CN109925718A - A kind of system and method for distributing the micro- end map of game - Google Patents
A kind of system and method for distributing the micro- end map of game Download PDFInfo
- Publication number
- CN109925718A CN109925718A CN201910032178.7A CN201910032178A CN109925718A CN 109925718 A CN109925718 A CN 109925718A CN 201910032178 A CN201910032178 A CN 201910032178A CN 109925718 A CN109925718 A CN 109925718A
- Authority
- CN
- China
- Prior art keywords
- map
- user
- game
- unit
- micro
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
Technical solution of the present invention includes a kind of system and method for distributing the micro- end map of game, for realizing: acquisition game client is uploaded to the user data of server；Training sample is constructed according to user data, in turn, Time series forecasting model is created based on training sample and time recurrent neural network；It obtains user and logins the ground graphic sequence having logged in the period for publish game, user, which is calculated, using Time series forecasting model will enter the sequence number of map, and executes and load in advance.The invention has the benefit that historical data of this method by analysis user's login map, training Time series forecasting model push away user's map that expectation accesses in next step according to the login list of the map of user to be determined is counter；Map is downloaded in advance using the micro- end of game, reduces waiting for downloads the time for user, promotes user experience.
Description
Technical field
The present invention relates to a kind of system, method and devices for distributing the micro- end map of game, belong to computer game field.
Background technique
The miniature client of game (referred to as micro- end) general small volume usually only includes some allocation lists, trip in installation kit
Actor model of playing acts audio, the resources such as map, actor model and the movement special efficacy in new hand village.As long as user spends the short period
Can download installation carry out game experiencing, during experience micro- end can backstage to other resources that do not download (mainly
Map resource) starting downloading, user can with unaware " downloading when playing ".The user experience of this process is preferable, subtracts
The very long waiting process of download games whole resource.This can effectively solve that most of network game client volumes are larger, user
Download the problem that the set-up time is long, login user is low, game user survival rate is low.
In view of map resource packet is more and larger, micro- end downloads resource Bao Shiyou sequencing on backstage and downloading needs
Want the time.When user will use and access the resource downloaded not in time in micro- end installation kit, under micro- end can be into server
It carries the map and is loaded into game.This process will use family and interrupt game operation and the completion that waits for downloads, this is influenced significantly
Ground user experience.How accurately to predict that every user it is expected the map resource used, and allow micro- end download in advance from the background with
User experience is promoted, this is the technical problem of a professional.According to known document, mature technology Accurate Prediction is not every for industry
The access order of position user's to map.Traditional prediction algorithm is generally basede on popular prediction or current map associative classification, i.e.,
The priority of downloading is determined according to the temperature and frequency of use of map, or the pass of map is counted and calculated according to current map
Join the frequency, downloads priority to determine.These methods do not consider that user logs in the timing dependence and user property of map,
Effect is poor in practical application, and accuracy rate is not high.Briefly, it is to have certain purpose that user, which logs in map, for example beats and blames, beats
Copy, collection or completion appointed task etc.；It is relevant that user from a upper map jumps the next map of access, for example connects
New task is received to go to operate to new map.Association between map is designed by game scenario, active task, and player role state etc. is more
A factor influences.Traditional method is difficult to find out the rule of timing from so multifactor.
According to investigation, in the micro- end field of game, industry research, which is concentrated mainly on micro- end, installs and later plays mechanism in downloading
Realization.A kind of for example method and system using the downloading of Script controlling micro- end of CN104503784A- proposes a kind of micro- end
Downloading rule, including establish downloading level mechanism, analysis and optimize resource downloading sequence, subsequent task is prejudged according to current task
Required resource is simultaneously downloaded in advance, merges downloading etc..A kind of on-line loaded method of game resource of CN105988809A- and micro- end
Engine describes on-line loaded method and multiple modules, including game resource pond, update module, resource information listing file, money
Source memory module, receiving module, multithreading download module and pushing module.CN105847429A- is a kind of to realize micro- end downloading money
The method in source provides micro- end design method that account more than one shares interaction.
In map resource prediction, traditional prediction algorithm is generally basede on popular prediction or current map associative classification.
Wherein popular prediction is that counting user logs in the frequency of map and determines the downloading priority of map according to the frequency, i.e., popular
Map is preferentially downloaded.Associative classification is the association frequency logged in map log between statistical map two-by-two from user；When given
The map that user to be determined is currently located, find out with the associated all maps of the map, and according to association frequency sequence, before row
Preferential downloading.These two kinds of methods do not consider that user logs in the timing dependence of map, and effect is poor in practical applications, accuracy rate
It is not high.More than solving the problems, such as, the present invention logs in log by excavating the map of user, is found out using timing mining algorithm
The order of priority rule of access, and the map that will be logged in using every user of the rule Accurate Prediction are jumped between map.This hair
Bright domain generality is very strong, and the method for the present invention all can be used in the field for being related to article access order.
Traditional machine learning method (such as SVM, logistic regression and feedforward neural network) is most of to be all based on independence
The premise of hypothesis all will not carry out explicit model the time.This write method can only handle the input of independence, i.e., previous defeated
Entering with the latter input is absolutely not relationship.But in the data of timing, the input of front and subsequent input are that have
Relationship.For example, when understand in short look like when, the isolated each word for understanding the words be it is inadequate, need to handle these
The entire sequence that word connects, this just has semanteme.A small amount of method (such as depth confidence network) is by node and temporal forerunner
Node is connected with descendant node, by the sliding window of context by time implicit model.But these models are all ignored
The actual conditions relied on when long in time series data, for example with time window size be that 5 models being trained can not be answered and be needed
The problem of wanting 6 time inputs；Cause performance bad.
Some method (such as Markov-chain model) can be answered Time Dependent explicit expression, but due to calculating
It is miscellaneous to spend big and be difficult to practical application.Specifically, Markov-chain model uses dynamic programming algorithm, is on computational complexity
Square of state space size.Secondly, the size of Model transfer table (state of two any point-in-times shifts track probability)
It is square of state space size；If hidden state quantity is larger, computational complexity increases in square number.In addition, hiding shape
The growth of state space increases with the size exponentially of window；Cause computation complexity excessive.
Summary of the invention
The present invention is provided the present invention provides a kind of system and method for distributing the micro- end map of game, using timing recurrence mind
Dependency Specification is compared with Markov-chain model when capturing long through network model, and the current state of this model is dependent on current defeated
Enter and the network state of a upper time step, hidden layer includes the network state information of any time point, nearly equivalent to appointing
The time context for length of anticipating.And the quantity of different conditions can increase and exponential increase with the number of nodes of hidden layer.For example
Each node indicates that N number of hiding node layer may also indicate that 2 n times power state, and expressiveness is superpower using 2 values.When potential
Ability to express growth of index with the growth of number of nodes, is deduced and the complexity of training increases at quadratic power, calculates the time
Complexity is little, has actual application ability.Judging from the experimental results, the present invention program can capture the high discrimination between map
Timing information, and the map that Accurate Prediction player will log in.
Technical solution of the present invention includes a kind of system for distributing the micro- end map of game, which includes data access list
Member, timing training unit and map predicting unit, it is characterised in that: data access unit is uploaded for obtaining game client
To the user data of game server；Timing training unit, for constructing training sample according to user data, in turn, based on instruction
Practice sample and time recurrent neural network creates Time series forecasting model；Predicting unit publishes game for obtaining a login of user
Period in the ground graphic sequence that has logged on, user, which is calculated, using Time series forecasting model will enter the sequence number of map,
And it executes and loads in advance.
According to the system of the power distribution micro- end map of game, data access unit therein is obtained on game client
The user data for reaching game server includes user attribute data and user behavior data, in which: the user attribute data
For the affiliated higher level's attribute data of game role, wherein each micro- end game includes there are many higher level's attribute data, every kind of higher level is swum
Play attribute has including multiple game roles；The user behavior data include once login publish game during accessed
The map id list crossed, wherein sequencing arrangement of the map id list to access.
According to the system of the power distribution micro- end map of game, timing training unit therein is specifically included: training number
According to module, is handled for carrying out a point group to users multiple in game according at least two higher level's attributes, obtain multiple user groups, often
A user group includes multiple users and its corresponding map access sequence；Temporal model module, for accessing every user
Ground graphic sequence inputted, using timing recurrent neural network training input ground graphic sequence obtain for predicting that user will
Access the temporal model of map.
According to the system of the power distribution micro- end map of game, temporal model module therein be further used for executing with
Lower step: the ground graphic sequence that each user is accessed is denoted as { x as input1,x2,...,xt, xtRepresent No. id of map；
Building has the hiding layer state of timing information, is denoted as { h1,h2,...,ht}；Based on hiding layer state htWith input xt, Yi Jiyou
The hiding layer state bring timing in upper layer relies on when long, constructs mapping function output estimation map value, obtains estimating map id value
It is denoted as { y1,y2,...,yt}。
According to the system of the power distribution micro- end map of game, mapping function therein further include: the hiding stratiform
State htBy mapping function fwAccording to hiding layer state ht-1With the x of inputtIt generates, its calculation formula is ht=fw(ht-1,xt), output
ytBy mapping function parameter W and current hiding layer state htIt generates, its calculation formula is yt=Whyht；The mapping function fwAre as follows:
ht=ot⊙tanh(ft⊙ct-1+it⊙tanh(Wxcxt+Whcht-1+bc)), for mapping function fw, wherein ot⊙ tanh () is
Output unit, ft⊙ct-1+it⊙ tanh () is input unit, Wxcxt+Whcht-1+bcTo forget unit.
According to the system of the power distribution micro- end map of game, output unit therein is specifically included: for according to defeated
Enter signal and hiding layer state and conditionally carries out corresponding output, calculation formula ht=ot⊙tanh(ct), specifically: using
Sigmoid layer building ctThe middle information o exportedt, by otIt is mapped between (0,1)；Using tanh layers ctAll regularization to-
Between 1 and 1；To two category information multiplied by weight of above-mentioned steps, result is exported.
According to the system of the power distribution micro- end map of game, input unit therein is specifically included: conditionally determining
The fixed value that internal hiding layer state is updated from input signal specifically is used to generate new candidate value c by tanh layerst(-
1,1), its calculation formula isThe state hidden layer c that will do not updatedt-1And ftIt is multiplied, and deletes extra information；
Add it*ctUnit update is carried out as new content to be added, its calculation formula is
According to the system of the power distribution micro- end map of game, forgetting unit therein is specifically included: for input
And/or output information carries out conditional screening, specifically, according to the output h of last momentt-1With current input xtTo generate
One 0 to 1 vector ft, and with ftDecide whether the information c for allowing last moment to acquiret-1By or part pass through, wherein 0 table
Showing does not allow any information to pass through, and 1 indicates that all information is allowed to pass through, its calculation formula is: ft=σ (Wxcxt+Whcht-1+bc)。
According to the system of the power distribution micro- end map of game, timing training unit unit therein includes model training
Module, it is characterised in that: for use at any time carry out backpropagation BPTT algorithm training is done to model, wherein BPTT algorithm
It is the training algorithm for circulation layer, after training, can get mapping function fwParameter and hidden layer state parameter
{h1,h2,...,ht}。
According to the system of the power distribution micro- end map of game, predicting unit therein specifically includes: model selects mould
Block is judged for docking the user base attribute held in a subtle way, is belonged to according to higher level locating for user base determined property user
Property, and then select corresponding prediction model；Model prediction module, the ground graphic sequence { x for input1,x2,...,xt, timing mould
Type is according to yt=WhyhtCalculate each map candidate y of next possible accesstProbability, the maximum map id of select probability makees
For prediction result.
Technical solution of the present invention further includes a kind of method for distributing the micro- end map of game, method includes the following steps:
Obtain the user data that game client is uploaded to server；Training sample is constructed according to user data, in turn, based on training sample
This and time recurrent neural network create Time series forecasting model；It obtains user and logins and had logged in the period for publish game
Ground graphic sequence, user, which is calculated, using Time series forecasting model will enter the sequence number of map, and executes and load in advance.
The invention has the benefit that this method logs in the historical data of map, training time series forecasting by analysis user
Model logs in the anti-map for pushing away user and it is expected access in next step of list according to the map of user to be determined；It is mentioned using the micro- end of game
Preceding downloading map reduces waiting for downloads the time for user, promotes user experience.
Detailed description of the invention
Fig. 1 show the system framework figure of embodiment according to the present invention；
Fig. 2 show the map access list sample schematic diagram of embodiment according to the present invention；
Fig. 3 show the temporal model training flow chart of embodiment according to the present invention；
Fig. 4 a show recurrent neural network training logic chart；
Fig. 4 b show the neural metwork training sample of embodiment according to the present invention；
Fig. 5 a show the data output y of embodiment according to the present inventiontSchematic diagram；
Fig. 5 b show the mapping function f of embodiment according to the present inventionwSchematic diagram；
Fig. 6 show the predictably map flow chart of embodiment according to the present invention.
Specific embodiment
It is carried out below with reference to technical effect of the embodiment and attached drawing to design of the invention, specific structure and generation clear
Chu, complete description, to be completely understood by the purpose of the present invention, scheme and effect.
It should be noted that unless otherwise specified, when a certain feature referred to as " fixation ", " connection " are in another feature,
It can directly fix, be connected to another feature, and can also fix, be connected to another feature indirectly.In addition, this
The descriptions such as the upper and lower, left and right used in open are only the mutual alignment pass relative to each component part of the disclosure in attached drawing
For system.The "an" of used singular, " described " and "the" are also intended to including most forms in the disclosure, are removed
Non- context clearly expresses other meaning.In addition, unless otherwise defined, all technical and scientific terms used herein
It is identical as the normally understood meaning of those skilled in the art.Term used in the description is intended merely to describe herein
Specific embodiment is not intended to be limiting of the invention.Term as used herein "and/or" includes one or more relevant
The arbitrary combination of listed item.
It will be appreciated that though various elements, but this may be described using term first, second, third, etc. in the disclosure
A little elements should not necessarily be limited by these terms.These terms are only used to for same type of element being distinguished from each other out.For example, not departing from
In the case where disclosure range, first element can also be referred to as second element, and similarly, second element can also be referred to as
One element.The use of provided in this article any and all example or exemplary language (" such as ", " such as ") is intended merely to more
Illustrate the embodiment of the present invention well, and unless the context requires otherwise, otherwise the scope of the present invention will not be applied and be limited.
Fig. 1 show the system framework figure of embodiment according to the present invention.System mainly has three parts composition, is capable respectively
For data access unit 101, timing training unit 102 and map predicting unit 103.Specifically,
Data access unit, this unit be responsible for accessing from client upload onto the server in two class data, it is as follows:
User property: the camp and school (game higher level attribute) etc. of user role in gaming；Different camps and door
The map scene of group, user's experience is different, and needs to do personalized map prediction according to information above.
User behavior: the map id list that user accessed during once logining and publishing game.User is stepping on
The primary experience period can be regarded as to publishing by entering game, collect its logged ground Figure List in this period, list saves
The login precedence of map, this order contain related law.Sample can refer to Fig. 2, this is that prediction user will visit
Ask the essential information source of map.
Timing training unit, attribute and user behaviors log based on user, constructs training sample, for generating a timing
Prediction model.
Map predicting unit, given user login the ground graphic sequence for publishing that the experience period has logged on, this unit at one
It is responsible for predicting next map that it will be accessed.
Fig. 3 show the temporal model training flow chart of embodiment according to the present invention.Attribute and behavior based on user
Training sample is constructed in log, for generating a Time series forecasting model.Training process refers to Fig. 3.Consist of two parts, wraps
Point all living creatures is included into training data and training temporal model.Specifically,
Training data: in view of the scene that the user of different schools and camp is seen is different, the map for needing to access
Resource is also different.In order to more accurately find the access related law of all types of user image resource over the ground, present invention basis first
Camp and school's information are to tenant group.Then construction training data is customized for each group, for training temporal model to have
Effect portrays the order rule that each group user accesses map resource.In an experiment, game has 4 camps and 13 schools, is cut into 4
× 13 user groups.Training data is that each group user is logining the access time sequence table for publishing to map in the experience period, is such as schemed
2。
Temporal model: the present invention uses the time recurrent neural network to map access order modeling with shot and long term memory
And excavate order rule.The network one kind periodically connects to capture order in serialized data and believe by hiding node layer
Breath, can predict the data of serializing.Different from other feedforward neural networks, timing recurrent neural network can save
A kind of state of context, or even can be stored in arbitrarily long contextual window, learn, express relevant information.For example exist
Text field, front and back word is not independent in a sentence；When prediction sentence next word what is, generally require
Know the word of front.Timing recurrent neural network can effectively capture information (i.e. timing) before and after the context between word, quilt
It is widely used in video to understand, the fields such as speech recognition and text-processing, performance is preferable.The present invention borrows this method applied to one
A new field, i.e. the map prediction at the micro- end of game.
Fig. 4 a show recurrent neural network training logic chart.Specifically, the ground graphic sequence each user access is made
For input, it is denoted as { x1,x2,...,xt, xtThe ID number of map is represented, building has the hiding layer state of timing information, is denoted as
{h0,h1,..., ht}；Based on hiding layer state ht and input xt, and the long Shi Yi of layer state bring timing hidden by upper layer
Rely, constructs mapping function output estimation map value, be denoted as { y1,y2,...,yt}.Logic chart refers to Fig. 4 a.Trained process is exactly
Pass through the anti-parameter for releasing mapped function relation of training data and hidden layer state parameter { h0,h1,...,ht}。
Fig. 4 b show the neural metwork training sample of embodiment according to the present invention.With the access map timing number of user
According to as training source, for example " 283,293,102,2,73 ", timing recurrent neural network can construct Function Mapping relationship, reference
Fig. 5 a.Specifically, when user is when ID is 283 map, the map that next expection is gone is 293；Neural network can basis
Input 283 constructs a mapping function to output 293
Fig. 5 a show the data output y of embodiment according to the present inventiontSchematic diagram.Wherein currently hide layer state htBy
Mapping function fW() is according to hiding layer state ht-1With input xtIt generates, specific formula refers to formula 1-1；Export ytBy mapping function
Parameter W and currently hiding layer state htIt generates, specific formula refers to formula 1-2.
ht=fw(ht-1,xt) formula 1-1
yt=WhyhtFormula 1-2
Fig. 5 b show the mapping function f of embodiment according to the present inventionwSchematic diagram.Dependence Problem when in order to solve long,
The present invention uses formula 1-3 as mapping function fW()。
ht=ot⊙tanh(ft⊙ct-1+it⊙tanh(Wxcxt+Whcht-1+bc)), formula 1-3
This mapping function joined one and judge the useful unit of information.The information for only meeting algorithm certification can just be stayed
Under, the information not being inconsistent then passes through forgetting door and passes into silence.Dependence Problem when this Forgetting Mechanism can effectively solve long in timing, it is special
The problem of being not computation complexity.This mapping function consists of three parts, including input unit, forgets unit, and output list
Member.
Forget unit: the unit is responsible for conditionally determining what information thrown away from block, with the extensive energy of lift scheme
Power.According to the output h of last momentt-1With current input xtTo generate one 0 to 1 vector ft, and with ftDecide whether to allow
The information c that one moment acquiredt-1By or part pass through.Wherein 0 indicate " any information is not allowed to pass through ", 1 indicates " to allow all letters
Breath passes through ".With specific reference to formula 1-4
ft=σ (Wxcxt+Whcht-1+bc), formula 1-4
Input unit: the unit is responsible for conditionally determining updating the value of internal hiding layer state from input signal.It is first
Tanh layers are first passed through to be used to generate new candidate value~ct, tanh layers of main function are that numerical value is all grouped between -1 and 1, ginseng
Examine formula 1-5；
Then old state ct-1 is multiplied with ft, some information for being not desired to retain are forgotten about；Then if*~ct is added
Unit update is carried out as new content to be added, with reference to formula 1-6.
Output unit: the unit hides layer state according to the inside of input signal and block and conditionally determines what is exported.
Sigmoid layer building c is used firsttThe middle information o exportedt, wherein sigmoid is often used as the threshold function table of neural network,
It will be between variable mappings to 0,1；Followed by tanh layers ctAll regularization is between -1 and 1；Finally these two types of information weights
Heavy phase multiplies, and output is as a result, participate in formula 1-7.
ht=ot⊙tanh(ct), formula 1-7
The present invention does training to model using progress backpropagation BPTT algorithm at any time.BPTT algorithm is for circulation layer
Training algorithm, including three steps:
1. the output valve of each neuron of forward calculation；
2. the error entry value of each neuron of retrospectively calculate, it is the local derviation of weighting input of the error function E to neuron j
Number；
3. calculating the gradient of each weight, weight finally is updated with stochastic gradient descent algorithm again.
After training, it can get mapping function fWThe parameter and hidden layer state parameter { h of ()0,h1,...,
ht}。
Fig. 6 show the predictably map flow chart of embodiment according to the present invention.Given user logins at one publishes body
The ground graphic sequence that the period has logged on is tested, this unit is responsible for predicting next map that it will be accessed, and process refers to Fig. 6.Needle
To user to be determined, suitable temporal model is selected according to its attribute (including school and camp) of unit 401 first, and utilize
The model gives a forecast, reference unit 402.
Model selection: in view of the map scene that the user of different schools and camp is seen is different, unit 3 is according to each
Group user trains corresponding model correspondingly.In an experiment, game has 4 camps and 13 schools, is cut into 4 × 13
A user group；Each user group trains corresponding temporal model, that is, has 4 × 13 models.First according to user's to be determined
Attribute, i.e., affiliated school and camp, select corresponding temporal model from 4 × 13 models.
Model prediction: the ground graphic sequence { x based on input1,x2,...,xt, temporal model calculates next according to formula 1-2
The each map candidate y that may be accessedtProbability, the maximum map ID of select probability is as prediction result.
Specifically, temporal model thinks that each input data is related with the preceding how many data inputted successively.For example exist
The ground graphic sequence often occurred in training data is " ... ABCDBCEDF ... "；As map " D " that user is accessed, timing mould
Type calculates discovery when the user has accessed map " B " and " C " before access " D ", then prediction output at this time is the general of B
Rate is bigger；When for received data if it is " C " and " E ", prediction output at this time is the general of F to the user before access " D "
Rate is bigger.It is " BCD " when giving the ground graphic sequence that user to be determined is accessed, then temporal model is by calculating next possibility
All maps of access are candidate, find the maximum probability of B, predict that the map results of access are B by the user.
Technical solution of the present invention further provides the compliance test result to present system.This unit verifies the project
Effect, the configuration of experiment and result are as follows:
Verifying configuration: individual forecasting result is applied in the micro- end map distribution of game；It is tested by AB, user is divided equally
At two groups, wherein A group application time series forecasting is as a result, the old download mechanism of B group application (including popular map be associated with map).When
It is primary to calculate hit for the map clock synchronization that prediction result and role will log in.Hit rate verifying is accessed not by statistical map
With the ease for use and user experience situation of mechanism, wherein hit rate is equal to hit-count divided by request number of times.Experience real data
Verifying, the hit rate more popular ground drawing method of inventive algorithm are promoted to 32.5%, and more popular ground drawing method is promoted to 23.4%,
It is significantly better than conventional method.
Applicant of the present invention thinks that this method logs in the historical data of map, training time series forecasting mould by analysis user
Type logs in the anti-map for pushing away user and it is expected access in next step of list according to the map of user to be determined.Further, game is utilized
Map is downloaded at micro- end in advance, reduces waiting for downloads the time for user, promotes user experience.According to known document, industry is without maturation
Method Accurate Prediction map access order.Traditional prediction algorithm is generally basede on popular prediction or map association analysis,
These methods do not consider that user logs in the timing dependence and user property of map, and predictablity rate is not high.Not with conventional method
Together, the present invention makes full use of the precedence rule accessed between map, and the precision of prediction is promoted using the order rule, is knowing
Other accuracy rate is obviously improved than conventional method.In an experiment, the present invention tests in the historical data that map accesses, prediction
Hit rate average out to 86.5%.The present invention is applied to the result of prediction in micro- end map individual character distribution of game, to every use
Its map that will access of family individual forecasting, hit rate more popular ground drawing method are promoted to 32.5%, and more popular ground drawing method mentions
It is upgraded to 23.4%, business Huge value.
It should be appreciated that the embodiment of the present invention can be by computer hardware, the combination of hardware and software or by depositing
The computer instruction in non-transitory computer-readable memory is stored up to be effected or carried out.Standard program skill can be used in method
Art-includes that the non-transitory computer-readable storage media configured with computer program is realized in computer program, wherein such as
Storage medium of this configuration operates computer in a manner of specific and is predefined --- according to describing in a particular embodiment
Method and attached drawing.Each program can be realized with the programming language of level process or object-oriented with logical with computer system
Letter.However, if desired, the program can be realized with compilation or machine language.Under any circumstance, the language can be compiling or
The language of explanation.In addition, the program can be run on the specific integrated circuit of programming for this purpose.
In addition, the operation of process described herein can be performed in any suitable order, unless herein in addition instruction or
Otherwise significantly with contradicted by context.Process described herein (or modification and/or combination thereof) can be held being configured with
It executes, and is can be used as jointly on the one or more processors under the control of one or more computer systems of row instruction
The code (for example, executable instruction, one or more computer program or one or more application) of execution, by hardware or its group
It closes to realize.Computer program includes the multiple instruction that can be performed by one or more processors.
Further, method can be realized in being operably coupled to suitable any kind of computing platform, including but
It is not limited to PC, mini-computer, main frame, work station, network or distributed computing environment, individual or integrated meter
Calculate machine platform or communicated with charged particle tool or other imaging devices etc..Each aspect of the present invention can be to be stored in
No matter machine readable code on non-transitory storage medium or equipment is moveable or is integrated to calculate and put down to realize
Platform, such as hard disk, optically read and/or write-in storage medium, RAM, ROM, so that it can be read by programmable calculator, when depositing
Storage media or equipment can be used for configuration and operation computer to execute process described herein when being read by computer.In addition,
Machine readable code, or part thereof can be transmitted by wired or wireless network.When such media include in conjunction with microprocessor or
When other data processors realize the instruction or program of above step, the disclosure herein includes that these and other are different types of non-
Temporary computer readable storage medium.When programming according to the method for the present invention with technology, the invention also includes computer sheets
Body.
Computer program can be applied to input data to execute functions herein, be deposited to convert input data with generating
It stores up to the output data of nonvolatile memory.Output information can also be applied to one or more output equipment such as displays.
In the preferred embodiment of the invention, the data of conversion indicate physics and tangible object, including the physics generated on display
Describe with the particular visual of physical objects.
More than, only presently preferred embodiments of the present invention, the invention is not limited to above embodiment, as long as its with
Identical means reach technical effect of the invention, all within the spirits and principles of the present invention, any modification for being made, equivalent
Replacement, improvement etc., should be included within the scope of the present invention.Its within the scope of the present invention technical solution and/
Or embodiment can have a variety of different modifications and variations.
Claims (11)
1. a kind of system for distributing the micro- end map of game, which includes that data access unit, timing training unit and map are pre-
Survey unit, it is characterised in that:
Data access unit, the user data for being uploaded to game server for obtaining game client；
Timing training unit in turn, is based on training sample and time recurrent neural for constructing training sample according to user data
Network creation Time series forecasting model；
Predicting unit logins the ground graphic sequence having logged in the period for publish game for obtaining user, pre- using timing
User, which is calculated, in survey model will enter the sequence number of map, and executes and load in advance.
2. the system of the distribution micro- end map of game according to claim 1, which is characterized in that the data access unit obtains
Taking game client to be uploaded to the user data of game server includes user attribute data and user behavior data, in which:
The user attribute data is the affiliated higher level's attribute data of game role, wherein each micro- end game includes that there are many higher levels
Attribute data, every kind of higher level's game attributes have including multiple game roles；
The user behavior data include once login publish game during the map id list that accessed, wherein map
Sequencing arrangement of the id list to access.
3. the system of the distribution micro- end map of game according to claim 1, which is characterized in that the timing training unit tool
Body includes:
Training data module is handled for carrying out a point group to users multiple in game according at least two higher level's attributes, is obtained more
A user group, each user group include multiple users and its corresponding map access sequence；
Temporal model module, the ground graphic sequence for accessing every user are inputted, are instructed using timing recurrent neural network
The ground graphic sequence for practicing input is obtained for predicting that user will access the temporal model of map.
4. it is according to claim 1 distribution the micro- end map of game system, which is characterized in that the temporal model module into
One step is for executing following steps:
The ground graphic sequence that each user is accessed is denoted as { x as input1,x2,...,xt, xtRepresent No. id of map；
Building has the hiding layer state of timing information, is denoted as { h1,h2,...,ht}；
Based on hiding layer state htWith input xt, and by upper layer hide layer state bring timing it is long when dependence, building mapping letter
Number output estimation map values, obtain estimating map id value and are denoted as { y1,y2,...,yt}。
5. the system of the distribution micro- end map of game according to claim 4, which is characterized in that the mapping function also wraps
It includes:
The hiding layer state htBy mapping function fwAccording to hiding layer state ht-1With the x of inputtIt generates, its calculation formula is ht
=fw(ht-1,xt), export ytBy mapping function parameter W and current hiding layer state htIt generates, its calculation formula is yt=Whyht；
The mapping function fwAre as follows: ht=ot⊙tanh(ft⊙ct-1+it⊙tanh(Wxcxt+Whcht-1+bc)), for mapping function
fw, wherein ot⊙ tanh () is output unit, ft⊙ct-1+it⊙ tanh () is input unit, Wxcxt+Whcht-1+bcTo forget
Unit.
6. the system of the distribution micro- end map of game according to claim 4, which is characterized in that the output unit specifically wraps
It includes:
For conditionally carrying out corresponding output, calculation formula h according to input signal and hiding layer statet=ot⊙tanh
(ct), specifically:
Use Sigmoid layer building ctThe middle information o exportedt, by otIt is mapped between (0,1)；
Using tanh layers ctAll regularization is between -1 and 1；
To two category information multiplied by weight of above-mentioned steps, result is exported.
7. the system of the distribution micro- end map of game according to claim 4, which is characterized in that the input unit specifically wraps
It includes:
Conditionally determine that the value that internal hiding layer state is updated from input signal specifically is used to generate by tanh layers
New candidate value ct(- 1,1), its calculation formula is
The state hidden layer c that will do not updatedt-1And ftIt is multiplied, and deletes extra information；
Add it*ctUnit update is carried out as new content to be added, its calculation formula is
8. the system of the distribution micro- end map of game according to claim 4, which is characterized in that the forgetting unit specifically wraps
It includes:
For carrying out conditional screening to input and/or output information, specifically, according to the output h of last momentt-1With work as
Preceding input xtTo generate one 0 to 1 vector ft, and with ftDecide whether the information c for allowing last moment to acquiret-1By or part
Pass through, wherein 0 indicates that any information is not allowed to pass through, 1 indicates that all information is allowed to pass through, its calculation formula is: ft=σ
(Wxcxt+Whcht-1+bc)。
9. the system of the distribution micro- end map of game according to claim 3 or 4, the timing training unit unit includes mould
Type training module, it is characterised in that:
Backpropagation BPTT algorithm is carried out at any time training is done to model, wherein BPTT algorithm is for circulation layer for using
Training algorithm, after training, can get mapping function fwParameter and hidden layer state parameter { h1,h2,...,
ht}。
10. the system of the distribution micro- end map of game according to claim 1, which is characterized in that the predicting unit is specific
Include:
Model selection module is judged for docking the user base attribute held in a subtle way, is used according to user base determined property
Higher level's attribute locating for family, and then select corresponding prediction model；
Model prediction module, the ground graphic sequence { x for input1,x2,...,xt, temporal model is according to yt=WhyhtIt calculates next
Each map candidate y of a possible accesstProbability, the maximum map id of select probability is as prediction result.
11. a kind of method for distributing the micro- end map of game, which is characterized in that method includes the following steps:
Obtain the user data that game client is uploaded to server；
Training sample is constructed according to user data, in turn, time series forecasting is created based on training sample and time recurrent neural network
Model；
It obtains user and logins the ground graphic sequence having logged in the period for publish game, be calculated using Time series forecasting model
User will enter the sequence number of map, and execute and load in advance.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910032178.7A CN109925718B (en) | 2019-01-14 | 2019-01-14 | System and method for distributing game micro-terminal map |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910032178.7A CN109925718B (en) | 2019-01-14 | 2019-01-14 | System and method for distributing game micro-terminal map |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109925718A true CN109925718A (en) | 2019-06-25 |
| CN109925718B CN109925718B (en) | 2023-04-28 |
Family
ID=66984921
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910032178.7A Active CN109925718B (en) | 2019-01-14 | 2019-01-14 | System and method for distributing game micro-terminal map |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109925718B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110841280A (en) * | 2019-11-06 | 2020-02-28 | 福建天晴在线互动科技有限公司 | Method for online dynamic loading and checking of resources of micro-terminal |
| CN111729305A (en) * | 2020-06-23 | 2020-10-02 | 网易（杭州）网络有限公司 | Map scene preloading method, model training method, device and storage medium |
| CN112947969A (en) * | 2021-03-10 | 2021-06-11 | 支付宝(杭州)信息技术有限公司 | Page off-screen rendering method, device, equipment and readable medium |
| CN114791893A (en) * | 2021-12-15 | 2022-07-26 | 许磊 | Serialization system for random data access |
| CN114912082A (en) * | 2021-12-15 | 2022-08-16 | 许磊 | General computing task collaboration system |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101358589A (en) * | 2003-11-24 | 2009-02-04 | 麦克罗斯塔克公司 | Microvalve device suitable for controlling a variable displacement compressor |
| CN104503784A (en) * | 2014-12-16 | 2015-04-08 | 珠海金山网络游戏科技有限公司 | Method and system for controlling micro-client downloading by using script |
| US20150365350A1 (en) * | 2014-06-12 | 2015-12-17 | Netease (Hangzhou) Network Co., Ltd. | Resource downloading method and apparatus |
| CN107198884A (en) * | 2017-05-26 | 2017-09-26 | 合肥充盈信息科技有限公司 | A kind of method that scene of game is efficiently loaded into |
| CN108764090A (en) * | 2018-05-18 | 2018-11-06 | 腾讯大地通途（北京）科技有限公司 | It is a kind of regionality unusual fluctuation determine method, apparatus, server and storage medium |
-
2019
- 2019-01-14 CN CN201910032178.7A patent/CN109925718B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101358589A (en) * | 2003-11-24 | 2009-02-04 | 麦克罗斯塔克公司 | Microvalve device suitable for controlling a variable displacement compressor |
| US20150365350A1 (en) * | 2014-06-12 | 2015-12-17 | Netease (Hangzhou) Network Co., Ltd. | Resource downloading method and apparatus |
| CN104503784A (en) * | 2014-12-16 | 2015-04-08 | 珠海金山网络游戏科技有限公司 | Method and system for controlling micro-client downloading by using script |
| CN107198884A (en) * | 2017-05-26 | 2017-09-26 | 合肥充盈信息科技有限公司 | A kind of method that scene of game is efficiently loaded into |
| CN108764090A (en) * | 2018-05-18 | 2018-11-06 | 腾讯大地通途（北京）科技有限公司 | It is a kind of regionality unusual fluctuation determine method, apparatus, server and storage medium |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110841280A (en) * | 2019-11-06 | 2020-02-28 | 福建天晴在线互动科技有限公司 | Method for online dynamic loading and checking of resources of micro-terminal |
| CN111729305A (en) * | 2020-06-23 | 2020-10-02 | 网易（杭州）网络有限公司 | Map scene preloading method, model training method, device and storage medium |
| CN111729305B (en) * | 2020-06-23 | 2024-02-09 | 网易（杭州）网络有限公司 | Map scene preloading method, model training method, device and storage medium |
| CN112947969A (en) * | 2021-03-10 | 2021-06-11 | 支付宝(杭州)信息技术有限公司 | Page off-screen rendering method, device, equipment and readable medium |
| CN114791893A (en) * | 2021-12-15 | 2022-07-26 | 许磊 | Serialization system for random data access |
| CN114912082A (en) * | 2021-12-15 | 2022-08-16 | 许磊 | General computing task collaboration system |
| CN114791893B (en) * | 2021-12-15 | 2023-05-09 | 许磊 | Serialization system for random data access |
| CN114912082B (en) * | 2021-12-15 | 2023-05-09 | 许磊 | General computing task collaboration system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109925718B (en) | 2023-04-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109925718A (en) | A kind of system and method for distributing the micro- end map of game | |
| CN110119844B (en) | Robot motion decision method, system and device introducing emotion regulation and control mechanism | |
| CN111282267B (en) | Information processing method, information processing apparatus, information processing medium, and electronic device | |
| CN110366734A (en) | Optimization neural network framework | |
| CN108701265A (en) | Learning Service provides device | |
| CN111242310B (en) | Feature validity evaluation method and device, electronic equipment and storage medium | |
| CN107241213A (en) | A kind of web service composition method learnt based on deeply | |
| Käser et al. | When to stop? Towards universal instructional policies | |
| Jia | Efficient computing budget allocation for simulation-based policy improvement | |
| CN109784959A (en) | A kind of target user's prediction technique, device, background server and storage medium | |
| EP4290351A1 (en) | Environment modeling method and apparatus based on decision flow graph, and electronic device | |
| CN111494964B (en) | Virtual article recommendation method, model training method, device and storage medium | |
| US20210097443A1 (en) | Population-based training of machine learning models | |
| CN107273979A (en) | The method and system of machine learning prediction are performed based on service class | |
| CN112070310A (en) | Loss user prediction method and device based on artificial intelligence and electronic equipment | |
| Watts et al. | Does cumulative advantage affect collective learning in science? An agent-based simulation | |
| CN112163671A (en) | New energy scene generation method and system | |
| CN113469372A (en) | Reinforcement learning training method, device, electronic equipment and storage medium | |
| CN112990958A (en) | Data processing method, data processing device, storage medium and computer equipment | |
| Argerich et al. | Tutor4RL: Guiding Reinforcement Learning with External Knowledge. | |
| Mazzolini et al. | Heaps' law, statistics of shared components, and temporal patterns from a sample-space-reducing process | |
| CN112245934B (en) | Data analysis method, device and equipment for virtual resources in virtual scene application | |
| Xu et al. | Meta-learning via weighted gradient update | |
| CN113761355A (en) | Information recommendation method, device, equipment and computer readable storage medium | |
| Needle | Management strategy evaluation for North Sea haddock |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| CB02 | Change of applicant information | ||
| CB02 | Change of applicant information |
Address after: 519080 Room 102, 202, 302, 402 Room 327 Room 102, 202 Room 329 Room 302, 325 Qiandao Ring Road, Tangjiawan Town, High-tech Zone, Zhuhai City, Guangdong Province Applicant after: Zhuhai Jinshan Digital Network Technology Co.,Ltd. Address before: 519000 building 3, Jinshan Software Park, 325 Qiandao Ring Road, Xiangzhou District, Zhuhai City, Guangdong Province Applicant before: ZHUHAI KINGSOFT ONLINE GAME TECHNOLOGY Co.,Ltd. |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |