KR20230103616A - System and Method for Recommending Deep Learning Framework based on Feature Selection and Scoring Algorithms - Google Patents

Abstract

The present invention receives user specifications as an input and recommends the optimal deep learning framework for implementing the user's algorithm, so that the optimal deep learning framework suitable for the learning target can be applied. A feature selection and scoring algorithm It relates to a system and method for recommending a deep learning framework using a framework property/feature input unit for inputting model template and framework property information for framework selection; a user inputting an input deep learning model to be converted and data property information to be used. A specification input unit; a model data analysis unit and a framework selection unit; a deep learning framework recommendation unit that receives framework properties/characteristics and user specifications and recommends a deep learning framework using a feature selection and scoring algorithm; a deep learning framework It includes; a recommendation framework information output unit that outputs recommended framework information and framework optimization model design recommended by the recommendation unit.

Description

System and Method for Recommending Deep Learning Framework based on Feature Selection and Scoring Algorithms}

The present invention relates to selection of a deep learning framework, and specifically, it is possible to recommend an optimal deep learning framework for implementing a user's algorithm by receiving a user specification as an input, thereby providing an optimal deep learning framework suitable for a learning target. It relates to a deep learning framework recommendation system and method using a feature selection and scoring algorithm that can be applied.

Deep learning is also called deep learning, which performs machine learning learning using artificial neural networks (ANNs) with multiple layers. Therefore, deep learning is not a completely different concept from machine learning, but a type of machine learning.

In machine learning, a person had to analyze and determine which feature to extract from among the various features of the data to be learned, but in deep learning, the machine automatically learns by extracting features from the data to be learned. As such, the biggest difference between deep learning and machine learning can be seen in whether or not the machine learns on its own.

Therefore, deep learning can be defined as a technology in which machines automatically learn important patterns and rules from large-scale data and make decisions or predictions based on them.

Numerous algorithms of various types, such as DNN, CNN, RNN, RBM, and DBN, have been developed and used for artificial neural network algorithms used in deep learning. In many cases, two or more algorithms are mixed and used to solve a single problem. lost.

It is very inefficient to have to implement a new implementation every time an already proven algorithm is used.

The deep learning framework provides a number of libraries that have already been verified and various deep learning algorithms that have been pre-trained, allowing developers to use them quickly and easily.

This frees developers from the exhausting work of implementing redundant functions and helps them focus on developing core algorithms to solve problems.

Built using neural networks, deep learning is inspired by the structure and function of the brain. Deep learning is a tool that drives revolutionary advances in AI.

Most deep learning frameworks, such as PyTorch, Tensorflow, ONNX, etc. use directed acyclic graphs to represent neural networks.

To use deep learning effectively, we need to define algorithms for model functions such as predicting and identifying objects, and we need a large amount of data to train the model for better accuracy.

There are many frameworks for such deep learning, and each framework has a different purpose.

Tensorflow is an open source platform developed by Google. It is JavaScript-based and provides various tools for building and deploying models in the browser.

The built model can be deployed to mobile and embedded devices, and supports large-scale production environments.

PyTorch is an open-source platform based on the Torch library developed by Facebook that aims to accelerate the entire process from research prototyping to production deployment, and is suitable for applications such as natural language processing and computer vision.

ONNX is a project jointly developed by Microsoft and Facebook. It is a system designed for machine learning and deep learning model development, provides access to hardware optimization, and provides a hardware system-wide performance optimization library.

Keras is an open source framework that runs on Tensorflow, Theano, Microsoft Cognitive Toolkit, and PlaidML. It provides high speed by providing data parallelism, and it is easy to accelerate model data training time.

Sonnet, developed by DeepMind, is built on top of Tensorflow, provides advanced libraries for building complex neural network structures, and develops and creates basic Python objects corresponding to specific parts of a neural network.

MXNET is an open source framework that enables fast model training and is highly scalable. It supports multiple programming languages such as C++, Python, Julia, Matlab, and JavaScript, and supports multiple GPUs with fast context switching.

Gluon is an open source framework that provides a concise API for defining machine learning and deep learning models by leveraging optimized neural network components.

It is suitable for developing and testing neural network models, and is suitable for developers new to machine learning. In addition, there are various deep learning frameworks, and each deep learning framework has different advantages. Among them, there is a need to select and present the optimal deep learning framework according to the user's purpose.

Deep learning frameworks can produce good results by selecting a framework that is suitable for the target to be analyzed or predicted.

Therefore, there is a demand for the development of useful technologies for selecting and applying the optimal deep learning framework for a desired learning target.

Republic of Korea Patent Publication No. 10-2021-0123110 Republic of Korea Patent No. 10-2257028

The present invention is to solve the problem of the prior art deep learning framework selection technology, and to recommend the optimal deep learning framework for implementing the user's algorithm by receiving user specifications as input, The purpose of this study is to provide a deep learning framework recommendation system and method using a feature selection and scoring algorithm that enables efficient application of the deep learning framework.

The present invention is a deep learning framework using a feature selection and scoring algorithm that outputs a recommended deep learning framework and framework optimization model design most suitable for the user's algorithm so that the user can select the optimal deep learning framework according to his or her purpose. Its purpose is to provide a recommendation system and method.

The present invention is a feature selection and scoring algorithm that allows users to easily build algorithms by using built-in functions of frameworks, thereby saving time in implementing algorithms and verifying whether recommended frameworks can be used in the target environment. The purpose is to provide a deep learning framework recommendation system and method using

Other objects of the present invention are not limited to the above-mentioned objects, and other objects not mentioned above will be clearly understood by those skilled in the art from the description below.

To achieve the above object, a deep learning framework recommendation system using a feature selection and scoring algorithm according to the present invention includes a model template for framework selection and a framework property/feature input unit for inputting framework property information; A user specification input unit that inputs the input deep learning model and data property information to be used; a model data analysis unit and a framework selection unit, which receive framework properties/characteristics and user specifications, and recommend deep learning frameworks using feature selection and scoring algorithms. A deep learning framework recommendation unit that performs; a recommendation framework information output unit that outputs recommended framework information and framework optimization model design recommended by the deep learning framework recommendation unit;

Here, the model data analysis unit includes a grammar analysis unit that checks for grammatical errors in user specifications and a data input analysis unit that performs data validation by taking as input a path of a specification file used to include training data It is characterized by doing.

In addition, the framework selection unit includes an algorithm extraction unit that extracts a user algorithm using a procedure analyzer and a function analyzer, a feature selection unit that analyzes the relationship between the extracted user algorithm and input data, and obtains characteristics of the input data; It is characterized in that it includes a framework scoring unit that assigns a score to the deep learning framework based on the feature.

In addition, the feature selection unit includes a user specification analysis unit that analyzes the user specification when the user algorithm is extracted in the algorithm extraction unit, a framework information analysis unit that analyzes framework information necessary for the user specification, and a user that analyzes the user algorithm. It is characterized in that it includes an algorithm analyzer and an input data feature extractor that extracts features of the input data by analyzing the relationship between the extracted user algorithm and the input data.

And the framework scoring unit includes a framework information patching unit for patching predefined deep learning framework information, a user specification feature patching unit for patching the characteristics of the user specification obtained by the feature selection process of the feature selection unit, and a user specification feature patching unit. A score generation unit that generates scores for each feature defined in the deep learning framework if the feature value of the feature of is present in the feature values of the feature of the currently selected deep learning framework, and a deep learning frame in which all features of the user specification are selected. If supported as a framework, a support framework list addition unit that adds the framework to the list of supported frameworks, and a support framework list that returns a list of supported frameworks so that the deep learning framework with the highest score can be recommended. It is characterized in that it includes a return part.

In addition, the user specification code used as the deep learning input of the framework recommendation system includes information that can extract the features necessary for selecting a deep learning framework, and the information that can extract the features is the purpose of the project, It is characterized by including a backend environment, training data, and user algorithm items written in pseudocode.

To achieve another object, a deep learning framework recommendation method using a feature selection and scoring algorithm according to the present invention includes a framework property/feature input step of inputting model template and framework property information for framework selection; User specification input step of entering data attribute information to be used with the deep learning model; receiving framework attributes/characteristics and user specifications through the model data analysis process and framework selection process, and recommending a deep learning framework using feature selection and scoring algorithms It is characterized by including; a deep learning framework recommendation step of doing; a recommendation framework information output step of outputting recommended framework information and framework optimization model design recommended in the deep learning framework recommendation step.

Here, the model data analysis process includes a grammar analysis process for checking user specifications for grammatical errors and a data input analysis process for performing data validation by taking as input a path of a specification file including training data. do.

And the framework selection process includes an algorithm extraction process of extracting a user algorithm using a procedure analyzer and a function analyzer, a feature selection process of obtaining characteristics of input data by analyzing the relationship between the extracted user algorithm and input data, It is characterized in that it includes a framework scoring process for assigning scores to the deep learning framework based on the selected feature.

And, in the feature selection process, when the user algorithm is extracted in the algorithm extraction process, the user specification analysis process for analyzing the user specification, the framework information analysis process for analyzing the framework information necessary for the user specification, and the user algorithm analysis It is characterized in that it includes a user algorithm analysis process and an input data feature extraction process of extracting features of the input data by analyzing the relationship between the extracted user algorithm and the input data.

In addition, the framework scoring process includes a framework information patching process for patching predefined deep learning framework information, a user specification feature patching process for patching user specification characteristics obtained by the feature selection process, and a user specification feature patching process. If the feature value of exists in the feature value of the feature of the currently selected deep learning framework, a score generation process of generating a score for each feature defined in the deep learning framework, and all features of the user specification are transferred to the selected deep learning framework. If supported, a process of adding a supported framework to the list of supported frameworks and a process of returning a list of supported frameworks to return a list of supported frameworks so that the deep learning framework with the highest score can be recommended It is characterized in that it includes.

The deep learning framework recommendation system and method using the feature selection and scoring algorithm according to the present invention as described above implements the following effects.

First, it receives user specifications as input and recommends the optimal deep learning framework for implementing the user's algorithm so that the optimal deep learning framework suitable for the learning target is applied efficiently.

Second, it outputs the recommended deep learning framework and framework optimization model design most suitable for the user's algorithm so that the user can select the optimal deep learning framework according to his or her purpose.

Third, users can use the built-in functions of the framework to easily build the algorithm, saving time in implementing the algorithm and checking whether the recommended framework can be used in the target environment.

1 is a block diagram of a deep learning framework recommendation system using a feature selection and scoring algorithm according to the present invention.
2 is a detailed configuration diagram of the model data analysis unit
3 is a detailed configuration diagram of a framework selection unit
4 is a detailed configuration diagram of a feature selection unit
5 is a detailed configuration diagram of the framework scoring unit
6 is a configuration diagram showing an example of a text file format user specification code in a user-defined format
7 is a flow chart showing a method for recommending a deep learning framework using a feature selection and scoring algorithm according to the present invention.
8 is an operational flow chart showing a detailed process for selecting a framework
9 is a flow chart showing a detailed process of framework scoring

Hereinafter, a preferred embodiment of a deep learning framework recommendation system and method using a feature selection and scoring algorithm according to the present invention will be described in detail.

Features and advantages of the deep learning framework recommendation system and method using the feature selection and scoring algorithm according to the present invention will become clear through the detailed description of each embodiment below.

1 is a block diagram of a deep learning framework recommendation system using a feature selection and scoring algorithm according to the present invention.

The system and method for recommending a deep learning framework using a feature selection and scoring algorithm according to the present invention receives user specifications as an input and recommends the optimal deep learning framework for implementing the user's algorithm, thereby providing the optimal This is to ensure that the application of the deep learning framework is made efficient.

To this end, the present invention may include a configuration for outputting a recommended deep learning framework most suitable for the user's algorithm and a framework optimization model design so that the user can select the optimal deep learning framework according to his or her purpose.

A deep learning framework recommendation system using a feature selection and scoring algorithm according to the present invention, as shown in FIG. 1, includes a framework property/feature input unit 10 for inputting model template and framework property information for framework selection; It consists of a user specification input unit 20 for inputting the input deep learning model to be converted and data property information to be used, a model data analysis unit 30a, and a framework selection unit 30b to receive framework properties/characteristics and user specifications. A deep learning framework recommendation unit 30 that recommends a deep learning framework using a feature selection and scoring algorithm, and outputs recommendation framework information that outputs recommendation framework information most suitable for the user's algorithm and framework optimization model design. It includes part 40.

In the deep learning framework recommendation system using the feature selection and scoring algorithm according to the present invention, information on various deep learning frameworks is defined in advance. After performing matching with input user specification information, a list of supported frameworks is returned, and then the framework with the highest score is recommended.

Information that can be included in user specification information includes 'purpose of program', 'device to perform deep learning', 'model size', 'data type', and the like.

The detailed configuration of the model data analysis unit 30a is as follows.

2 is a detailed configuration diagram of a model data analysis unit.

As shown in FIG. 2, the model data analysis unit 30a receives the syntax analysis unit 21 that checks whether or not there is a grammatical error in the user specification and the path of the specification file, which is a form used to include training data, as input. It includes a data input analysis unit 22 that performs validation.

The detailed configuration of the framework selection unit 30b is as follows.

3 is a detailed configuration diagram of a framework selection unit.

As shown in FIG. 3, the framework selection unit 30b analyzes the relationship between the algorithm extraction unit 31 that extracts the user algorithm using a procedure analyzer and a function analyzer, and the extracted user algorithm and input data to input data. It includes a feature selection unit 32 that obtains characteristics of data and a framework scoring unit 33 that assigns scores to the most suitable deep learning framework based on the selected characteristics.

The detailed configuration of the feature selection unit 32 is as follows.

4 is a detailed configuration diagram of a feature selection unit.

As shown in FIG. 4, the feature selection unit 32 includes a user specification analysis unit 41 that analyzes the user specification when the algorithm extraction unit 31 extracts the user algorithm, and analyzes framework information necessary for the user specification. A framework information analysis unit 42 that analyzes user algorithms, a user algorithm analysis unit 43 that analyzes user algorithms, and an input data feature extraction unit that extracts characteristics of input data by analyzing the relationship between the extracted user algorithm and input data. (44).

The detailed configuration of the framework scoring unit 33 is as follows.

5 is a detailed configuration diagram of a framework scoring unit.

As shown in FIG. 5, the framework scoring unit 33 is a framework information patching unit 51 for patching predefined deep learning framework information and a user obtained by the feature selection process of the feature selection unit 32. [ score (feature value) * weight (feature)] to generate a score of the corresponding feature, and if all features of the user specification are supported by the selected deep learning framework, the framework is supported. Includes a support framework list addition unit 54 for adding to the list of frameworks, and a support framework list return unit 55 for returning a list of supported frameworks so that the deep learning framework with the highest score can be recommended. do.

6 shows an example of a user-specified code in a text file format in a user-defined format. The user-specified code is a deep learning input for a framework recommendation system, and includes project goals, backend environments, training data, and pseudocode. It includes information for extracting necessary features for selecting a deep learning framework, such as the user's algorithm.

The deep learning framework recommendation method using the feature selection and scoring algorithm according to the present invention is described in detail as follows.

7 is a flowchart illustrating a deep learning framework recommendation method using a feature selection and scoring algorithm according to the present invention.

The deep learning framework recommendation method using the feature selection and scoring algorithm according to the present invention includes a user specification input step (S701) receiving user specifications as input, and model data for grammar analysis and data input analysis using a model data analyzer. An analysis step (S702) and a deep learning framework recommendation step (S703) of recommending a deep learning framework based on feature selection through algorithm extraction, feature selection, and framework scoring using a framework selection algorithm, and and a recommended framework information output step (S704) of outputting the most suitable recommended deep learning framework and framework optimization model design.

The detailed description of the model data analysis steps is as follows.

The model data analysis step is to analyze the grammar and data input of user specifications through the grammar analyzer and data input analyzer.

At this stage, the grammar analyzer checks the user specification for grammatical errors.

For example, a system error occurs when information other than the information previously defined in the system is input as the input of the frontend target device (frontend_target_device).

If there are no grammatical errors, the user specifications entered as input for model data analysis are output without system errors.

The data input analyzer performs data validation by taking the path of the specification file containing the training data as input, and if invalid data is found, a system error occurs. If there is no error, the system does not generate an error. Assign the input data to the appropriate variables.

The specific steps for recommending a deep learning framework are as follows.

8 is an operational flowchart illustrating a detailed process for selecting a framework.

The framework selection process consists of an algorithm extraction step (S801), a feature selection step (S802), and a framework scoring step (S803).

In the algorithm extraction step, a user algorithm is extracted using a procedure analyzer and function analyzer.

In the feature selection step, the relationship between the extracted user algorithm and the input data is analyzed to obtain the features of the input data.

In the framework scoring step, the most suitable deep learning framework is scored based on the selected features.

Then, based on the scores given in the recommended framework information output step, information on the recommended deep learning framework and framework optimization model design are output.

Here, in the feature selection step (S802), if the user algorithm is extracted in the algorithm extraction step, the user specification analysis step of analyzing the user specification, the framework information analysis step of analyzing the framework information necessary for the user specification, and the user algorithm It includes a user algorithm analysis step of performing analysis, and an input data feature extraction step of extracting features of the input data by analyzing the relationship between the extracted user algorithm and the input data.

In FIG. 8 , four characteristics of input data are largely represented, which may be further added according to specific implementation.

The features are classified into data size, dynamic graph, CNN, and CUDA, and each feature gives a score to various deep learning frameworks.

For example, for dynamic graphs, you can give Tensorflow, PyTorch, and MXNet 4, 10, and 4 points respectively.

And the detailed configuration of the framework scoring step (S803) is as follows.

9 is a flowchart showing a detailed process of framework scoring.

The framework scoring step (S803) includes a framework information patching step (S901) of patching predefined deep learning framework information and a user specification feature patching step (S902) of patching user specification characteristics obtained by the feature selection process. ), and matching feature values of selected framework features and user-specified feature values, if the feature values of user-specified features exist in the feature values of features of the currently selected deep learning framework, the feature values defined in the deep learning framework are A score generation step of generating a score for each feature (S905), and a step of adding a list of supported frameworks to add the framework to the list of supported frameworks if all features of the user specification are supported by the selected deep learning framework ( S906) and a supported framework list returning step (S97) of returning a list of supported frameworks so that the deep learning framework with the highest score can be recommended.

Framework scoring algorithms are essential to recommend optimal deep learning frameworks for selected characteristics of user specifications.

The system starts by patching various predefined deep learning framework information, and then patches the features of user specifications obtained through the 'feature selection' process.

If the feature value of the feature of the user specification exists in the feature value of the feature of the currently selected deep learning framework, 'score (feature value) * weight (feature)' for each feature defined in the deep learning framework is performed to obtain the corresponding Generate a score for a feature.

This is done for every feature in the user's specification, and as a result the score for that framework is obtained by summing the scores of all features.

Through this, if all features of the user specification are supported by the selected deep learning framework, the framework is added to the list of supported frameworks, and if the selected deep learning framework does not support the feature value of the feature included in the user specification In this case, the same process is performed by selecting another deep learning framework that exists in the system.

When the above process is performed for all deep learning frameworks existing in the system, a list of supported frameworks is returned and the system recommends the deep learning framework with the highest score among them.

In the example of framework scoring, for example, in the case of 'device' among various characteristics of user specification, it is assumed that the characteristic of user specification is 'device' and the characteristic value is 'android'.

Assuming that the system framework is Tensorflow, first check whether there is a 'device' feature among Tensorflow features, and if not, select another deep learning framework. (CUDA, TPU, CPU, Android, etc.) exist.

After checking whether each feature value is the same as the feature specified by the user, if it is the same, a score of the feature value can be obtained, and the obtained score is multiplied by the weight of the feature to perform scoring for the feature.

This is equally applied to all other features, and framework scoring is performed by summing scores for all features.

The deep learning framework recommendation system and method using the feature selection and scoring algorithm according to the present invention described above receives user specifications as input and recommends the optimal deep learning framework for implementing the user's algorithm, thereby learning target This is to ensure that the application of the optimal deep learning framework for the application is made efficiently.

According to the present invention, a user can use a built-in function of a framework to easily build an algorithm, thereby saving time in implementing the algorithm and checking whether a recommended framework can be used in a target environment.

As described above, it will be understood that the present invention is implemented in a modified form without departing from the essential characteristics of the present invention.

Therefore, the specified embodiments should be considered from an explanatory point of view rather than a limiting point of view, and the scope of the present invention is shown in the claims rather than the foregoing description, and all differences within the equivalent range are considered to be included in the present invention. will have to be interpreted

10. Framework property/characteristic input part
20. User specification input unit
30. Deep Learning Framework Recommendation Unit
40. Recommendation framework information output unit

Claims (11)

a framework property/characteristic input unit for inputting model template and framework property information for selecting a framework;
a user specification input unit for inputting input deep learning model to be converted and data attribute information to be used;
A deep learning framework recommendation unit composed of a model data analysis unit and a framework selection unit that receives framework properties/characteristics and user specifications and recommends a deep learning framework using a feature selection and scoring algorithm;
A deep learning framework recommendation system using a feature selection and scoring algorithm, comprising: a recommendation framework information output unit configured to output recommendation framework information and framework optimization model design recommended by the deep learning framework recommendation unit. The method of claim 1, wherein the model data analysis unit,
a grammar analysis unit that checks for grammatical errors in user specifications;
A deep learning framework recommendation system using a feature selection and scoring algorithm, comprising a data input analysis unit that performs data validation by taking as input a path of a specification file, which is a form used to include training data. The method of claim 1, wherein the framework selection unit,
An algorithm extraction unit that extracts a user algorithm using a procedure analyzer and a function analyzer;
a feature selector that obtains features of the input data by analyzing the relationship between the extracted user algorithm and the input data;
A deep learning framework recommendation system using a feature selection and scoring algorithm, comprising a framework scoring unit for assigning scores to the deep learning framework based on selected features. The method of claim 3, wherein the feature selection unit,
A user specification analysis unit that analyzes the user specification when the user algorithm is extracted by the algorithm extraction unit;
A framework information analysis unit for analyzing framework information required for user specifications;
A user algorithm analysis unit that analyzes a user algorithm;
A deep learning framework recommendation system using a feature selection and scoring algorithm, comprising an input data feature extractor for extracting features of the input data by analyzing the relationship between the extracted user algorithm and the input data. The method of claim 3, wherein the framework scoring unit,
A framework information patch unit for patching predefined deep learning framework information;
a user-specification feature patching unit for patching the characteristics of the user-specification obtained by the feature selection process of the feature selection unit;
A score generator for generating a score for each feature defined in the deep learning framework if the feature value of the feature specified by the user exists in the feature value of the feature of the currently selected deep learning framework;
If all characteristics of the user specification are supported by the selected deep learning framework, a support framework list addition unit for adding the framework to the list of supported frameworks;
A deep learning framework recommendation system using a feature selection and scoring algorithm, comprising a support framework list return unit that returns a list of frameworks that support the deep learning framework with the highest score to be recommended. The method of claim 1, wherein the user specification code used as the deep learning input of the framework recommendation system is:
It includes the information that can extract the features needed to select a deep learning framework, and the information that can extract the features includes the purpose of the project, the backend environment, training data, and user algorithm items written in pseudocode. A deep learning framework recommendation system using a feature selection and scoring algorithm comprising a. a framework property/characteristic input step of inputting model template and framework property information for selecting a framework;
A user specification input step of inputting input deep learning model to be converted and data attribute information to be used;
A deep learning framework recommendation step of receiving framework properties/characteristics and user specifications through a model data analysis process and a framework selection process, and recommending a deep learning framework using a feature selection and scoring algorithm;
A method for recommending a deep learning framework using a feature selection and scoring algorithm, comprising: a recommendation framework information output step of outputting recommended framework information and framework optimization model design recommended in the deep learning framework recommendation step. The method of claim 7, wherein the model data analysis process,
A grammatical analysis process of checking user specifications for grammatical errors;
A method for recommending a deep learning framework using a feature selection and scoring algorithm, comprising a data input analysis process for performing data validation by taking a pass of a specification file including training data as an input. The method of claim 7, wherein the framework selection process,
An algorithm extraction process of extracting a user algorithm using a procedure analyzer and a function analyzer;
A feature selection process of obtaining features of the input data by analyzing the relationship between the extracted user algorithm and the input data;
A method for recommending a deep learning framework using a feature selection and scoring algorithm, comprising a framework scoring process for assigning scores to the deep learning framework based on selected features. The method of claim 9, wherein the feature selection process,
In the algorithm extraction process, when the user algorithm is extracted, a user specification analysis process of analyzing the user specification;
A framework information analysis process of analyzing framework information necessary for user specifications;
A user algorithm analysis process for analyzing a user algorithm;
A method for recommending a deep learning framework using a feature selection and scoring algorithm, comprising an input data feature extraction process of extracting features of the input data by analyzing the relationship between the extracted user algorithm and the input data. The method of claim 9, wherein the framework scoring process,
A framework information patching process for patching predefined deep learning framework information;
a user-specification feature patching process for patching the user-specified feature obtained by the feature selection process;
A score generation process of generating a score for each feature defined in the deep learning framework if the feature value of the feature specified by the user exists in the feature value of the feature of the currently selected deep learning framework;
If all features of the user specification are supported by the selected deep learning framework, a process of adding the framework to the list of supported frameworks;
A method for recommending a deep learning framework using a feature selection and scoring algorithm, comprising a process of returning a list of supported frameworks so that a deep learning framework with the highest score can be recommended.

Images