion message, acquiring data set information for training the first model and training information of the first model, the training information including: operator module information and operator parameters corresponding to each operator;
merging the data set information, the operator module information and operator parameters corresponding to operators into a training record;
and generating the modeling data wide table according to the training record.
2. The model search space generation method of claim 1, wherein the operator module comprises a feature engineering strategy module and a model training module.
3. The method of generating a model search space according to claim 1, wherein after the step of generating a model search space according to the recommendation, the method further comprises:
searching in the model search space based on the target data set information to obtain a first search result;
a business model is created based on the first search result.
4. The method of generating a model search space according to claim 1, wherein the modeling task information further includes: searching parameters and/or modeling scenarios, after the step of generating a model search space according to the recommendation result, the method further comprising:
searching in the search space based on the target data set information and the search parameter and/or the modeling scene to obtain a second search result;
and creating a business model based on the second search result.
5. The model search space generation method of claim 3 or 4, further comprising:
after the business model is completed, target model training information obtained by the business model is obtained;
updating the modeling data wide table according to the target data set information and the target model training information;
and adjusting the recommendation model according to the updated modeling data wide table.
6. The method of generating a model search space according to claim 1, wherein the step of generating a model search space according to the recommendation result includes:
screening the recommendation result according to the data type of the operator parameter in the recommendation result to obtain a screening result;
generating the model search space based on the screening results.
7. An apparatus for model search space generation, the apparatus comprising:
the system comprises a first acquisition module, a second acquisition module and a third acquisition module, wherein the first acquisition module is used for acquiring modeling task information which comprises target data set information;
the processing module is used for processing the target data set information by using a pre-trained recommendation model for recommending a model training method to obtain a recommendation result;
the first generation module is used for generating a model search space according to the recommendation result;
the recommendation result comprises a feature engineering strategy and a model training algorithm, and the model search space comprises: an operator module space and an operator parameter space;
the second acquisition module is used for acquiring a modeling data wide table to obtain sample data for training the recommendation model; the modeling data wide table comprises historical modeling data;
the second generation module is used for carrying out model training according to the sample data and generating the recommendation model;
the second acquisition module includes: an acquisition sub-module and a processing sub-module,
the obtaining sub-module is configured to, in response to a first model training completion message, obtain data set information used for training the first model and training information of the first model, where the training information includes: operator module information and operator parameters corresponding to each operator;
the processing submodule comprises:
the merging unit is used for merging the data set information, the operator module information and the operator parameters corresponding to the operators into a training record;
and the generating unit is used for generating the modeling data wide table according to the training record.
8. A model search space generation system comprising a processor, a memory and a computer program stored on the memory and executable on the processor, the computer program when executed by the processor implementing the steps of the model search space generation method of any one of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210037063.9A CN114385876B (en) | 2022-01-13 | 2022-01-13 | Model search space generation method, device and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210037063.9A CN114385876B (en) | 2022-01-13 | 2022-01-13 | Model search space generation method, device and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114385876A CN114385876A (en) | 2022-04-22 |
| CN114385876B true CN114385876B (en) | 2022-09-13 |
Family
ID=81202258
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210037063.9A Active CN114385876B (en) | 2022-01-13 | 2022-01-13 | Model search space generation method, device and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114385876B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114663662B (en) * | 2022-05-23 | 2022-09-09 | 深圳思谋信息科技有限公司 | Hyper-parameter searching method, device, computer equipment and storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8356000B1 (en) * | 2000-04-13 | 2013-01-15 | John R. Koza | Method and apparatus for designing structures |
| CN112149838A (en) * | 2020-09-03 | 2020-12-29 | 第四范式（北京）技术有限公司 | Method, device, electronic equipment and storage medium for realizing automatic model building |
| CN112541595A (en) * | 2020-12-30 | 2021-03-23 | 中国建设银行股份有限公司 | Model construction method and device, storage medium and electronic equipment |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107507068A (en) * | 2017-09-02 | 2017-12-22 | 广东奡风科技股份有限公司 | A kind of financial product real-time recommendation method based on random forests algorithm |
| CN110457333B (en) * | 2019-07-29 | 2021-10-22 | 招商局金融科技有限公司 | Data real-time updating method and device and computer readable storage medium |
| CN110991658A (en) * | 2019-11-28 | 2020-04-10 | 重庆紫光华山智安科技有限公司 | Model training method and device, electronic equipment and computer readable storage medium |
| CN113869521A (en) * | 2020-06-30 | 2021-12-31 | 华为技术有限公司 | Method, device, computing equipment and storage medium for constructing prediction model |
| CN113128702B (en) * | 2021-04-15 | 2024-11-19 | 杭州电子科技大学 | A neural network adaptive distributed parallel training method based on reinforcement learning |
-
2022
- 2022-01-13 CN CN202210037063.9A patent/CN114385876B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8356000B1 (en) * | 2000-04-13 | 2013-01-15 | John R. Koza | Method and apparatus for designing structures |
| CN112149838A (en) * | 2020-09-03 | 2020-12-29 | 第四范式（北京）技术有限公司 | Method, device, electronic equipment and storage medium for realizing automatic model building |
| CN112541595A (en) * | 2020-12-30 | 2021-03-23 | 中国建设银行股份有限公司 | Model construction method and device, storage medium and electronic equipment |
Non-Patent Citations (1)
| Title |
|---|
| 面向网络空间表达的数据模型设计及处理方法;王映雪等;《测绘科学技术学报》;20191215(第06期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114385876A (en) | 2022-04-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111914085B (en) | Text fine granularity emotion classification method, system, device and storage medium | |
| CN111611488B (en) | Information recommendation method and device based on artificial intelligence and electronic equipment | |
| CN111815432B (en) | Financial service risk prediction method and device | |
| CN112418320B (en) | Enterprise association relation identification method, device and storage medium | |
| CN111582341B (en) | User abnormal operation prediction method and device | |
| CN109376061A (en) | A kind of information processing method and system | |
| CN118036756B (en) | Method, device, computer equipment and storage medium for large model multi-round dialogue | |
| CN112699945A (en) | Data labeling method and device, storage medium and electronic device | |
| CN114385876B (en) | Model search space generation method, device and system | |
| CN116341059A (en) | Tunnel intelligent design method based on similarity | |
| CN111539390A (en) | Small target image identification method, equipment and system based on Yolov3 | |
| CN112200275B (en) | Artificial neural network quantification method and device | |
| CN112130896B (en) | Neural network model migration method and device, electronic equipment and storage medium | |
| CN114238402A (en) | Alarm data processing method and device, storage medium and electronic equipment | |
| CN115660695A (en) | Customer service personnel label portrait construction method and device, electronic equipment and storage medium | |
| CN115600818A (en) | Multi-dimensional scoring method and device, electronic equipment and storage medium | |
| CN118070916A (en) | Marketing scene prediction model training method and device, electronic equipment and storage medium | |
| CN113688989A (en) | Deep learning network acceleration method, device, equipment and storage medium | |
| CN111061774A (en) | Method, device, electronic device and storage medium for judging accuracy of search results | |
| CN119025684B (en) | A multi-caliber data management method, device and equipment based on knowledge graph | |
| CN110569436A (en) | network media news recommendation method based on high-dimensional auxiliary information | |
| CN116341990B (en) | Knowledge management evaluation method and system for infrastructure engineering | |
| CN116226527B (en) | Digital community treatment method for realizing behavior prediction through resident big data | |
| CN119273438A (en) | Risk control strategy mining method, device, equipment and storage medium based on large model | |
| CN117216530A (en) | Model information determining method, device and system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |