JP2019095973A - Information processor and information processing program - Google Patents

Abstract

To efficiently learn an artificial intelligence agent playing an electronic game using a plurality of objects.SOLUTION: A player log collection part 52 acquires a player log to be data showing a game progress history of a player related to a game. An expression learning unit 54 performs expression learning on the basis of the player log stored in a player log DB36 and learns feature vectors expressing features of respective characters. A learning part 56 with a teacher has AI agents 42 learn strategies about use of the respective characters while using feature vectors of the respective characters learned by the expression learning unit 54 on the basis of the player log stored to the player log DB36. An AI match processing part 58 has the AI agents 42 that have learned match each other in the game, and an AI log collection part 62 acquires AI logs. A statistic data generation part 64 generates statistic data 44a related to a use result of the characters on the basis of the AI logs.SELECTED DRAWING: Figure 5

Description

  The present invention relates to an information processing apparatus and an information processing program, and more particularly to an information processing apparatus and an information processing program for providing an electronic game.

  BACKGROUND Conventionally, an information processing apparatus that provides an electronic game is known. As an electronic game provided by such an information processing apparatus, an electronic game using a plurality of objects (in-game objects) such as characters and cards is known. As an example, there is a battle game in which a player uses a plurality of objects while fighting against an opponent who is a computer or another player.

  On the other hand, research on machine learning has been actively conducted in recent years. For example, research has been conducted on deep learning using multi-layered neural networks. Such deep learning is used to learn an artificial intelligence (AI) agent playing an electronic game, and an AI agent having higher performance than before (that is, a game is “good”) is realized. For example, Non-Patent Document 1 discloses a technique related to a deep neural network applied to an AI agent playing Go.

David Silver, et al, “Mastering the Game of Go with Deep Neural Networks and Tree Search”, Nature 529, 484-489, (28 January 2016)

  In electronic games such as Go and Shogi, the types of objects that can be used are limited. For example, in Go, only Shiraishi and Kuroishi are the types of objects, and in Shogi, there are only limited types of objects such as king, dragon, flying, horse, horn, and the like. In order to learn an AI agent playing an electronic game, it is necessary to identify and learn an object used in the electronic game. In the case of Go and Shogi, as described above, since there are few types of objects used, even if identification is performed by individually labeling each object by a method such as one hot vector, a particular problem Did not occur.

  However, when learning an AI agent playing an electronic game in which a large number of objects, for example, a few hundred or more, are used, labeling each object individually represents the objects. The vector will have a huge number of dimensions. As a result, there may arise a problem that the processing for game play in the AI agent or the learning processing of the AI agent becomes complicated.

  In addition, among many objects, there may be objects having similar characteristics (for example, game effects when used). When learning an AI agent, it may be sufficient to similarly learn an object having similar characteristics. In other words, since objects with similar features need not be identified with one another, identifying and learning objects with similar features may be inefficient.

  An object of the present invention is to more efficiently learn an artificial intelligence agent playing an electronic game using a plurality of objects.

  The present invention shows a game progress history of a player related to an electronic game using a plurality of objects, and an expression learning learning a feature vector representing a feature of each of the objects based on a game log including a use result of the plurality of objects. At least using the unit, the game log, and the feature vector of each of the objects learned by the expression learning unit, the artificial intelligence agent who plays the electronic game learns a strategy regarding the use of the plurality of objects An information processing apparatus comprising: a strategy learning unit.

  Preferably, the strategy learning unit is at least a supervised learning unit that performs supervised learning on the artificial intelligence agent based on the game log, and at least a reinforcement learning unit that performs reinforcement learning on the artificial intelligence agent. It is characterized by including one.

  Desirably, a statistic indicating a game progress history of the electronic game using the learned artificial intelligence agent, and generating statistical data on the use result of the object based on an AI log including the use result of the plurality of objects A data generation unit and a display control unit for displaying the statistical data on a display unit are further included.

  Preferably, in the competition of the electronic game using the artificial intelligence agent, an object under development which can not be used by a general player is used, and the AI log includes a use result of the object under development, and the statistical data The generating unit generates statistical data on the use result of the object under development.

  Preferably, the AI log includes a game progress history of the electronic game by the learned artificial intelligence agents and a use result of the plurality of objects.

  Preferably, the statistical data includes a winning percentage when the object is used, a probability of activating a special effect which can be triggered by the object when the object is used, and a distribution of effects generated when the object is used. And at least one of the above.

  The present invention also provides a computer with a game progress history of a player related to an electronic game using a plurality of objects, and a feature vector representing the feature of each of the objects based on a game log including usage results of the plurality of objects. Using the plurality of objects in an artificial intelligence agent that plays the electronic game using at least an expression learning unit that learns, the game log, and the feature vector of each of the objects learned by the expression learning unit And an information processing program characterized by causing it to function as a strategy learning unit for learning a strategy related to

  According to the present invention, an artificial intelligence agent playing an electronic game using a plurality of objects can be learned more efficiently.

It is a structure schematic diagram of a game system concerning this embodiment. It is a figure which shows the example of the game screen of the game which the game system which concerns on this embodiment provides. It is a structure schematic of a planner terminal. It is a structure schematic of a player terminal. It is a structure schematic of a game server. It is a figure which shows the example of a content of character DB. It is a figure which shows the example of a content of player log DB. It is a figure which shows the example of a display of several statistical data. It is a flowchart which shows the flow of a process of the game server which concerns on this embodiment.

  Hereinafter, embodiments of the present invention will be described.

  FIG. 1 shows a schematic configuration diagram of a game system 10 according to the present embodiment. The game system 10 is a developer of an electronic game provided by the game system 10 (hereinafter simply referred to as “the game”), more specifically, a planner used by a planner who is a creator of an object used in the game. A terminal 12, a player terminal 14 used by a player of the game, and a game server 16 as an information processing apparatus are included. The planner terminal 12 and the game server 16 and the player terminal 14 and the game server 16 are communicably connected to each other via the communication line 18. The communication line 18 is configured by, for example, the Internet or a local area network (LAN). Although one planner terminal 12 and one player terminal 14 are shown in FIG. 1, a plurality of each may be provided.

  In the game system 10, when the planner accesses the game server 16 using the planner terminal 12, the present game including the creation of an object can be developed. Then, the game server 16 transmits various game information to the player terminal 14 in response to the request for providing the main game transmitted from the player terminal 14, whereby the present game is provided.

  Here, the contents of the game will be described. An example of a game screen displayed on the player terminal 14 is shown in FIG. This game is to play a match against an enemy player using a plurality of characters as a plurality of objects. The enemy player may be another player other than the player, or may be a computer (game server 16). Specifically, on the board 20 in which a plurality of squares are defined as a game field, the player and the enemy player alternately arrange the characters in the manner of Othello (registered trademark). In FIG. 2, the rows of each box are indicated by numerals “1” to “6” and the columns by alphabetic characters “A” to “F”.

  The white circles on the board 20 indicate the player's stone (white stone), and the black circles indicate the enemy player's stone (black stone). The character with the character name (for example, "C0055") in the white circle on the board 20 indicates the character of the player (the player character), and the character with the character name in the black circle is the enemy player's Indicates a character (enemy character). When an enemy character or black stone is sandwiched between two self characters or one white stone and one self character, the enemy character or black stone can be changed (turned over) to white stone. When one's own character or white stone is inserted between two enemy characters or one black stone and one enemy character, the own character or white stone is changed to black stone. As in the normal Othello, two white stones and two black stones may be placed on the four central squares of the board 20 at the start of the game. That is, white stones may be placed in 3C and 4D squares, and black stones may be placed in 3D and 4C squares.

  The player is a plurality of characters defined in the present game, and selects a plurality of characters from the characters acquired by the player (available in the present game). For example, in the present embodiment, 16 characters are selected. The sixteen character groups selected by the player are called a deck. The player can acquire a new character according to the progress of the game or by charging. As shown in FIG. 2, four characters selected from the deck are displayed on the screen. The four characters are called a handpiece 22. The handpieces 22 may be randomly selected from the decks by the game server 16 or may be selectable by the player from the decks. The user selects an arbitrary character from the four handpieces 22 and places the selected character in a box on the board 20. When one character out of the handpieces 22 is placed on the board 20, one character selected from the deck is newly added to the handpieces 22.

  Although the details will be described later, a plurality of parameters are set for each character. For example, parameters such as an attack power, a skill effect, and a skill activation condition are set. When the character is placed on the board 20, it is possible to give the opponent a damage according to the offensive power of the character. Alternatively, when the skill activation condition of the character is satisfied, the skill as a special effect of the character is executed, and the game effect according to the skill effect (or the skill effect and the aggression) is activated. .

  An HP (hit point) is set for each of the player and the enemy player. In FIG. 2, it is shown that the maximum HP of the player is 250 and the current HP is 180. An opponent's HP can be reduced by the fact that the character is placed on the board 20 or by executing a skill. Also, HP may be able to recover depending on the skill of the character. And if HP becomes 0 first, it is a loss, the other will win.

  Thus, the game provided by the game system 10 is a game in which the player plays a match against the opponent while using a plurality of characters.

  FIG. 3 shows a schematic configuration of the planner terminal 12. The planner terminal 12 may be, for example, a general computer.

  The control unit 12a includes, for example, a CPU (Central Processing Unit) or a microcomputer, and controls each unit of the planner terminal 12 in accordance with a program stored in a storage unit 12e described later.

  The communication unit 12 b includes, for example, a network adapter, and has a function of communicating with the game server 16 via the communication line 18.

  The input unit 12c includes, for example, a mouse, a keyboard, or a touch panel. The input unit 12 c is for inputting a planner's instruction to the planner terminal 12.

  The display unit 12 d is configured to include, for example, a liquid crystal display and the like, and displays various screens. In particular, a game development screen or the like is displayed on the display unit 12d.

  The storage unit 12 e includes, for example, a hard disk, a random access memory (RAM), or a read only memory (ROM). The storage unit 12e stores a program for operating each unit of the planner terminal 12. Alternatively, various control data or various processing result data may be stored in the storage unit 12e.

  FIG. 4 shows a schematic configuration of the player terminal 14. In the present embodiment, a portable terminal such as a smartphone or a tablet terminal is used as the player terminal 14, but the player terminal 14 may be a stationary computer.

  The control unit 14a includes, for example, a CPU or a microcomputer, and controls each unit of the player terminal 14 according to a client program of a game stored in a storage unit 14e described later.

  The communication unit 14 b includes, for example, a network adapter, and has a function of communicating with the game server 16 via the communication line 18.

  The input unit 14 c includes, for example, a touch panel or a button. The input unit 14 c is for inputting an instruction of the player to the player terminal 14.

  The display unit 14d is configured to include, for example, a liquid crystal display and the like, and is a person who displays various screens. In particular, a game screen or the like is displayed on the display unit 14d.

  The storage unit 14 e includes, for example, a RAM or a ROM. The storage unit 14e stores a game client program. Alternatively, various control data or various game processing result data may be stored in the storage unit 14e.

  A schematic diagram of the game server 16 is shown in FIG.

  The communication unit 30 includes, for example, a network adapter, and has a function of communicating with the planner terminal 12 and the player terminal 14 via the communication line 18.

  The storage unit 32 includes, for example, a hard disk, a RAM, or a ROM. The storage unit 32 stores a game host program as an information processing program. By the host program being executed, the game progresses and the respective units of the game server 16 operate. In addition, as shown in FIG. 5, the storage unit 32 stores a statistical data group 44 including a character DB 34, a player log DB 36, a distributed representation DB 38, an AI log DB 40, an AI agent 42, and a plurality of statistical data 44a. Ru.

  The character DB 34 is a database that is defined in the present game (that is, by the host program) and stores data regarding characters that can be used by the player or the enemy player in the game. Specifically, in the character DB 34, each parameter set for each character is stored. In addition, in this game, it is possible to use thousands of characters, and it is a specification in which new characters are sequentially added.

  An example of the content of the character DB 34 is shown in FIG. In FIG. 6, the character DB 34 is shown in a table format, but the data format of the character DB 34 is not limited to this. In FIG. 6, one record corresponds to one character. In the present embodiment, each character has a character name, an attribute indicating the feature of the character, a rare degree indicating the difficulty of obtaining the character, and a criterion of damage to the opponent when the character is placed on the board 20 The attack power, the skill name, the skill activation condition indicating the condition for activating the skill of the character, and the parameter of the skill effect which is the effect when the skill is activated. Note that “normal attack” in the column of skill effects means that the character is placed on the board 20 and damages the opponent based on the value of the offensive power without activating the skill.

  These parameters are defined by the planner. Of course, the item of parameters is not limited to these, and each character may have parameters for other items.

  As shown in FIG. 6, in the present embodiment, the skill activation condition of each character is, for example, "HP is 5% or less of maximum HP" or "3 or more own characters with attribute P1 on the board" The conditions regarding the game situation are stipulated. Therefore, it is difficult for the planner to properly set the skill effect or the offensive power. Assuming that the skill activation condition is simply given as a probability as a numerical value, it is relatively easy for the planner to set an appropriate skill effect or aggression based on the probability (numerical value) of the skill activation. Become. However, when the skill activation condition is defined in the condition related to the game situation, it is not easy to grasp the activation probability of the skill of the character, and it is difficult to appropriately set the skill effect or the aggression of the new character. doing.

  In addition, even as a player, it is difficult to appropriately select a character to be included in the deck because the skill activation condition of each character is defined by the condition related to the game situation. That is, even as a player, it is not easy to grasp the probability with which the character activates the skill even if the skill activation condition of the character is viewed.

  Returning to FIG. 5, various data are stored in the player log DB 36, the distributed representation DB 38, and the AI log DB 40, respectively. Details of data stored in these databases will be described later.

  The AI agent 42 is an artificial intelligence program that plays the game. The AI agent 42 is learned by a method using deep learning described later. Details of the learning method of the AI agent 42 will be described later. The AI agent 42 in the present embodiment is configured to include a multi-layered neural network. Since the multilayer neural network is defined by various parameters (layer structure, neuron structure in each layer, weight or bias of each neuron, etc.), the storage unit 32 stores these parameters.

  Statistical data 44a is statistical data on the use result of the character. In the present embodiment, the statistical data 44 a is generated for each character, and a plurality of statistical data 44 a corresponding to each character is the statistical data group 44. The statistical data 44a is generated by a statistical data generation unit 64 described later, and the details thereof will be described later.

  The control unit 50 includes, for example, a CPU, a graphics processing unit (GPU), or a microcomputer, and controls each unit of the game server 16 in accordance with a host program of the game stored in the storage unit 32. Further, as shown in FIG. 5, the control unit 50 controls the player log collecting unit 52, the expression learning unit 54, the supervised learning unit 56, the AI battle processing unit 58, the reinforcement learning unit 60, the AI log collecting unit according to the host program. It also functions as the statistical data generator 64 and the display controller 66. Hereinafter, these functions of the control unit 50 will be described.

  The player log collecting unit 52 acquires a player log, which is data indicating a game progress history of a player regarding the game, and stores the player log in the player log DB 36. The player log collecting unit 52 acquires the player log when the game play by the player ends or during the game play. Incidentally, the player referred to here may include the developer of the game, and the player log collecting unit 52 may acquire a log regarding test play by the developer. In any case, the player log is a log related to the actual game play by a human.

  An example of the content of the player log DB 36 is shown in FIG. Although FIG. 7 also shows the player log DB 36 in a table format, the data format of the player log DB 36 is not limited to this. As described above, since the first player and the second player alternately arrange the characters on the board 20 as described above, the number of turns in the player log DB 36 corresponds to the first player in the current turn. Alternatively, a “placement character” indicating the character placed on the board 20 by the second player, a “placement place” indicating the place where the character is placed in the turn, and the number of times the opponent's stone or enemy character was turned over in the turn "Number turned over" indicating that, "Skill activation" indicating whether or not the skill of the character placed in the current turn is activated, and "Skill effect value" indicating the effect (for example, the amount of damage) generated by the activation of the skill Including the item of ". Of course, the items shown in FIG. 7 are an example, and the player log may include other items. For example, items such as the actual damage amount given to the opponent by the normal attack of the character may be included.

  As described above, the player log uses the result ("number of flipped cards", "skill activation", or "when the first player or the second player uses each character (that is, placed on the board 20) when each character is used. "Skill effect value" is included.

  The expression learning unit 54 performs expression learning based on the player log as a game log stored in the player log DB 36. Thereby, a feature vector representing the feature of each character registered in the character DB 34 is learned. That is, the expression learning unit 54 learns the distributed expression of each character used in the present game. The feature vector (distributed expression) of each character learned by the expression learning unit 54 is stored in the distributed expression DB 38. In addition, the expression learning unit 54 uses the AI log as a game log to indicate the game progress history by the AI agent 42 by the AI battle processing unit 58 described later, instead of the player log or in combination with the player log. You may

  The feature vector according to the present embodiment is, for example, a five-dimensional vector, and each element indicates a strategic feature (for example, "I want to put in a corner", "I want to use in the early stage") in this game. it is conceivable that.

  The expression learning unit 54 performs expression learning using, for example, a learning device for expression learning configured by a neural network. Specifically, based on the game state (a stone or the arrangement state of characters on the board 20, etc.) of a player log before a character is used, and the result (next state) that the character is used. , The value of each element of the feature vector of the character is learned.

  As described above, it is possible to represent a large number of characters by lower dimensional vectors by decentralized expression of each character. In addition, characters having similar strategic characteristics in the game will be represented by similar feature vectors. Note that various known learning methods can be used as the expression learning method.

  The supervised learning unit 56 causes the AI agent 42 to learn a strategy regarding the use of each character using at least the player log stored in the player log DB 36 and the feature vector of each character learned by the expression learning unit 54. The strategy relating to the use of each character is a strategy including, in a certain game state, what character the character should be used in, or at which position on the board 20 the character should be placed.

  Specifically, the supervised learning unit 56 determines the feature vector of the character existing on the board or in the hand at the game time, the arrangement position, the number of turns, the action state such as the actionable hand and the hit point, and the action information The AI agent 42 is made to learn using learning data whose output is the evaluation value of the input action information in the input game state, and the like). The evaluation value is calculated in advance based on the player log. That is, the supervised learning unit 56 causes the AI agent 42 to learn by supervised learning.

  In the learning in the supervised learning unit 56, since the learning is performed after each character is expressed by the feature vector, the amount of operation in the learning can be reduced as compared to the case where each character is individually labeled. In addition, since characters having similar characteristics are represented by similar feature vectors, learning on a certain character will perform learning on a character similar to the character. That is, the effect of generalization of learning can be expected.

  The AI battle processing unit 58 performs a process of playing a match between the AI agents 42 learned by the supervised learning unit 56 or further learned by the reinforcement learning unit 60 described later. In the competition between the AI agents 42, the AI agents 42 may be left to select characters to be put into the deck. Further, the AI battle processing unit 58 extracts a trend deck which is a deck frequently used by the user from the player log stored in the player log DB 36, and causes the AI agent 42 to use the trend deck. May be

  In addition, the AI battle processing unit 58 may include a character under development before release, that is, a character under development that can not be used by a general player, in at least one of the decks of the two AI agents 42 who compete.

  The reinforcement learning unit 60 performs reinforcement learning, which is learning to strengthen the AI agent 42. Reinforcement learning is a method different from supervised learning, and for example, Q learning can be used as reinforcement learning. Specifically, in reinforcement learning, the AI agent 42 is rewarded based on the result (for example, winning or losing) of a series of actions selected by the AI agent 42 from a certain game state, and the AI agent 42 Learn the action to be taken based on the reward.

  In the present embodiment, the reinforcement learning unit 60 performs learning in reinforcement learning based on the battle result of the AI agents 42 by the AI battle processing unit 58. The reinforcement learning unit 60 may perform reinforcement learning based on the player log stored in the player log DB 36. In addition, in reinforcement learning, the reinforcement learning unit 60 uses the result of the battle between the AI agent 42 and the NPC (Non Player Character; rule-based AI that is not the AI agent 42) or the battle result between the AI agent 42 and a human player. Learning may be performed on the basis of this.

  In this game, the action that the AI agent 42 can take differs depending on the game state. For example, the position where the character can be arranged differs depending on the state of the board 20, and the character that can be used originally differs depending on the type of the handpiece 22. Therefore, in the present embodiment, reinforcement learning is performed by using a certain game state and the action taken by the AI agent 42 as both inputs and outputting an index indicating the value of the inputted action.

  Also in reinforcement learning, it is preferable to perform learning using the feature vector of each character learned by the expression learning unit 54. Thereby, the same effect as the effect in the supervised learning unit 56 can be obtained in reinforcement learning.

  In the present embodiment, at least one of the supervised learning unit 56 and the reinforcement learning unit 60 corresponds to a strategy learning unit. That is, strategic learning that causes the AI agent 42 to learn a strategy related to the use of each character is either supervised learning or reinforcement learning or both.

  The AI log collection unit 62 indicates the game progress history of the present game between the AI agents 42 by the AI battle processing unit 58, acquires an AI log that is data including use results of a plurality of characters, and stores it in the AI log DB 40 Do. Since the AI log has the same items as the player log, detailed description of the AI log is omitted here. The difference between the player log and the AI log is only whether it indicates the history of the game progress by the player or indicates the history of the game progress by the AI agent 42. The AI log collecting unit 62 may acquire not only logs obtained by the battle between the AI agents 42 but also logs obtained by the battle between the AI agent 42 and the NPC as an AI log.

  When the AI battle processing unit 58 includes the character under development in the deck of the AI agent 42, the AI log collection unit 62 can acquire an AI log including the use result of the character under development. As described above, according to the battle between the AI agents 42, it is possible to obtain a log when the character under development is actually used in the game prior to the release of the character under development.

  The statistical data generation unit 64 generates statistical data 44 a based on the AI log stored in the AI log DB 40. As described above, the statistical data 44a is generated according to each character, and relates to the use result of each character. Of course, when the AI log includes the use result of the character under development, the statistical data generation unit 64 generates statistical data 44a on the use result of the character under development. In the following description regarding the statistical data 44a, the term "character" basically includes characters under development.

  For example, statistical data 44a may include the winning percentage when using a character. In this case, the statistical data generation unit 64 extracts, from the AI log, whether or not the focused character and the AI agent 42 used when the focused character are used have won, and calculates the winning percentage for the focused character. .

  Also, the statistical data 44a may include the firing probability of the skill when using the character. In this case, the statistical data generation unit 64 extracts, from the AI log, whether the focused character and the skill have been activated when the focused character is used, and calculates the skill activation probability for the focused character.

  Also, the statistical data 44a may include the distribution of effects generated when characters are used. In this case, the statistical data generation unit 64 extracts the attention character and the skill effect value when the attention character activates the skill from the AI log, and calculates the distribution of the skill effect value related to the attention character. Alternatively, the effect generated when the character is used may include the amount of damage given to the opponent when the character attacked normally, that is, the statistical data generation unit 64 selects from among the AI log. The target character and the damage given to the opponent by the normal attack of the target character are extracted, and the distribution of the normal attack damage of the target character is calculated.

  As described above, in the present embodiment, the statistical data 44a is at least one of the winning percentage when using a character, the activation probability of a skill when using a character, and the distribution of effects generated when using a character. Contains one. Of course, the statistical data 44a may include other data.

  The statistical data generation unit 64 generates statistical data 44 a for a plurality of characters registered in the character DB 34, ideally, all characters by the above-described processing. A statistical data group 44 is thereby formed.

  The display control unit 66 causes the display unit 12 d of the planner terminal 12 to display the statistical data 44 a generated by the statistical data generation unit 64 in accordance with a request from the planner terminal 12 (that is, from the planner). The display control unit 66 also causes the display unit 14 d of the player terminal 14 to display the statistical data 44 a generated by the statistical data generation unit 64 in accordance with a request from the player terminal 14 (that is, from the player). The display control unit 66 does not cause the display unit 14 d of the player terminal 14 to display the statistical data 44 a regarding the character under development.

  FIG. 8 shows a display example of a plurality of statistical data 44a in the present embodiment. Of the statistical data group 44, statistical data 44a to be displayed may be selected in accordance with an instruction from the planner or the player. For example, to display a plurality of statistical data 44a on skills of a plurality of characters having common parameters for parameter items (rare degree, attribute, skill activation condition, skill effect, kind of skill, etc.) designated by the planner or player. Can. For example, FIG. 8 shows a state in which a plurality of statistical data 44 a for a character having a rare degree “S” in the statistical data group 44 and having a damage skill is displayed. Preferably, the plurality of statistical data 44a are displayed comparably. For example, as shown in FIG. 8, a plurality of statistical data 44a are displayed side by side. The single statistical data 44a selected by the planner or the player may be displayed. In the present specification, the plurality of statistical data 44 a displayed on the planner terminal 12 or the player terminal 14 are collectively referred to as a statistical graph 70.

  In the present embodiment, the statistical graph 70a is displayed literally in the form of a graph. Character names are arranged on the horizontal axis of the statistical graph 70a. In the example of FIG. 8, the plurality of characters are arranged such that the skill activation probability of the leftmost character is the highest, and the skill activation probability decreases from left to right.

  The vertical axis of the statistical graph 70a indicates the skill effect value and the skill activation probability. Specifically, the left vertical axis indicates the skill effect value, and the right vertical axis indicates the skill activation probability. In the display example of FIG. 8, the distribution of the skill effect value of each character is represented by a box-and-whisker plot. Specifically, the lower limit of the trap represents the 5th percentile of the skill effect value (the effect value of 5% from the bottom when the distribution of damage is arranged in descending order), and the lower limit of the box represents the 25th percentile of the skill effect value. The horizontal line in the box represents the median skill effect value, the upper limit of the box represents the 75th percentile of the skill effect value, and the upper limit of the trap represents the 95th percentile of the skill effect value.

  On the other hand, as for the skill activation probability, points are plotted on the graph for each character. In the example of FIG. 8, points of skill activation probabilities for a plurality of characters are connected to form a skill activation probability graph. As described above, in the example of FIG. 8, since the characters are arranged in the order of skill activation probability, it can be understood that the skill activation probability graph shows a lower skill activation probability as it goes to the right.

  Since each character included in the statistical graph 70a shown in FIG. 8 has the same rareness, it is desirable to have the same value. Therefore, from the viewpoint of the skill activation probability and the skill effect value, for these characters, the skill effect value should be smaller as the skill activation probability is higher, and the skill effect value should be larger as the skill activation probability is lower. Conversely, if the skill activation value is high but the skill effect value is high, the character becomes a character capable of frequently activating a strong skill, and the character has a greater value than other characters of the same rare degree. It will be. In addition, if the skill activation value is low even though the skill activation probability is low, the effect of the activated skill becomes poor even though the skill activation probability is low, and the character is lower than other characters with the same rareness. It will be worthless. Thus, if the skill activation probability and the skill effect value are not balanced among the characters having the same value, the game balance is broken.

  In the statistical graph 70 of FIG. 8, the skill effect value is smaller as the skill activation probability is higher, and the skill effect value is larger as the skill activation probability is lower. The parameters of each character are set relatively appropriately. It can be said that it represents On the other hand, for example, when there is a character having a high skill effect value even though the skill activation probability is high or a character having a low skill effect value even though the skill activation probability is low, according to the statistical graph 70, the planner It can be easily grasped that the character (abnormal character) exists. The balance between the skill activation probability of the abnormal character and the skill effect value is not balanced because an inappropriate value (abnormal value) is set to any of the parameters of the abnormal character. Therefore, it can be said that the statistical graph 70 is for detecting an abnormal value possessed by the abnormal character, and the planner can easily grasp the character having the abnormal value by the statistical graph 70. Note that the abnormal value objectively means a value that causes the character to perform a performance (strength) not suitable for the value (in the present embodiment, “rare degree”), and subjectively And means a value that causes the character to exhibit the planner's unintended performance.

  The statistical graph 70 is useful not only for the planner but also for the player. For example, when the player selects a character to be used in the present game (that is, a character to be put into the deck), the statistical graph 70 can be referred to.

  The outline of the game system 10 according to the present embodiment is as described above. According to the present embodiment, after each character is represented by the feature vector by the expression learning unit 54, learning of the AI agent 42 is performed by the supervised learning unit 56 or the reinforcement learning unit 60. This enables the AI agent 42 to learn more efficiently. Specifically, as described above, the reduction of the operation amount for learning by compression of the dimension of the vector representing the character and the generalization of the learning are realized.

  Further, in the present embodiment, statistical data 44 a is generated based on the learned AI agents 42 or an AI log generated by the competition between the AI agent 42 and the NPC. Since the AI agent 42 or the NPC can continuously fight as long as the game server 16 is in operation, it is possible to acquire a large amount of AI logs as compared with the player log. Thereby, the credibility of the statistical data 44a is improved. Furthermore, although the player log can not be acquired because the character under development is not released, it is possible to acquire an AI log regarding the character under development by making the AI agent 42 play a match, that is, statistics about the character under development Data 44a can be generated. This enables the planner to more appropriately set the parameters of the character under development.

  The flow of processing of the game server 16 will be described below according to the flowchart shown in FIG.

  In step S <b> 10, the player log collection unit 52 collects player logs and stores them in the player log DB 36.

  In step S12, the expression learning unit 54 learns the feature vector of each character based on the player log acquired in step S10.

  In step S14, the supervised learning unit 56 performs supervised learning on the AI agent 42, using the feature vector of each character obtained in step S12, based on the player log acquired in step S10.

  In step S16, the AI battle processing unit causes the AI agents 42 learned in step S14 to battle in the game. The reinforcement learning unit 60 performs reinforcement learning to further strengthen the AI agent 42 learned in step S <b> 14 based on the battle result between the AI agents 42. Also, the AI log collection unit 62 acquires an AI log regarding a battle between the AI agents 42. The battle between the AI agents 42 is repeatedly executed, thereby strengthening the AI agent 42 and acquiring a large amount of AI logs.

  In step S18, the statistical data generation unit 64 generates a plurality of statistical data 44a related to the use result of each character based on the AI log obtained in step S16.

  In step S20, in response to a request from the planner terminal 12 or the player terminal 14, the display control unit 66 causes the planner terminal 12 or the player terminal 14 to display the statistical data 44a generated in step S18.

  As mentioned above, although embodiment which concerns on this invention was described, this invention is not limited to the said embodiment, A various change is possible unless it deviates from the meaning of this invention.

  DESCRIPTION OF SYMBOLS 10 game system, 12 planner terminal, 12a, 14a, 50 control part, 12b, 14b, 30 communication part, 12c, 14c input part, 12d, 14d display part, 12e, 14e, 32 storage part, 14 player terminal, 16 game Server, 34 character DB, 36 player log DB, 38 distributed expression DB, 40 AI log DB, 42 AI agent, 44 statistical data group, 44a statistical data, 52 player log collecting unit, 54 expression learning unit, 56 supervised learning unit , 58 AI battle processing unit, 60 reinforcement learning unit, 62 AI log collecting unit, 64 statistical data generating unit, 66 display control unit, 70 statistical graph.

The present invention shows a game progress history of a player regarding an electronic game using a plurality of objects, and learns a feature vector representing a strategic feature of each of the objects based on a game log including use results of the plurality of objects. Using the plurality of objects in an artificial intelligence agent playing the electronic game using at least an expression learning unit, the game log, and the feature vector of each object learned by the expression learning unit And a strategy learning unit for learning

The present invention also provides a computer with a game progress history of a player related to an electronic game using a plurality of objects, and based on a game log including usage results of the plurality of objects, strategic features of each of the objects The artificial intelligence agent playing the electronic game using at least an expression learning unit learning a feature vector to be represented, the game log, and the feature vector of each object learned by the expression learning unit, the plurality of artificial intelligence agents It is an information processing program characterized by causing it to function as a strategy learning unit that learns a strategy related to use of an object.

The present invention shows a game progress history of a player related to an electronic game using a plurality of objects whose effects on the electronic game vary according to the game situation of the electronic game, and a game log including the use results of the plurality of objects. Using at least the expression learning unit for learning the feature vector representing the strategic feature of each of the objects, the game log, and the feature vector of each of the objects learned by the expression learning unit; An information processing apparatus comprising: a strategy learning unit that causes an artificial intelligence agent playing a game to learn a strategy related to the use of the plurality of objects.

Further, the present invention shows a game progress history of a player related to an electronic game using a plurality of objects of which the effects on the electronic game change according to the game situation of the electronic game according to the game situation of the electronic game, and the use results of the plurality of objects An expression learning unit for learning a feature vector representing a feature of each of the objects based on a game log including the game log, the game log, and the feature vectors of the objects learned by the expression learning unit; An information processing program characterized by causing an artificial intelligence agent playing the electronic game to learn a strategy related to the use of the plurality of objects and to function as a strategy learning unit.

Claims (7)

An expression learning unit that learns a game progress history of a player related to an electronic game using a plurality of objects, and learns a feature vector representing a feature of each of the objects based on a game log including a use result of the plurality of objects;
A strategy learning unit that causes an artificial intelligence agent playing the electronic game to learn a strategy related to the use of the plurality of objects using at least the game log and the feature vector of each of the objects learned by the expression learning unit. When,
An information processing apparatus comprising: The strategy learning unit includes at least one of a supervised learning unit that performs supervised learning on the artificial intelligence agent based on the game log, and a reinforcement learning unit that performs reinforcement learning on the artificial intelligence agent. ,
An information processing apparatus according to claim 1, characterized in that. A statistical data generation unit that generates a statistical data on a use result of the object based on an AI log including a game progress history of the electronic game using the learned artificial intelligence agent and including a use result of the plurality of objects When,
A display control unit for displaying the statistical data on a display unit;
The information processing apparatus according to claim 1, further comprising: In the competition of the electronic game using the artificial intelligence agent, an object under development that can not be used by a general player is used,
The AI log contains usage results of the object under development,
The statistical data generation unit generates statistical data on use results of the object under development.
The information processing apparatus according to claim 3, characterized in that: The AI log includes a game progress history of the electronic game by the learned artificial intelligence agents and a use result of the plurality of objects.
The information processing apparatus according to claim 3 or 4, characterized in that: The statistical data includes at least a winning percentage when the object is used, a probability of firing a special effect which can be triggered by the object when the object is used, and a distribution of effects generated when the object is used. Including one,
The information processing apparatus according to any one of claims 3 to 5, characterized in that: Computer,
An expression learning unit that learns a game progress history of a player related to an electronic game using a plurality of objects, and learns a feature vector representing a feature of each of the objects based on a game log including a use result of the plurality of objects;
A strategy learning unit that causes an artificial intelligence agent playing the electronic game to learn a strategy related to the use of the plurality of objects using at least the game log and the feature vector of each of the objects learned by the expression learning unit. When,
An information processing program characterized in that it functions as: