Information processing methods, devices, electronic devices and storage media

By using a distributed system for feature extraction, labeling, and independent processing of game data by storage nodes, the problem of coupling between game AI training sample generation and game logic is solved, thereby improving the efficiency of game AI training and the gaming experience.

CN116401007BActive Publication Date: 2026-11-10NETEASE (HANGZHOU) NETWORK CO LTD
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202310240557.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-03-08
Publication Date
2026-11-10
Estimated Expiration
2043-03-08

AI Technical Summary

Technical Problem

Existing game AI training sample generation technology is highly coupled with game logic, causing game lag and affecting the gaming experience.

Method used

A distributed system is adopted, including feature extraction nodes, label annotation nodes, and storage nodes, which respectively handle feature extraction, label annotation, and sample storage of game data, and generate training samples independently of game logic.

Benefits of technology

This reduces the coupling between training sample generation and game logic, decreases game lag, and improves the generation of game AI training samples and the overall gaming experience.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116401007B_ABST
    Figure CN116401007B_ABST
Patent Text Reader

Abstract

This invention provides an information processing method, apparatus, electronic device, and storage medium; applied to a game AI decision-making system, the game AI decision-making system comprising a distributed system, the distributed system comprising feature extraction nodes, labeling nodes, and storage nodes, the method comprising: when the feature extraction node receives target game data, performing feature extraction on the target game data to obtain target feature data; labeling player operation data in the target game data using the labeling nodes to obtain target label data, the target label data corresponding to the target model configuration; and storing the target feature data and target label data in a target file server using the storage nodes to obtain target game AI training samples in the target file server, the target game AI training samples being used for training the target game AI model.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to an information processing method, apparatus, electronic device, and storage medium. Background Technology

[0002] Artificial intelligence (AI) in games is a type of game robot used to simulate human decision-making behavior in games. It is also known as anthropomorphic player and can play a role in a variety of game scenarios.

[0003] Game AI's decision-making behavior requires model training based on a large amount of real player behavior data to accurately predict outcomes. Existing game AI training samples are typically extracted, labeled, and saved online simultaneously with the game's execution. This means that training sample generation intrudes into the game's logic, burdening the game's operation and impacting the gaming experience. Therefore, existing training sample generation techniques are too tightly coupled with game logic and are prone to causing game lag. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide an information processing method, apparatus, electronic device and storage medium to reduce the coupling between training sample generation and game logic, and to reduce the stuttering phenomenon caused by training sample generation in the game.

[0005] In a first aspect, embodiments of the present invention provide an information processing method applied to a game AI decision-making system. The game AI decision-making system includes a distributed system, comprising feature extraction nodes, labeling nodes, and storage nodes. The method includes: when the feature extraction node receives target game data, performing feature extraction on the target game data to obtain target feature data, the target feature data corresponding to a target model configuration, the target model configuration indicating configuration information of a target game AI model, the target game AI model simulating player operations, and the target game data including player operation data; labeling the player operation data in the target game data using the labeling nodes to obtain target label data, the target label data corresponding to the target model configuration; and storing the target feature data and the target label data in a target file server using the storage nodes to obtain target game AI training samples in the target file server, the target game AI training samples being used for training the target game AI model.

[0006] Secondly, embodiments of the present invention provide an information processing device applied to a game AI decision-making system. The game AI decision-making system includes a distributed system, which includes feature extraction nodes, labeling nodes, and storage nodes. The device includes: an extraction module, used to extract features from the target game data when the feature extraction node receives target game data, to obtain target feature data, the target feature data corresponding to a target model configuration, the target model configuration indicating the configuration information of a target game AI model, the target game AI model simulating player operations, and the target game data including player operation data; a labeling module, used to label the player operation data in the target game data through the labeling nodes, to obtain target label data, the target label data corresponding to the target model configuration; and a generation module, used to store the target feature data and the target label data to a target file server through the storage nodes, to obtain target game AI training samples in the target file server, the target game AI training samples being used for training the target game AI model.

[0007] Thirdly, embodiments of the present invention provide an electronic device, including a processor and a memory, wherein the memory stores machine-executable instructions that can be executed by the processor, and the processor executes the machine-executable instructions to implement the above-described information processing method.

[0008] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing computer-executable instructions. When the computer-executable instructions are invoked and executed by a processor, the computer-executable instructions cause the processor to implement the above-described information processing method.

[0009] The embodiments of the present invention bring the following beneficial effects:

[0010] The aforementioned information processing method, apparatus, electronic device, and storage medium are applied to a game AI decision-making system. The game AI decision-making system includes a distributed system, comprising feature extraction nodes, labeling nodes, and storage nodes. The method includes: when the feature extraction node receives target game data, performing feature extraction on the target game data to obtain target feature data, which corresponds to a target model configuration. The target model configuration indicates the configuration information of the target game AI model, which simulates player operations. The target game data includes player operation data. The labeling nodes label the player operation data in the target game data to obtain target label data, which corresponds to the target model configuration. The storage nodes store the target feature data and the target label data in a target file server to obtain target game AI training samples in the target file server. These target game AI training samples are used for training the target game AI model. In this approach, the distributed system in the game AI decision-making system can perform feature extraction, labeling, and sample storage of game data in parallel through different nodes. This allows the game logic to only collect game data, while the specific training sample generation logic is completed in the distributed system. This reduces the coupling between training sample generation and game logic, and also reduces the stuttering caused by training sample generation in the game.

[0011] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained in accordance with the structures particularly pointed out in the description, claims and drawings.

[0012] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0013] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0014] Figure 1 This is a flowchart of one embodiment of the information processing method in this invention;

[0015] Figure 2This is a flowchart of another embodiment of the information processing method in this invention;

[0016] Figure 3 A schematic diagram of an information processing device provided in an embodiment of the present invention;

[0017] Figure 4 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” or “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0020] For ease of understanding, the specific process of this embodiment of the invention is described below. In this embodiment, the information processing method is applied to a game AI decision-making system. The game AI decision-making system includes a distributed system, which comprises feature extraction nodes, label annotation nodes, and storage nodes. It is understood that the distributed system is deployed through a distributed framework, such as the container cluster management framework K8S (Kubernetes) or the distributed service framework Dubbo, to provide distributed computing services in the game AI decision-making system. A complete distributed computing service / function provided by the distributed system is completed by multiple nodes. A physical machine, virtual machine, Docker image, independent process, server, etc., in the distributed system can all be a node in the distributed system. In this embodiment, the distributed system includes at least three task nodes: feature extraction nodes, label annotation nodes, and storage nodes. In one embodiment, the distributed system also includes a management node, which coordinates the task nodes and performs task scheduling, enabling the task nodes to cooperate to achieve multi-process parallel processing of the target task. In this embodiment, the target task includes, but is not limited to, the task of generating game AI training samples and the decision-making task of the game AI, etc., and is not specifically limited here.

[0021] In one embodiment, the distributed system includes at least one distributed server, each of which includes multiple task nodes. These task nodes are designated as feature extraction nodes, label annotation nodes, and storage nodes, respectively. Each task node corresponds to one process, enabling a single distributed server to perform multi-process parallel processing of the target task. In another embodiment, the distributed system includes multiple distributed servers, each corresponding to one task node. These multiple distributed servers respectively correspond to feature extraction nodes, label annotation nodes, and storage nodes, allowing the multiple distributed servers to cooperate in performing multi-process parallel processing of the target task.

[0022] Please see Figure 1 One embodiment of the information processing method in this invention includes:

[0023] Step S10: When the feature extraction node receives the target game data, it extracts features from the target game data to obtain target feature data. The target feature data corresponds to the target model configuration. The target model configuration is used to indicate the configuration information of the target game AI model. The target game AI model is used to simulate player operations. The target game data includes player operation data.

[0024] In this embodiment, the feature extraction node is used to extract features from the target game data. When other task processing units (such as servers) in the game AI decision-making system, or other nodes in the distributed system, collect the target game data, they send the target game data to the feature extraction node. When the feature extraction node receives the target game data, it extracts features from the target game data to obtain the target feature data. Unlike traditional feature extraction performed in the model, the feature extraction in this embodiment is decoupled from the model and processed independently in the feature extraction node. This reduces the workload of the model, improves processing efficiency, and, due to the independence of the feature extraction node, allows feature processing to be performed in parallel with other tasks (such as labeling), thereby improving the efficiency of game AI training sample generation and game AI decision-making, and ultimately enhancing the game experience.

[0025] In one implementation, the target game data includes player operation data and all data that can influence player operation decisions. Different target games correspond to different target game data. For example, in a Multiplayer Online Battle Arena (MOBA) game, the target game data may include global game data (such as game duration, economic information, monster spawn settings, map obstruction information, and vision information), faction data (such as team size, faction economic information, and faction experience information), and individual unit data (such as health information, mana information, equipment information, individual economic information, and individual experience information). Specific details are not limited here. It is understood that the player operation data in the target game data refers to the input data generated by the player's terminal operations, such as movement direction, movement speed, and skill release. Specific details are not limited here.

[0026] It is understood that the information processing method provided in this embodiment of the invention can generate game AI training samples for different games, as well as game samples for different game AIs within the same game. This embodiment sets the configuration information of game AI models corresponding to different games or different game AIs through target model configuration. Different configuration information indicates the data standards for feature data and / or label data required by different game AI models. This ensures that the acquisition of game AI training samples, the training of game AI models, and the application of game AI models do not require intrusion into the game logic, and can be flexibly configured according to different model requirements, improving the flexibility and adaptability of the game AI. In one embodiment, since the target model configuration is usually not modified after it is determined, the target model configuration can be obtained from the feature extraction node, extracted from the target game data, or obtained from any other storage location; specific details are not limited here. In this embodiment, the target feature data is extracted according to the configuration information of the target game AI model as indicated by the target model configuration. Therefore, the target feature data corresponds to the target model configuration. For example, if the target model configuration indicates that the configuration information of the target game AI model includes target feature data in JSON format, then when the feature extraction node performs feature extraction on the target game data, the generated target feature data will include feature data in JSON format. The specific details are not limited here.

[0027] It should be noted that the target game AI model is used to simulate player operations, making the AI ​​in the target game more human-like. The more mature the target game AI model is trained, the closer its output behavioral decisions are to the behavioral decisions of real players in the game. This implementation method collects game AI training samples almost imperceptibly during the game process operated by real players, and uses them to train the target game AI model, so that its output behavioral decisions are closer to the behavioral decisions of real players in the game, thereby improving the player's game experience.

[0028] In one implementation, to improve the transmission efficiency of target game data, this implementation uses target game data corresponding to a time unit as the processing object. Specifically, during the game process of the target game, target game data corresponding to each time unit is collected in real time and sent to a feature extraction node. When the feature extraction node receives the target game data corresponding to each time unit, it performs feature extraction on the target game data corresponding to each time unit to obtain target feature data corresponding to each time unit. A time unit can be preset according to specific business needs, such as 1 frame, 100ms, 1 second, etc., and is not limited here. In another implementation, the target game data can also be the target game data corresponding to a game session. During the game process of the target game, target game data corresponding to each time unit is collected in real time. When the target game session ends, the target game data corresponding to the target game session is sent to the feature extraction node. When the feature extraction node receives the target game data for the target game session, it performs feature extraction on the target game data corresponding to the target game session to obtain target feature data corresponding to the target game session.

[0029] Step S20: Label the player operation data in the target game data through the label annotation node to obtain the target label data, which corresponds to the target model configuration;

[0030] In this embodiment, to analyze and simulate the actions of real players, target label data corresponding to player action data in the target game data is generated through label annotation nodes according to the configuration information of the target game AI model as indicated by the target model configuration. This allows for the rapid acquisition of a large number of labeled game AI training samples for training the target game AI model. This embodiment can quickly generate data labels through standardized model configuration information, thereby rapidly obtaining a large number of game AI training samples and improving the accuracy of game AI behavior decisions. For example, assuming the player's movement direction is divided into 12 equal parts, with each part corresponding to a label, then the movement direction data in the player's action data can be used to determine the corresponding part and the label of that part, thus labeling the movement direction data and obtaining the target label data.

[0031] In one implementation, steps S10 and S20 can be executed in parallel. The feature extraction node corresponds to one process, and the labeling node corresponds to another process. By having the processes corresponding to the feature extraction node and the labeling node perform feature extraction on the target game data and label the player operation data in the target game data in parallel, the target feature data and target label data are obtained simultaneously. This improves the efficiency of generating game AI training samples. Furthermore, since the generation of target game data and the generation of game AI training samples are independent of each other and do not affect each other, the impact of the generation of game AI training samples on the operation of the game itself can be reduced. This reduces the coupling between the generation of training samples and the game logic, and reduces the occurrence of lag caused by the generation of training samples.

[0032] In one implementation, in order to ensure that the target feature data and target label data in the game AI training samples correspond one-to-one, the target label data and target feature data in this implementation have the same time unit. That is, if each time unit corresponds to one target feature data, then each time unit also corresponds to one target label data. If one game session corresponds to one target game data, then one game session also corresponds to one target label data and one target feature data. The specifics are not limited here.

[0033] Step S30: Store the target feature data and target label data to the target file server through the storage node to obtain the target game AI training sample in the target file server. The target game AI training sample is used for training the target game AI model.

[0034] In this embodiment, the storage node combines target feature data and target label data to obtain target game AI training samples. These training samples are then stored on the target file server, resulting in the target game AI training samples stored on the target file server. In one embodiment, when combining target feature data and target label data, they are combined according to a one-to-one correspondence between time units or game sessions. For example, assuming the time unit is a frame, target feature data and target label data with the same frame number are combined to obtain target game AI training samples with corresponding frame numbers. This facilitates subsequent processing of target feature data and target label data within the same time unit, improving data processing efficiency.

[0035] In one implementation, to meet the large file throughput requirements of game AI training samples during game AI model training, the storage node combines target feature data and target label data to obtain target game AI training samples, then asynchronously stores them in a preset file system. The target game AI training samples in the preset file system are then uploaded to the target file server via cloud storage services. Thus, when game AI training samples are needed, the required number of game AI training samples can be downloaded directly from the target file server without the need to collect training samples in real time, thereby improving the efficiency of game AI model training.

[0036] The information processing method provided by the above implementation method enables the distributed system in the game AI decision-making system to perform feature extraction, labeling, and sample storage of game data in parallel through different nodes. This allows the game logic to collect game data only, while the specific training sample generation logic is completed in the distributed system. This reduces the coupling between training sample generation and game logic, and also reduces the stuttering caused by training sample generation in the game.

[0037] Please see Figure 2 In this embodiment of the invention, the information processing method is applied to a game AI decision-making system. The game AI decision-making system includes a distributed system and a game server. The distributed system includes feature extraction nodes, labeling nodes, and storage nodes. Another embodiment of the information processing method in this invention includes:

[0038] Step S201: When the feature extraction node receives the target game data, it extracts features from the target game data to obtain target feature data. The target feature data corresponds to the target model configuration. The target model configuration is used to indicate the configuration information of the target game AI model. The target game AI model is used to simulate player operations. The target game data includes player operation data.

[0039] In one implementation, the game AI decision-making system further includes a game server and a supervised learning server. Before step S201, the system further includes: saving game recordings for each game session on the game server to obtain target game recordings for each game session on the game server; offline retrieving target game recordings for each game session from the game server via the supervised learning server, and parsing the retrieved target game recordings for each game session to obtain target game data for at least one game session on the supervised learning server; and sending the target game data for at least one game session to the feature extraction node via the supervised learning server.

[0040] This embodiment provides an offline method for generating game AI training samples. To reduce the network resource requirements of the game server during the game process, after each game session ends, the game recording of the current game session is saved on the game server, resulting in the target game recording corresponding to each game session on the game server. When it is necessary to generate game AI training samples, the supervised learning server offline retrieves the target game recordings corresponding to all game sessions on the game server and parses the target game recordings to restore them to target game data, resulting in the target game data corresponding to at least one game session on the supervised learning server. The target game data can be the target game data corresponding to each time unit, and one game session corresponds to the target game data corresponding to multiple frames. Next, the supervised learning server sends a feature extraction request to the feature extraction node and sends the target game data corresponding to at least one game session to the feature extraction node. The feature extraction node responds to the feature extraction request. When the feature extraction node receives the target game data, it performs feature extraction on the target game data to obtain target feature data. This implementation reduces the collection of game data during the game process by fetching game recordings offline, thereby reducing the impact of training the game AI on the game itself. The game only needs to save the corresponding game recording at the end of each game, which greatly reduces the stuttering caused by training sample generation.

[0041] In one embodiment, the game AI decision-making system further includes a game server and a supervised learning server. Before step S201, the system further includes: collecting and packaging target game data for each time unit during the game process on the game server to obtain target game data corresponding to each time unit on the game server; and sending the target game data corresponding to each time unit to the feature extraction node through the game server. This embodiment uses an online game AI training sample generation method. Since the target game data is temporary data generated in real time during the game and is not stored for a long time, the game server collects and packages the target game data for each time unit during each game session to obtain target game data corresponding to each time unit on the game server. The game server then sends the target game data corresponding to each time unit to the feature extraction node in real time. When the feature extraction node receives the target game data, it performs feature extraction on the target game data to obtain target feature data. This embodiment, by sending target game data in real time during the game process, makes the game data required for game AI training and game AI decision-making more accurate and real-time, meeting the needs of real-time online training and real-time decision-making.

[0042] In one embodiment, the target game data includes a target model configuration, which indicates the data format and data standard of the feature data required by the target game AI model. Step S201 includes: when the feature extraction node receives the target game data corresponding to each time unit, based on the data format and data standard of the feature data required by the target game AI model indicated by the target model configuration, the feature extraction node generates standard feature data for the target game data corresponding to each time unit, thus obtaining the target feature data corresponding to each time unit. In this embodiment, the target model configuration is included in the target game data. The target model configuration indicates the data format and data standard of the feature data required by the target game AI model. When the feature extraction node receives the target game data corresponding to each time unit, it generates standard feature data according to the data format and data standard indicated by the target model configuration, thus obtaining the target feature data corresponding to each time unit. This embodiment can set different model configurations in the game data of different games, enabling the feature extraction node to generate feature data according to different model data requirements based on the received model configuration, improving the efficiency and standardization of feature data generation.

[0043] Step S202: Label the player operation data in the target game data through the label annotation node to obtain the target label data, which corresponds to the target model configuration;

[0044] In one implementation, the target game data includes a target model configuration, which indicates the data format and data standard required for labeling the target game AI model. Step S202 includes: based on the data format and data standard required for labeling the target game AI model as indicated by the target model configuration, generating standard labels for player operation data corresponding to each time unit through labeling nodes, thus obtaining target label data corresponding to each time unit. In this implementation, the target model configuration is also used to indicate the data format and data standard required for labeling the target game AI model. When labeling player operation data in the target game data through labeling nodes, standard player operation data labels are generated according to the data format and data standard indicated by the target model configuration, thereby obtaining target label data corresponding to each time unit. This implementation allows different model configurations to be set in the game data of different games, enabling labeling nodes to perform labeling according to different model data requirements based on the model configuration in the game data, thereby improving the efficiency and standardization of labeling.

[0045] Step S203: Store the target feature data and target label data to the target file server through the storage node to obtain the target game AI training sample in the target file server. The target game AI training sample is used for training the target game AI model.

[0046] In one embodiment, the game AI decision-making system further includes a supervised learning server. After step S203, the system further includes: the supervised learning server responding to a model training instruction by reading target game AI training samples from a target file server; and training a preset target game AI model using the read target game AI training samples to obtain a pre-trained target game AI model. In this embodiment, after obtaining the target game AI training samples, the target game AI model can be trained. Specifically, the supervised learning server responding to a model training instruction downloads the target game AI training samples from the target file server, and then trains the preset target game AI model using the target feature data and target label data in the target game AI training samples to obtain a pre-trained target game AI model. The preset target game AI model can be an untrained basic model or a pre-trained target game AI model; the specific method is not limited here. This embodiment separates the supervised learning server from the distributed system, enabling the generation and application of training samples to be separated, making the game AI decision-making process more independent and flexible. The generation of training samples also does not need to be coupled with the game logic, thereby reducing the stuttering caused by the generation of training samples.

[0047] In one embodiment, step S203 includes: cyclically reading target feature data and target label data corresponding to each time unit from a preset storage queue via a storage node, and combining the target feature data and target label data according to the time unit to obtain a target game AI training sample; then asynchronously saving the target game AI training sample to a target file server via multiple processes to obtain the target game AI training sample in the target file server. This embodiment stores the target feature data and target label data corresponding to each time unit in a queue. Specifically, after the feature extraction node generates the target feature data corresponding to each time unit, it adds the generated target feature data to a preset storage queue. After the label annotation node generates the target label data corresponding to each time unit, it also adds the generated target label data to the preset storage queue. The storage node then sequentially reads the target feature data and target label data corresponding to each time unit from the storage queue, and combines the read target feature data and target label data one-to-one according to the time unit to obtain the target game AI training sample. Next, the target game AI training sample is asynchronously stored in the target file server to obtain the target game AI training sample in the target file server. This embodiment avoids storage node blocking and improves data storage efficiency.

[0048] Step S204: The game server responds to the game AI decision-making instructions of the target game by collecting the current game data for each time unit through the game process in the game server, and obtains the current game data corresponding to each time unit. The current game data includes the target game data other than player operation data.

[0049] In this step, when the target game requires game AI, the game server responds to the target game's game AI decision-making instructions by collecting current game data for each time unit during the game process. This current game data does not include player action data, but only target game data other than player action data, such as global game data, faction data, and individual unit data, etc. Specific details are not limited here. This implementation differs from training sample generation because it applies a game AI model to predict / decide the behavior of the game AI within the game; therefore, prediction / decision-making can be completed without collecting player action data.

[0050] Step S205: The game server sends the current game data for each time unit to the feature extraction node, and the feature extraction node extracts features from the current game data corresponding to each time unit to obtain the current feature data corresponding to each time unit. The current feature data corresponds to the target model configuration.

[0051] In this step, the game server sends the current game data for each time unit to the feature extraction node in real time. The feature extraction node then extracts features from the data to obtain the current feature data corresponding to each time unit. It should be noted that the feature extraction node extracts features from the current game data in the same way it extracts features from the target game data. Both the extracted current feature data and the extracted target feature data conform to the target model configuration; however, specific limitations are not specified here.

[0052] Step S206: The target game AI model performs action prediction on the current feature data corresponding to each time unit, obtains the action prediction result of the target game AI, and executes the action prediction result of the target game AI in the target game through the game server.

[0053] In this embodiment, the pre-trained target game AI model can be ported from the supervised learning server to any node in the game server or distributed system. The feature extraction node sends the current feature data corresponding to each time unit to the server or node where the target game AI model is located. The pre-trained target game AI model performs action prediction / decision on the current feature data corresponding to each time unit, thereby obtaining the action prediction result of the target game AI. If the action prediction result is in the game server, there is no need to transmit the action prediction result again, and the action prediction result can be directly executed on the target game AI in the target game through the game server. If the action prediction result is not in the game server, the game server needs to receive the action prediction result before it can execute the action prediction result on the target game AI in the target game. The specifics are not limited here.

[0054] In one embodiment, the game AI decision-making system further includes a supervised learning server. Step S206 includes: sending the current feature data corresponding to each time unit to the supervised learning server via a feature extraction node; performing action prediction on the current feature data corresponding to each time unit using a pre-trained target game AI model in the supervised learning server; obtaining action prediction results for each time unit; sending the action prediction results for each time unit to the game server via the supervised learning server; the game server receiving the action prediction results for each time unit from the supervised learning server; and executing the action prediction results for each time unit on the target game AI in the target game. In this embodiment, after obtaining the current feature data corresponding to each time unit, the feature extraction node can send the current feature data to the supervised learning server. The target game AI model in the supervised learning server performs action prediction / decision-making on the current feature data to obtain the action prediction results for each time unit of the target game AI. Then, the action prediction results for each time unit are sent to the game server. The game server applies the action prediction results to the target game AI, causing the target game AI to execute the action prediction results and make actions consistent with the action prediction results, thereby achieving human-like behavior in the game AI.

[0055] The information processing method provided by the above implementation method enables the distributed system in the game AI decision-making system to perform feature extraction, labeling, and sample storage of game data in parallel through different nodes. This allows the game logic to collect game data only, while the specific training sample generation logic is completed in the distributed system. This reduces the coupling between training sample generation and game logic, and also reduces the stuttering caused by training sample generation in the game.

[0056] For the corresponding method embodiments described above, see [link to relevant documentation]. Figure 3The diagram illustrates an information processing device applied to a game AI decision-making system. The game AI decision-making system includes a distributed system comprising feature extraction nodes, labeling nodes, and storage nodes. The device includes: an extraction module 30, used to extract features from the target game data received by the feature extraction nodes to obtain target feature data, which corresponds to a target model configuration. The target model configuration indicates the configuration information of the target game AI model, which simulates player operations. The target game data includes player operation data. A labeling module 32 is used to label the player operation data in the target game data through the labeling nodes to obtain target label data, which corresponds to the target model configuration. A generation module 34 is used to store the target feature data and the target label data in a target file server through the storage nodes to obtain target game AI training samples in the target file server. The target game AI training samples are used for training the target game AI model.

[0057] The aforementioned information processing device, the distributed system in the game AI decision-making system, can perform feature extraction, labeling, and sample storage of game data in parallel through different nodes. This allows the game logic to only collect game data, while the specific training sample generation logic is completed in the distributed system. This reduces the coupling between training sample generation and game logic, and also reduces the stuttering caused by training sample generation in the game.

[0058] Optionally, the above-mentioned game AI decision-making system further includes a supervised learning server, and the above-mentioned device further includes: a reading module, used by the supervised learning server to read the target game AI training samples from the target file server in response to a model training instruction; and a training module, used to train a preset target game AI model using the read target game AI training samples to obtain a pre-trained target game AI model.

[0059] Optionally, the above-mentioned game AI decision-making system further includes a game server, and the above-mentioned device further includes: a collection module, used by the game server in response to the game AI decision-making instructions of the target game, to collect current game data for each time unit through the game process in the game server, to obtain current game data corresponding to each time unit, the current game data including target game data other than the player operation data; a sending module, used by the game server to send the current game data for each time unit to the feature extraction node, and to perform feature extraction on the current game data corresponding to each time unit through the feature extraction node, to obtain current feature data corresponding to each time unit, the current feature data corresponding to the target model configuration; and a prediction module, used by the target game AI model to predict the action of the game AI based on the current feature data corresponding to each time unit, to obtain the action prediction result of the target game AI, and to execute the action prediction result of the target game AI in the target game through the game server.

[0060] Optionally, the aforementioned game AI decision-making system further includes a supervised learning server. The aforementioned prediction module is further configured to: send the current feature data corresponding to each time unit to the supervised learning server through the feature extraction node, and perform action prediction for the game AI based on the current feature data corresponding to each time unit using the pre-trained target game AI model in the supervised learning server, thereby obtaining the action prediction result for each time unit; send the action prediction result for each time unit to the game server through the supervised learning server, and the game server receives the action prediction result for each time unit sent by the supervised learning server and executes the action prediction result for each time unit on the target game AI in the target game.

[0061] Optionally, the aforementioned game AI decision-making system further includes a game server and a supervised learning server, and the aforementioned device further includes: a recording module, used to save game recordings of each game session through the game server to obtain target game recordings corresponding to each game session in the game server; a retrieval module, used to offline retrieve target game recordings corresponding to each game session in the game server through the supervised learning server, and perform target game data parsing on the retrieved target game recordings corresponding to each game session to obtain target game data corresponding to at least one game session in the supervised learning server; and a first distribution module, used to send the target game data corresponding to the at least one game session to the feature extraction node through the supervised learning server.

[0062] Optionally, the aforementioned game AI decision-making system further includes a game server and a supervised learning server, and the aforementioned device further includes:

[0063] The packaging module is used to collect and package the target game data for each time unit in the game process of the game server, so as to obtain the target game data corresponding to each time unit in the game server; the second distribution module is used to send the target game data corresponding to each time unit to the feature extraction node through the game server.

[0064] Optionally, the target game data mentioned above includes a target model configuration, which is used to indicate the data format and data standard of the feature data required by the target game AI model; the extraction module is further used to: when the feature extraction node receives the target game data corresponding to each time unit, based on the data format and data standard of the feature data required by the target game AI model indicated by the target model configuration, generate standard feature data for the target game data corresponding to each time unit through the feature extraction node, thereby obtaining the target feature data corresponding to each time unit.

[0065] Optionally, the target game data mentioned above includes a target model configuration, which is used to indicate the data format and data standard for the labeling required by the target game AI model; the labeling module is further used to: generate standard labels for the player operation data corresponding to each time unit through the labeling node based on the data format and data standard for the labeling required by the target game AI model indicated by the target model configuration, so as to obtain the target label data corresponding to each time unit.

[0066] Optionally, the above-mentioned generation module is further configured to: read the target feature data and the target label data corresponding to each time unit in the preset storage queue through the storage node in a loop, and combine the target feature data and the target label data according to the time unit to obtain the target game AI training sample; and save the target game AI training sample to the target file server asynchronously through multiple processes to obtain the target game AI training sample in the target file server.

[0067] This embodiment also provides an electronic device, including a processor and a memory. The memory stores machine-executable instructions that can be executed by the processor, and the processor executes the machine-executable instructions to implement the above-described information processing method. This electronic device can be a server or a terminal device.

[0068] See Figure 4 As shown, the electronic device includes a processor 100 and a memory 101. The memory 101 stores machine-executable instructions that can be executed by the processor 100. The processor 100 executes the machine-executable instructions to implement the above-described information processing method.

[0069] Furthermore, Figure 4 The electronic device shown also includes a bus 102 and a communication interface 103, with the processor 100, the communication interface 103 and the memory 101 connected via the bus 102.

[0070] The memory 101 may include high-speed random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Communication between this system network element and at least one other network element is achieved through at least one communication interface 103 (which can be wired or wireless), such as the Internet, wide area network, local area network, or metropolitan area network. The bus 102 may be an ISA bus, PCI bus, or EISA bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 4 The symbol is represented by a single double-headed arrow, but this does not mean that there is only one bus or one type of bus.

[0071] Processor 100 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of processor 100 or by instructions in software form. Processor 100 may be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it may also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this invention. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this invention can be directly manifested as execution by a hardware decoding processor, or execution by a combination of hardware and software modules in the decoding processor. The software module can reside in a readily available storage medium in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory 101. The processor 100 reads information from memory 101 and, in conjunction with its hardware, completes the steps of the method described in the foregoing embodiments, for example:

[0072] When the feature extraction node receives the target game data, it extracts features from the target game data to obtain target feature data. The target feature data corresponds to the target model configuration, which indicates the configuration information of the target game AI model. The target game AI model is used to simulate player operations, and the target game data includes player operation data. The labeling node labels the player operation data in the target game data to obtain target label data, which corresponds to the target model configuration. The storage node stores the target feature data and target label data in the target file server to obtain the target game AI training samples in the target file server. The target game AI training samples are used to train the target game AI model.

[0073] In this approach, the distributed system in the game AI decision-making system can perform feature extraction, labeling, and sample storage of game data in parallel through different nodes. This allows the game logic to only collect game data, while the specific training sample generation logic is completed in the distributed system. This reduces the coupling between training sample generation and game logic, and also reduces the stuttering caused by training sample generation in the game.

[0074] Optionally, the game AI decision-making system also includes a supervised learning server. After storing the target feature data and target label data to the target file server through the storage node to obtain the target game AI training samples in the target file server, the method further includes: the supervised learning server responding to the model training instruction by reading the target game AI training samples from the target file server; and training the preset target game AI model with the read target game AI training samples to obtain the pre-trained target game AI model.

[0075] Optionally, the game AI decision-making system also includes a game server. After storing the target feature data and target label data to the target file server through storage nodes to obtain the target game AI training samples in the target file server, the method further includes: the game server responding to the game AI decision-making instructions of the target game by collecting the current game data for each time unit through the game process in the game server to obtain the current game data corresponding to each time unit, the current game data including the target game data excluding player operation data; sending the current game data for each time unit to the feature extraction node through the game server, and performing feature extraction on the current game data corresponding to each time unit through the feature extraction node to obtain the current feature data corresponding to each time unit, the current feature data corresponding to the target model configuration; performing action prediction of the game AI on the current feature data corresponding to each time unit through the target game AI model to obtain the action prediction result of the target game AI, and executing the action prediction result of the target game AI in the target game through the game server.

[0076] Optionally, the game AI decision-making system also includes a supervised learning server, which uses the target game AI model to predict the actions of the game AI based on the current feature data corresponding to each time unit, obtains the action prediction results of the target game AI, and executes the action prediction results of the target game AI in the target game through the game server. This includes: sending the current feature data corresponding to each time unit to the supervised learning server through the feature extraction node, and using the target game AI model pre-trained in the supervised learning server to predict the actions of the game AI based on the current feature data corresponding to each time unit, obtaining the action prediction results for each time unit; sending the action prediction results for each time unit to the game server through the supervised learning server, and the game server receiving the action prediction results for each time unit sent by the supervised learning server, and executing the action prediction results for each time unit of the target game AI in the target game.

[0077] Optionally, the game AI decision-making system also includes a game server and a supervised learning server. Before the feature extraction node extracts features from the target game data to obtain target feature data when it receives target game data, the method further includes: saving game recordings for each game session on the game server to obtain the target game recordings for each game session on the game server; offline fetching the target game recordings for each game session on the game server through the supervised learning server, and parsing the target game data for each fetched game session to obtain the target game data for at least one game session on the supervised learning server; and sending the target game data for at least one game session on the supervised learning server to the feature extraction node.

[0078] Optionally, the game AI decision-making system also includes a game server and a supervised learning server. Before the feature extraction node extracts features from the target game data to obtain the target feature data when it receives the target game data, the method further includes: collecting and packaging the target game data for each time unit during the game process on the game server to obtain the target game data corresponding to each time unit on the game server; and sending the target game data corresponding to each time unit to the feature extraction node through the game server.

[0079] Optionally, the target game data includes a target model configuration, which indicates the data format and data standard of the feature data required by the target game AI model. When the feature extraction node receives the target game data, it performs feature extraction on the target game data to obtain target feature data. This includes: when the feature extraction node receives the target game data corresponding to each time unit, based on the data format and data standard of the feature data required by the target game AI model indicated by the target model configuration, it performs standard feature data generation on the target game data corresponding to each time unit to obtain the target feature data corresponding to each time unit.

[0080] Optionally, the target game data includes a target model configuration, which indicates the data format and data standard for labeling the target game AI model. The player operation data in the target game data is labeled by the label labeling node to obtain target label data, including: the data format and data standard for labeling the target game AI model as indicated by the target model configuration, and the standard label generation of the player operation data corresponding to each time unit through the label labeling node to obtain the target label data corresponding to each time unit.

[0081] Optionally, the target feature data and target label data are stored in the target file server through the storage node to obtain the target game AI training sample in the target file server. This includes: reading the target feature data and target label data corresponding to each time unit in the preset storage queue through the storage node in a loop, and combining the target feature data and target label data according to the time unit to obtain the target game AI training sample; and saving the target game AI training sample to the target file server asynchronously through multiple processes to obtain the target game AI training sample in the target file server.

[0082] This embodiment also provides a computer-readable storage medium storing computer-executable instructions. When these computer-executable instructions are invoked and executed by a processor, they cause the processor to implement the aforementioned information processing method, for example:

[0083] When the feature extraction node receives the target game data, it extracts features from the target game data to obtain target feature data. The target feature data corresponds to the target model configuration, which indicates the configuration information of the target game AI model. The target game AI model is used to simulate player operations, and the target game data includes player operation data. The labeling node labels the player operation data in the target game data to obtain target label data, which corresponds to the target model configuration. The storage node stores the target feature data and target label data in the target file server to obtain the target game AI training samples in the target file server. The target game AI training samples are used to train the target game AI model.

[0084] In this approach, the distributed system in the game AI decision-making system can perform feature extraction, labeling, and sample storage of game data in parallel through different nodes. This allows the game logic to only collect game data, while the specific training sample generation logic is completed in the distributed system. This reduces the coupling between training sample generation and game logic, and also reduces the stuttering caused by training sample generation in the game.

[0085] Optionally, the game AI decision-making system also includes a supervised learning server. After storing the target feature data and target label data to the target file server through the storage node to obtain the target game AI training samples in the target file server, the method further includes: the supervised learning server responding to the model training instruction by reading the target game AI training samples from the target file server; and training the preset target game AI model with the read target game AI training samples to obtain the pre-trained target game AI model.

[0086] Optionally, the game AI decision-making system also includes a game server. After storing the target feature data and target label data to the target file server through storage nodes to obtain the target game AI training samples in the target file server, the method further includes: the game server responding to the game AI decision-making instructions of the target game by collecting the current game data for each time unit through the game process in the game server to obtain the current game data corresponding to each time unit, the current game data including the target game data excluding player operation data; sending the current game data for each time unit to the feature extraction node through the game server, and performing feature extraction on the current game data corresponding to each time unit through the feature extraction node to obtain the current feature data corresponding to each time unit, the current feature data corresponding to the target model configuration; performing action prediction of the game AI on the current feature data corresponding to each time unit through the target game AI model to obtain the action prediction result of the target game AI, and executing the action prediction result of the target game AI in the target game through the game server.

[0087] Optionally, the game AI decision-making system also includes a supervised learning server, which uses the target game AI model to predict the actions of the game AI based on the current feature data corresponding to each time unit, obtains the action prediction results of the target game AI, and executes the action prediction results of the target game AI in the target game through the game server. This includes: sending the current feature data corresponding to each time unit to the supervised learning server through the feature extraction node, and using the target game AI model pre-trained in the supervised learning server to predict the actions of the game AI based on the current feature data corresponding to each time unit, obtaining the action prediction results for each time unit; sending the action prediction results for each time unit to the game server through the supervised learning server, and the game server receiving the action prediction results for each time unit sent by the supervised learning server, and executing the action prediction results for each time unit of the target game AI in the target game.

[0088] Optionally, the game AI decision-making system also includes a game server and a supervised learning server. Before the feature extraction node extracts features from the target game data to obtain target feature data when it receives target game data, the method further includes: saving game recordings for each game session on the game server to obtain the target game recordings for each game session on the game server; offline fetching the target game recordings for each game session on the game server through the supervised learning server, and parsing the target game data for each fetched game session to obtain the target game data for at least one game session on the supervised learning server; and sending the target game data for at least one game session on the supervised learning server to the feature extraction node.

[0089] Optionally, the game AI decision-making system also includes a game server and a supervised learning server. Before the feature extraction node extracts features from the target game data to obtain the target feature data when it receives the target game data, the method further includes: collecting and packaging the target game data for each time unit during the game process on the game server to obtain the target game data corresponding to each time unit on the game server; and sending the target game data corresponding to each time unit to the feature extraction node through the game server.

[0090] Optionally, the target game data includes a target model configuration, which indicates the data format and data standard of the feature data required by the target game AI model. When the feature extraction node receives the target game data, it performs feature extraction on the target game data to obtain target feature data. This includes: when the feature extraction node receives the target game data corresponding to each time unit, based on the data format and data standard of the feature data required by the target game AI model indicated by the target model configuration, it performs standard feature data generation on the target game data corresponding to each time unit to obtain the target feature data corresponding to each time unit.

[0091] Optionally, the target game data includes a target model configuration, which indicates the data format and data standard for labeling the target game AI model. The player operation data in the target game data is labeled by the label labeling node to obtain target label data, including: the data format and data standard for labeling the target game AI model as indicated by the target model configuration, and the standard label generation of the player operation data corresponding to each time unit through the label labeling node to obtain the target label data corresponding to each time unit.

[0092] Optionally, the target feature data and target label data are stored in the target file server through the storage node to obtain the target game AI training sample in the target file server. This includes: reading the target feature data and target label data corresponding to each time unit in the preset storage queue through the storage node in a loop, and combining the target feature data and target label data according to the time unit to obtain the target game AI training sample; and saving the target game AI training sample to the target file server asynchronously through multiple processes to obtain the target game AI training sample in the target file server.

[0093] The computer program products of the information processing method, apparatus, electronic device and storage medium provided in the embodiments of the present invention include a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the methods described in the preceding method embodiments. For specific implementation, please refer to the method embodiments, which will not be repeated here.

[0094] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system and apparatus described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0095] Furthermore, in the description of the embodiments of the present invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in the present invention based on the specific circumstances.

[0096] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0097] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0098] Finally, it should be noted that the above embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An information processing method, characterized in that, The method is applied to a game AI decision-making system, which includes a distributed system and a game server. The distributed system includes feature extraction nodes, labeling nodes, and storage nodes. When the feature extraction node receives the target game data, it performs feature extraction on the target game data to obtain target feature data. The target feature data corresponds to the target model configuration. The target model configuration is used to indicate the configuration information of the target game AI model. The target game AI model is used to simulate player operations. The target game data includes player operation data. The player operation data in the target game data is labeled using the label annotation nodes to obtain target label data, which corresponds to the target model configuration. The target feature data and the target label data are stored in the target file server through the storage node to obtain the target game AI training sample in the target file server. The target game AI training sample is used for training the target game AI model. The game server responds to the game AI decision-making instructions of the target game by collecting the current game data for each time unit through the game process in the game server, and obtains the current game data corresponding to each time unit. The current game data includes the target game data other than the player operation data. The game server sends the current game data for each time unit to the feature extraction node, and the feature extraction node performs feature extraction on the current game data corresponding to each time unit to obtain the current feature data corresponding to each time unit. The current feature data corresponds to the target model configuration. The target game AI model performs action prediction on the current feature data corresponding to each time unit to obtain the action prediction result of the target game AI, and then executes the action prediction result of the target game AI in the target game through the game server.

2. The method according to claim 1, characterized in that, The game AI decision-making system further includes a supervised learning server. After storing the target feature data and the target label data in the target file server through the storage node to obtain the target game AI training samples in the target file server, the method further includes: The supervised learning server responds to the model training instruction and reads the target game AI training samples from the target file server. The target game AI model is trained by reading the target game AI training samples to obtain the pre-trained target game AI model.

3. The method according to claim 1, characterized in that, The game AI decision-making system also includes a supervised learning server. The process of predicting the action of the game AI by analyzing the current feature data corresponding to each time unit using the target game AI model, obtaining the action prediction result of the target game AI, and then executing the action prediction result on the target game AI in the target game through the game server includes: The feature extraction node sends the current feature data corresponding to each time unit to the supervised learning server, and the target game AI model pre-trained in the supervised learning server performs action prediction of the game AI on the current feature data corresponding to each time unit to obtain the action prediction result for each time unit. The supervised learning server sends the action prediction results for each time unit to the game server. The game server receives the action prediction results for each time unit sent by the supervised learning server and executes the action prediction results for each time unit on the target game AI in the target game.

4. The method according to claim 1, characterized in that, The game AI decision-making system further includes a game server and a supervised learning server. Before the feature extraction node extracts features from the target game data to obtain target feature data when the feature extraction node receives the target game data, the method further includes: By saving the game recordings of each game session on the game server, the target game recording corresponding to each game session in the game server can be obtained. The supervised learning server offline retrieves the target game recordings corresponding to each game session from the game server, and performs target game data parsing on the retrieved target game recordings corresponding to each game session to obtain the target game data corresponding to at least one game session in the supervised learning server. The supervised learning server sends the target game data corresponding to each of the at least one game session to the feature extraction node.

5. The method according to claim 1, characterized in that, The game AI decision-making system further includes a game server and a supervised learning server. Before the feature extraction node extracts features from the target game data to obtain target feature data when the feature extraction node receives the target game data, the method further includes: During the game process of the game server, the target game data for each time unit is collected and packaged to obtain the target game data corresponding to each time unit in the game server. The game server sends the target game data corresponding to each time unit to the feature extraction node.

6. The method according to claim 1, characterized in that, The target game data includes a target model configuration, which indicates the data format and data standard of the feature data required by the target game AI model. When the feature extraction node receives the target game data, it performs feature extraction on the target game data to obtain target feature data, including: When the feature extraction node receives the target game data corresponding to each time unit, it generates standard feature data for the target game AI model corresponding to each time unit based on the data format and data standard of the feature data required by the target game AI model as indicated by the target model configuration, thereby obtaining the target feature data corresponding to each time unit.

7. The method according to claim 1, characterized in that, The target game data includes a target model configuration, which indicates the data format and data standard for the labeling required by the target game AI model. The step of tagging player operation data in the target game data through the tagging nodes to obtain target tag data includes: Based on the data format and data standard required for labeling the target game AI model as indicated by the target model configuration, standard labels are generated for the player operation data corresponding to each time unit through the label labeling node, thus obtaining the target label data corresponding to each time unit.

8. The method according to claim 1, characterized in that, The step of storing the target feature data and the target label data to the target file server through the storage node to obtain the target game AI training samples in the target file server includes: The target feature data and target label data corresponding to each time unit in the preset storage queue are read cyclically through the storage node, and the target feature data and target label data are combined according to the time unit to obtain the target game AI training sample; The target game AI training samples are saved to the target file server asynchronously through multiple processes, thus obtaining the target game AI training samples in the target file server.

9. An information processing device, characterized in that, An apparatus for use in a game AI decision-making system, the game AI decision-making system comprising a distributed system, the distributed system comprising feature extraction nodes, label annotation nodes, and storage nodes, the apparatus comprising: An extraction module is used to extract features from the target game data when the feature extraction node receives the target game data, thereby obtaining target feature data. The target feature data corresponds to the target model configuration, which is used to indicate the configuration information of the target game AI model. The target game AI model is used to simulate player operations, and the target game data includes player operation data. The annotation module is used to annotate player operation data in the target game data with the label annotation nodes to obtain target label data, which corresponds to the target model configuration. The generation module is used to store the target feature data and the target label data to the target file server through the storage node to obtain the target game AI training sample in the target file server. The target game AI training sample is used for training the target game AI model. The acquisition module is used by the game server to respond to the game AI decision-making instructions of the target game, and to acquire the current game data for each time unit through the game process in the game server, so as to obtain the current game data corresponding to each time unit. The current game data includes the target game data other than the player operation data. The sending module is used to send the current game data for each time unit to the feature extraction node through the game server, and to extract the features from the current game data corresponding to each time unit through the feature extraction node to obtain the current feature data corresponding to each time unit. The current feature data corresponds to the target model configuration. The prediction module is used to predict the action of the game AI based on the current feature data corresponding to each time unit through the target game AI model, obtain the action prediction result of the target game AI, and execute the action prediction result of the target game AI in the target game through the game server.

10. An electronic device, characterized in that, The method includes a processor and a memory, the memory storing machine-executable instructions that can be executed by the processor, the processor executing the machine-executable instructions to implement the information processing method according to any one of claims 1-8.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when invoked and executed by a processor, cause the processor to implement the information processing method according to any one of claims 1-8.

Citation Information

Patent Citations

  • Data processing method and device for game and medium

    CN110064205A