CN106528398A - Game software performance visual analysis method - Google Patents

Abstract

The invention discloses a game software performance visual analysis method. The method comprises the following steps of A, dividing a game software scene into a plurality of meshes; B, mapping frame rate values to the meshes; C, mapping the meshes with different frame rate values to corresponding colors through a thermodynamic diagram, thereby obtaining a frame rate distribution view; and D, based on frame rate value distribution of the meshes, calculating performance difference values of the game software scene in different regions, and drawing a performance comparison view. In combination with visual analysis means such as the game scene division, the thermodynamic diagram and the like, designers and developers can be assisted to locate potential problems in game software and make feasible improvement policies for realizing reasonable resource scheduling/utilization of the game software.

Description

The visual analysis method of Games Software performance

Technical field

The present invention relates to the performance evaluation technical field of computer game software, more particularly to one Plant the visual method for analyzing performance of Games Software.

Background technology

The continuous innovation of hardware is ordered about in the fast development of computer game industry, from desk-top figure Hardware is all constantly being updated to ambulatory handheld game station etc., to support the definition to scene And the higher and higher all kinds of computer games of resolution requirement.There are many factors affect to play Family game experiencing, including resolution, network delay and game render frame per second etc..Investigate these The impact of overall playability of the factor to playing has very important meaning.It is general with what is played And, game designer a can cater to extensive and polynary market environment in the urgent need to designing Game products.Therefore, understand different players are how to interact with game and experience is It is very necessary.

In many factors, game renders that frame per second is the most key, and it determines the entirety of game Stationarity and feeling of immersion.Game developer's raising of often trying one's best renders frame per second, but It is incident, it is to sacrifice resolution as cost, cause the distortion of scene of game.Not only such as This, the multiformity of game hardware (GPU, CPU and RAM etc.) can cause same trip Different numerical value is produced during play operation renders frame per second.

In sum, game developer is maximally utilising to the greatest extent hardware condition, and is based on On the basis of fully understanding that they render frame per second with game, for player provides one it is steadily true to nature Game running environment.With regard to the related visual analysis technology of frame rate of game data and performance evaluation, Mainly can be illustrated in terms of following two：Game performance analysis method and game data can Depending on changing method.

1. game performance research

The technical scheme of 1.1 game performance research prior arts

Resolution, network delay and render many factors such as frame per second and can affect the game body of player Test.Here focus on the impact for rendering frame per second.Many is all to close based on the research for rendering frame per second In terms of video quality, they have studied different frame per second be how to affect user viewing video when Experience.There are some technologies to be to weigh the study that user obtains information from different quality video Ability.However, the discussed video that works before is had very with the computer game of here It is different.Viewing video, even opens video conference etc., and the requirement to real-time, interactive is not all than Upper computer game.Claypool et al. proposes handing over based on different for first person shooting game A kind of unique sorting technique for mutually requiring so that can be observed by the quantitative analyses to frame per second Their impacts to player.These technologies are generally concentrated at the frame per second of research different range What the game experiencing and game for how affecting player was showed.

The shortcoming of 1.2 game performance research prior arts

The Different Effects that the frame per second of different range is produced only are studied in work before, not Overall performance to playing carries out quantitative assessment.The present invention proposes a game performance data Visual analysis framework, depth analysis are integrated into into the solution of complete set with intuitively visualization Scheme, provides a kind of science and quantitative interpretational criteria to analyze game performance.

2. game data visualization technique

The scheme of 2.1 game data visualization techniques

Visualization technique be using computer supported interaction and dynamic vision show displaying number According to internal feature, to obtain new viewpoint and form hypothetical conclusion.Wallner et al. is right It is existing that a literature review has been carried out in terms of game data visualization, however, game data Visualization is still within the comparison primary stage.Design one is gone to the data of a computer game Individual visualization scheme, first have to consider be it towards user.When system towards target use When family is development of games group and Game analysis teacher, from the technology that software visualization derives, than Such as daily record tracking, program structure and operation when behavior etc. may be suitable for.Another and game The new field that visualization is closely related is the visualization technique of sports game information, is whether existed In terms of data, or in terms of targeted customer and the possible concern of user, move and calculate Machine game suffers from many similar features.

The shortcoming of 2.2 game data visualization techniques

Game data major part all has time-space attribute, therefore, many space-time visualization technique quilts It is frequently used in the middle of game data, such as thermodynamic chart etc..Wallner etc. proposes one kind Visual analysis system PLATO, to time correlation and the game data of higher-dimension is analyzed. The many technologies of this system integration, such as subgraph match, path finding, data compare and gather Class and some other visualization techniques.However, the work of Wallner et al. is integrated with too much View, on the contrary to some specific analysis work not deeply.Dinara et al. is using visualization skill The behavior of player in art research computer game, goes to understand wherein potential pattern, and this is System supports cluster and comparative analysiss.However, their visualization scheme has all simply used most base The view of plinth, such as block diagram, pie chart etc., these views are for analysis game performance data also It is far from enough.

And in terms of the exercise data visualization technique related to game, existing work also it is few again It is few.Medler et al. motion and computer game data from the angle analysis of statistical graph, so And, the potentiality for visualizing in this respect are still underestimated, in work few in number, TennisViewer and TenniVis has carried out visual research to tennis match data, and recent Cox and Stasko etc. goes to excavate baseball using the block diagram and tree graph in Information Visualization Technology The information hidden in match, these work distinguish very big with terms of game performance data.

The disclosure of background above technology contents be only used for aid in understand the present invention inventive concept and Technical scheme, which not necessarily belongs to the prior art of present patent application, is not clearly demonstrate,proving According to showing the above in the case of the applying date of present patent application is disclosed, above-mentioned background Technology should not be taken to the novelty and creativeness for evaluating the application.

The content of the invention

It is an object of the invention to a kind of visual method for analyzing performance of Games Software is proposed, to solve What certainly above-mentioned prior art was present can not really help designer's positioning potential problems and then do Go out to improve the technical problem of decision-making.

For this purpose, the present invention proposes a kind of visual analysis method of Games Software performance, including it is as follows Step：

A. by Games Software scene partitioning be multiple grids；

B. frame rate value is mapped to the plurality of grid；

C. it is corresponding color to pass through thermodynamic chart by the mesh mapping with different frame rate values, is obtained Frame per second arrangement view；

D. the frame per second Distribution value based on the plurality of grid, calculates Games Software scene in not same district The performance difference value in domain, and rendering performance compares view.

Preferably, the visual analysis method also includes qualitative assessment Games Software in different tests The step of performance difference in stage, the performance difference is by terms of following formula (1) and formula (2) Obtain,

Wherein, num_mSampled point quantity in expression grid m, N are represented and are sampled in game test k Point quantity sum, n is all number of grids for being included into Performance Calculation, fps_kRepresent that game is surveyed The overall performance frame per second of examination k, fps_mRepresent the performance frame per second of grid m；

Wherein, fps_kRepresent the overall performance frame per second of game test k, fps_k+1Represent game test The overall performance frame per second of k+1, fps_dRepresent the poor performance of game test (k and k+1) twice Different, m, N and k are natural number, 1≤k≤200.

Preferably, the visual analysis method also includes the step of rejecting the deviation value in model.

Preferably, the method for the deviation value in the Rejection of samples includes：

S1. two original game performance difference curve data samples are obtained；

S2. calculate correlation coefficient；

S3. Tectonic facies array；

S4. the change of fischer Z and standardization；

S5. construct variance array；

S6. construct chi-square statisticss amount；

S7. deviation value is checked, such as there is significant difference, then delete corresponding number in initial data Return to step S2 after value；Correlation coefficient is regarded as if it there is no significant difference truly stable.

Preferably, the method for the deviation value in the Rejection of samples includes：

1) assume that two performance difference curves are respectively X=(x₁,x₂,...,x_n) and Y=(y₁,y₂,...,y_n), Pearson correlation coefficients r (X, Y) calculating is carried out first,

Wherein,It is the meansigma methodss of performance difference curve X,It is the meansigma methodss of performance difference curve Y；

2) pairing is deleted and constructs phase relation array

In x₁,x₂,...,x_nAnd y₁,y₂,...,y_nIn, remove i-th respectively (i=1,2 ..., n) to data (x_i,y_i), Remaining (n-1) is designated as into X respectively to sample_i=X x_iAnd Y_i=Y y_i, its correlation coefficient is r_i=r (X_i,Y_i), the phase relation array of sample is designated as (r₁,r₂,...,r_n)；

3) fischer transform is carried out to the phase relation array, is allowed to normal state, then enters rower Standardization so that variance is 1, obtains the phase relation array after normal state standardization (rz₁,rz₂,...,rz_n)；

Now, variance s of phase relation array_rIt is changed into s_r=1；

4) variance array and construction chi-square statisticss amount are constructed, in (rz₁,rz₂,...,rz_n) in, go successively Fall i-th (i=1,2 ..., n) individual value rz_i, the variance of n-1 value under complementationSo as to obtain square margin Group

5) investigate one by one in variance arrayWith s_rWhether=1 difference is notable, null hypothesises H₀With Alternative hvpothesis H₁Respectively：

Construction statistic

Preferably, the sizing grid in step A is set to user's scalable.

Preferably, in step C, low frame rate value is represented according to warm colour and cool colour represents high The mode of frame rate value maps corresponding color.

Preferably, the visual analysis method is also including qualitative assessment Games Software scene performance Step, first, determines the size and sampled point threshold value of the grid, and calculating falls into the grid Average frame per second and hits, then according to formula (3) calculate obtain scene of game performance scores Score,

Wherein, n is all number of grids for meeting condition, fps_iFor the average frame per second of the grid, Sample_num is the number of samples of the grid, and total_sample is that the grid for meeting condition owns Hits summation.

The beneficial effect that the present invention is compared with the prior art includes：By dividing with reference to scene of game And the visual analysis means such as thermodynamic chart, the present invention can help designer and developer's positioning game Potential problem in software, and feasible improvement decision-making is made, to realize that Games Software is rational Scheduling of resource/utilization.

In preferred technical scheme, the performance also to playing carries out qualitative assessment, realizes existing There is the irrealizable qualitative assessment of technology institute.

Description of the drawings

Fig. 1 is the workflow schematic diagram of visual method for analyzing performance of the invention；

Fig. 2 a, 2b, 2c, 2d are frame per second of sizing grid when being 5,10,20,50 respectively Thermodynamic chart (screenshot capture)；

Fig. 3 a, 3b are frame per second scattergram (screenshot capture, the prompt text tested twice respectively It is shown that the hits that last time tests and this is tested in the net region of selection and average frame Rate)；

Fig. 4 is the related-coefficient test method that specific embodiment of the invention interaction is deleted Flow chart；

Fig. 5 a, 5b, 5c, 5d, 5e, 5f, 5g, 5h, 5i are the correlation of initial data respectively Coefficient view, the variance view of initial data, the X 2 test view of initial data, delete Correlation coefficient view, the variance view of initial data, initial data after 1st data X 2 test view, delete correlation coefficient view after the 8th data, initial data The X 2 test view (screenshot capture) of variance view, initial data.

Specific embodiment

Abbreviation and Key Term definition:

GPU：Graphics Processing Unit, Graphics Processing Unit are commonly called as video card；

CPU：Central Processing Unit, central processing unit；

RAM：Random-Access Memory, random access memory are commonly called as internal memory；

MMO-RPG：Massive Multi-player online role-playing game, it is large-scale The online RPG (Role-playing game) of many people；

NPC：Non-player characters, non-player character；

FPS：Frame per second, number of pictures per second are commonly called as frame per second.

As shown in figure 1, being the brief workflow signal of the visual method for analyzing performance of the present embodiment Figure, after loading performance data are finished, frame per second arrangement view can show under all time steps first Frame per second overall distribution situation, and support user mutual so that user arbitrarily can check The frame per second distribution situation in any region on scene of game.In order to contrast the identical trip of different test phases The performance of play scene, the present embodiment provide the user quantitative evaluation in Performance comparision view Standard and assessment result, and support that the interactive iteration of user analyzes process, the interaction meeting again of user Simultaneously affect the model parameter evaluated, iteration several times after, finally given scene of game performance Assessment result.

Thermodynamic chart, can intuitively be presented some and not allow originally readily understood or expression number very much According to such as density, frequency, temperature etc. are used region instead and color is this is easier to be more readily understood Mode presented.The present invention is finally used by distinguishing high-density region and density regions Gradient template determines the density situation of regional.Gradient template is exactly a Ge Cong centers to four The picture of all gray values gradually transition change, is combined on whole background picture using gradient template After coordinate migration, the gradient map needed for just drawing out, on figure, the density situation of regional is just Can have at fingertips.Then color bar is used, it is different according to the value of density, will be from indigo plant to red mistake The rgb value for crossing is mapped in gradient map.Current data is recorded first with a global variable Middle highest frame per second, then when thermodynamic chart is drawn, deducts this highest with this global variable Frame rate value, and corresponding algorithm is adopted, draw out thermodynamic chart.Find in practice, daily record data In often have some abnormity points, such as frame rate value is very high, is close to 200 or so etc., these The introducing of data can cause the uneven (region that this abnormity point is located of the extreme of last thermodynamic chart Summarize substantial amounts of data, and cause other regions very unobvious), it is therefore desirable to handle in advance The carrying out of high frame per second is processed, and is unified into appropriate frame rate value (for example 60 frame) or indivedual numbers According to giving up.

Additionally, thermodynamic chart is denounced by many users with its abstract fuzzy expression all the time, number Value is mapped as information fuzzy caused by color, it is impossible to embody specific numeric distribution.The present invention Then interacted by way of accumulative perception or distributed number figure and be given, so as to macroscopic view can be obtained General picture, also have specific numeric distribution.

For certain of certain a game is once tested, can be with reference to frame per second data and its additional seat Mark data, analyze the low position of frame per second on map, and can be substantially by the general picture of thermodynamic chart Understand the frame per second distribution of whole scene.If however, testing map scene before and after will contrasting twice Performance rise or decline, be which problem cause (engine performance, fine arts resource, Planning playing method, number of players etc.), then it is unknown, if the individually number to testing twice According to performance tracking is carried out, repeat with the aforedescribed process to do twice, can only probably obtain a performance The qualitatively judgement of difference, cannot still determine that problem is located.Based on this, inventor is by frame per second Performance study continues in-depth, with reference to test data is analyzed twice in front and back.For having, note Record test in Same Scene twice under certain coordinate before and after frame per second information twice, i.e., in this coordinate On frame rate value and quantity.Due in most cases, testing twice on same coordinate The record for having frame per second information is considerably less, it is therefore necessary to which map is cut into size one by one Adjustable grid, coordinate is integrated in the controllable grid of this size.Thus, can be by Average frame per second in coordinate points is converted into the average frame per second of grid, adjusts sizing grid every time When, the frame per second information fallen in this grid needs to recalculate, you can obtain the average of the grid Frame per second, is mapped as frame per second thermodynamic chart then.It is illustrated in fig. 2 shown below, is sizing grid difference respectively For 5,10,20,50 when, each coordinate is integrated in its most close grid successively, Ran Houji Calculate the frame per second meansigma methodss for falling into all records of this grid.

Inventor is it is further proposed that technical scheme below：Poor performance i.e. for testing twice It is different, the frame per second data tested twice are shown in a width figure and by those performances rise or under The region of drop is significantly identified.Specifically, it is the frame per second thermodynamic chart phase that will test twice Subtract each other with the pixel value on coordinate, and be mapped to the color bar that two kinds of color transitions are constituted, difference Blue series space is mapped as positive, difference is negative to be mapped as red colour system space, following Fig. 3 It is shown.

The aforementioned feature for broadly having elaborated the present invention and design, so as to more preferably geographical The detailed description of the solution present invention.

With reference to specific embodiment and compare accompanying drawing the present invention is described in further detail. It is emphasized that what the description below was merely exemplary, rather than in order to limit the present invention's Scope and its application.

The visual analysis method of the Games Software performance of the present embodiment, comprises the steps：

A. by Games Software scene partitioning be multiple grids；

B. frame rate value is mapped to the plurality of grid；

C. it is corresponding color to pass through thermodynamic chart by the mesh mapping with different frame rate values, is obtained Frame per second arrangement view；

D. the frame per second Distribution value based on the plurality of grid, calculates Games Software scene in not same district The performance difference value in domain, and rendering performance compares view.

The frame per second arrangement view can show the overall distribution of current frame rate, and support user Interactive adjustment parameter setting, so as to affect final frame per second distribution results.First, the present invention will Scene of game map partitioning is to be made up of the grid of many fixed sizes, then, by different coordinates On frame rate value be mapped to these grids among.The present invention uses average frame rate value to represent The concrete numerical value of each grid, last the grid with different numerical value is reflected by we using thermodynamic chart Penetrate as different colors, warm colour represents low frame rate value and cool colour represents high frame rate value.Due to true In real game, scene of game map is made up of various different size of regions, however, In data plane, the actual size in every piece of region is not we determined that, in order to simplify calculating, I By scene map partitioning for user's adjustable size grid constitute.Therefore, frame per second numerical value meeting Fallen in different grids according to its coordinate figure, therefore, different grids contains quantity not Same sampled point.The grid with statistical significance is included in calculating process, Therefore, we are supplied to user adjust the interactive function of sampled point threshold value.

Additionally, the average frame per second of each grid has been obtained, can adopting according to shared by this grid Sample number accounts for the weight of whole hits, calculates then the Performance Score of the scene.When processing One sampled point threshold value can be set, sampled point quantity relatively low grid is filtered out, factor amount is too Few and no statistical significance.

The performance data difference that Performance comparision view is used in visualization game test twice, it is indicated that Change significantly in which scene areas performance and sample distribution, and provide a kind of quantitative Game performance evaluation methodology, to help customer analysis game performance whether to improve or decline. If we independently analyze the performance data of test of playing twice, perhaps can obtain one compared with For hazy sensations, but still cannot judge which region has obvious performance to become exactly Change, or be accurately positioned potential problem.Therefore, test to quantitatively weigh game twice Performance difference, the present invention formulated a kind of normal form, goes to weigh the performance difference of regional, So as to obtain overall performance difference situation.In view of sampled point quantity in different grids not Together, the overall performance of present invention definition scene can be obtained by such as following formula (1)：

num_mSampled point quantity in expression grid m, N represent the sampled point in game test k Quantity sum, n is all number of grids for being included into Performance Calculation, fps_kRepresent game test The overall performance frame per second of k, fps_mRepresent the performance frame per second of grid m.

Therefore, the performance difference of game test (k and k+1) is obtained by such as following formula (2) twice：

Wherein, fps_kRepresent the overall performance frame per second of game test k, fps_k+1Represent game test The overall performance frame per second of k+1.

Scene performance PTS can be obtained by such as following formula (2)：

Wherein, n is all number of grids for meeting condition, fps_iFor the average frame per second of the grid, Sampl_num is the number of samples of the grid, and total_sample is the grid institute for meeting condition Some hits summations, the performance difference use tested twice " (this test performance score-on Secondary test performance score)/last time test performance score " represent.

As performance difference result is affected with sampled point threshold value by sizing grid, in order to help User finally determines the performance difference of test of playing twice, and we fix a series of sampling thresholds (10,50,100 etc.), the selection of sampling threshold are determined according to the data volume of the scene of game It is fixed, for example successively sizing grid is circulated from 1 to 200, or for some sizes compared with Little scene of game map, sizing grid are circulated from 1 to 100.When sampling threshold is When 10, the scene overall performance frame per second when sizing grid is 1,2,3 grade is calculated respectively, Then performance difference value is calculated according to formula (2), thus, we have obtained a series of by difference The performance difference value that sampling threshold and sizing grid are generated.

However, different sampling thresholds and sizing grid can be unstable to the generation of performance difference result Influence of fluctuations, in order to obtain more true stable performance difference value, we adopt one kind The related-coefficient test method that interaction is deleted (delete, i.e. user's interaction in system interface by interaction Choose a certain deviation point to be deleted), the flow process of empirical method is as shown in figure 4, for twice The performance curve that game test is produced (is carried out to a series of sizing grids under different sampling thresholds The performance difference value that circulation is produced), correlation analysiss are carried out first, if this two poor performances Different curve strong correlation, then they are unrelated with sampling threshold, can be with the corresponding performance of each grid The average of value is representing final performance difference.However, in actual experiment, it is possible to meeting Some deviation values (fluctuate significantly value) occur can affect overall performance difference curve, Therefore, before correlation analysiss are carried out, need to exclude these deviation values.In order to exclude Impact of the deviation value to correlation coefficient, the present embodiment propose a kind of interactive iteration formula and delete with structure The method for making phase relation array, with easy, the objective analysis method of one kind, goes in Rejection of samples Affect correlation coefficient deviation value, so as to objective quantitative ground test samples linearly dependent coefficient it is true Reality and stability, make calculated sample correlation coefficient more true and reliable.

The correlation coefficient empirical method that interaction shown in Fig. 4 is deleted comprises the steps：

It is to obtain two original game performance difference curve data samples 101 (to swim twice first The performance curve that play test is produced), correlation coefficient 102 is then calculated, phase relation array is constructed 103, by the uncommon transform in Fil and standardization 104, variance array 105 is reconstructed, and is constructed Chi-square statisticss amount 106, the method for rejecting construction phase relation array finally by interactive iteration formula are come Determine deviation value 107 (the larger deviation value that fluctuates is determined by interaction), and judge whether There is significant difference 108.As existed, then corresponding numerical value 109 in initial data is deleted, and Being back to above calculating correlation coefficient step carries out the calculating of correlation again；If there is no significance difference It is different, then it is judged as that correlation coefficient is truly stable.

The interactive iteration formula rejects the concrete calculating process of the method for construction phase relation array such as Under：

1) assume that two performance difference curves are respectively X=(x₁,x₂,...,x_n) and Y=(y₁,y₂,...,y_n), Pearson correlation coefficients r (X, Y) calculating is carried out first, whereinIt is performance The meansigma methodss of difference curve X,It is the meansigma methodss of performance difference curve Y：

2) pairing is deleted and constructs phase relation array

In x₁,x₂,...,x_nAnd y₁,y₂,...,y_nIn, remove i-th respectively (i=1,2 ..., n) to data (x_i,y_i), Remaining (n-1) is designated as into X respectively to sample_i=X x_iAnd Y_i=Y y_i, its correlation coefficient is r_i=r (X_i,Y_i), the phase relation array of sample is designated as (r₁,r₂,...,r_n).Then to this correlation coefficient Group carries out fischer transform, is allowed to normal state, then is standardized so that variance is 1 To facilitate follow-up calculating, the phase relation array after normal state standardization is obtained (rz₁,rz₂,...,rz_n), now, variance s of phase relation array_rIt is changed into s_r=1.

3) variance array and construction chi-square statisticss amount are constructed, the undulatory property for investigating variance array is It is no to change.

In phase relation array (rz₁,rz₂,...,rz_n) in, remove i-th successively (i=1,2 ..., n) individual value rz_i(rz_i Represent the correlation coefficient after normal state standardization), the variance of n-1 value under complementation So as to obtain variance arrayThen investigate one by one in variance arrayWith s_r=1 Significantly (i.e. with the presence or absence of significant difference) whether difference.Null hypothesises H₀With alternative hvpothesis H₁Point It is not：

Construction statistic

Then, the present embodiment removes iteration null hypothesises H using interactive mode⁰With alternative hvpothesis H¹ Pairing delete process, after user mutual deletes a deviation value, can again dynamic generate New performance difference curve after deletion, continues to repeat above step until significant deviation Value is present, and the result for finally obtaining is considered reliable and stable correlation coefficient.Determine After final correlation coefficient, performance difference is asked for using the corresponding performance difference value of remaining grid Meansigma methodss as final performance difference value, as shown in Fig. 5 a-5i.

The interactive iteration of the present embodiment deletes quantitative measurement game performance differences method, sampling threshold Respectively 10 and 50, Fig. 5 a be the correlation coefficient view of initial data, Fig. 5 b be original number According to variance view, Fig. 5 c are the X 2 test views of initial data, Fig. 5 d, 5e, 5f point It is not that correlation coefficient view after deleting the 1st data, variance view and X 2 test are regarded Figure, Fig. 5 g, 5h, 5i are correlation coefficient figure after deleting the 8th data, variance respectively View and X 2 test view.After deleting first data, related system view, variance are regarded The fluctuation of figure and X 2 test view diminishes, after deleting the 8th data, the ripple of three views It is dynamic more stable, it was demonstrated that after deleting deviation value, remaining sample is more credible, so as to count The final performance difference for calculating is relatively reliable.

The significant difference (significant difference) is a statistical term, it It is the evaluation statistically to data variance.Its technical standard is such：Normal conditions Under, experimental result reaches 0.05 level or 0.01 level just it may be said that possessing aobvious between data Write difference or extremely notable.When conclusion is done, directivity should be described really and (be for example noticeably greater than Or significantly less than).Sig values (significant difference value) generally use P>0.05 represents diversity not Significantly；0.01<P<0.05 represents that diversity is notable；P<0.01 represents that diversity is extremely notable.This In we be the data that measure in inspection experiment, then it is poor when possessing significance between data Different, the null hypothesises of experiment can be overthrown, and alternative hypothesis is set up；If otherwise not having between data Standby significant difference, the then alternative hypothesis tested can be overthrown, and null hypothesises are supported. Here, level of significance α is given, x is carried out to each value in square margin group successively²Inspection, If certain s_riCorresponding statistic of testIn region of rejection, then refuse null hypothesises, Think in phase relation array (rz₁,rz₂,...,rz_n) in remove rz_iWhen, phase relation array undulatory property In there occurs significant changes, i.e. original sample, data are to (x_i,y_i) to sample correlation coefficient r's Affect than more significant；If all statistic of testAll be present in acceptance region, then it is believed that The undulatory property of phase relation array is not affected by any value, i.e., in phase relation array, nothing peels off Value is present, and sample correlation coefficient r is truly stable, can represent population correlation coefficient.The present invention Be using interactively method go choose deviation value deleted, be not by traditional method, X is given completely²Inspection goes to complete, and can so eliminate the effects of the act, and obtains more true and reliable Sample coefficient.

It would be recognized by those skilled in the art that above description is made numerous accommodations be it is possible, So embodiment is intended merely to describe one or more particular implementations.

Although having been described above the example embodiment of the present invention, it will be understood by those skilled in the art that can be with Which is variously modified and is replaced, without departing from the spirit of the present invention.Furthermore it is possible to do Go out many modifications so that particular case to be fitted to the designs of the present invention, without departing from being described herein Central concept of the present invention.So, the present invention is not only restricted to specific embodiment disclosed here, May also include belonging to all embodiments and its equivalent of the scope of the invention.

Claims (8)

1. a kind of visual analysis method of Games Software performance, it is characterised in that including following step Suddenly：

A. by Games Software scene partitioning be multiple grids；

B. frame rate value is mapped to the plurality of grid；

C. it is corresponding color to pass through thermodynamic chart by the mesh mapping with different frame rate values, is obtained Frame per second arrangement view；

D. the frame per second Distribution value based on the plurality of grid, calculates Games Software scene in not same district The performance difference value in domain, and rendering performance compares view.

2. the visual method for analyzing performance of Games Software as claimed in claim 1, its feature It is：The visual analysis method also includes qualitative assessment Games Software in different test phases The step of performance difference, the performance difference by being calculated with following formula (1) and formula (2),

${\mathrm{fps}}_k=\underset{m=1}{\overset{n}{\&Sigma;}}\frac{{\mathrm{num}}_m}{N}*{\mathrm{fps}}_m---\left(1\right)$

Wherein, num_mSampled point quantity in expression grid m, N are represented and are sampled in game test k Point quantity sum, n is all number of grids for being included into Performance Calculation, fps_kRepresent that game is surveyed The overall performance frame per second of examination k, fps_mRepresent the performance frame per second of grid m；

${\mathrm{fps}}_d=\frac{({\mathrm{fps}}_{k+1}-{\mathrm{fps}}_k)*100}{{\mathrm{fps}}_k}---\left(2\right)$

Wherein, fps_kRepresent the overall performance frame per second of game test k, fps_k+1Represent game test k+1 Overall performance frame per second, fps_dThe performance difference of game test (k and k+1) twice is represented, M, N and k are natural number, 1≤k≤200.

3. the visual method for analyzing performance of Games Software as claimed in claim 2, its feature It is：The visual analysis method also includes the step of rejecting the deviation value in model.

4. the visual method for analyzing performance of Games Software as claimed in claim 3, its feature It is that the method for the deviation value in the Rejection of samples includes：

S1. two original game performance difference curve data samples are obtained；

S2. calculate correlation coefficient；

S3. Tectonic facies array；

S4. the change of fischer Z and standardization；

S5. construct variance array；

S6. construct chi-square statisticss amount；

S7. deviation value is checked, such as there is significant difference, then delete corresponding number in initial data Return to step S2 after value；Correlation coefficient is regarded as if it there is no significant difference truly stable.

5. the visual method for analyzing performance of Games Software as claimed in claim 3, its feature It is that the method for the deviation value in the Rejection of samples includes：

1) assume that two performance difference curves are respectively X=(x₁,x₂,...,x_n) and Y=(y₁,y₂,...,y_n), Pearson correlation coefficients r (X, Y) calculating is carried out first,

$r=r(X,Y)=\frac{\underset{i=1}{\overset{n}{\&Sigma;}}(x_i-\stackrel{\&OverBar;}{x})(y_i-\stackrel{\&OverBar;}{y})}{\sqrt{\underset{i=1}{\overset{n}{\&Sigma;}}{(x_i-\stackrel{\&OverBar;}{x})}^2*\underset{i=1}{\overset{n}{\&Sigma;}}{(y_i-\stackrel{\&OverBar;}{y})}^2}}---\left(4\right)$

Wherein,It is the meansigma methodss of performance difference curve X,It is the meansigma methodss of performance difference curve Y；

2) pairing is deleted and constructs phase relation array

In x₁,x₂,...,x_nAnd y₁,y₂,...,y_nIn, remove i-th respectively (i=1,2 ..., n) to data (x_i,y_i), Remaining (n-1) is designated as into X respectively to sample_i=X x_iAnd Y_i=Y y_i, its correlation coefficient is r_i=r (X_i,Y_i), the phase relation array of sample is designated as (r₁,r₂,...,r_n)；

3) fischer transform is carried out to the phase relation array, is allowed to normal state, then enters rower Standardization so that variance is 1, obtains the phase relation array after normal state standardization (rz₁,rz₂,...,rz_n)；

Now, variance s of phase relation array_rIt is changed into s_r=1；

4) variance array and construction chi-square statisticss amount are constructed, in (rz₁,rz₂,...,rz_n) in, go successively Fall i-th (i=1,2 ..., n) individual value rz_i, the variance of n-1 value under complementationSo as to obtain square margin Group

5) investigate one by one in variance arrayWith s_rWhether=1 difference is notable, null hypothesises H₀With Alternative hvpothesis H₁Respectively：

$\begin{array}{c}{H}_0:s_{r_i}=s_r\\ H_1:s_{r_i}\&NotEqual;s_r\end{array}---\left(5\right)$

Construction statistic ${\mathrm{\&chi;}}_i^2=\frac{(n-1)s_{r_i}}{s_r}.$

6. the visual method for analyzing performance of Games Software as claimed in claim 1, its feature It is：Sizing grid in step A is set to user's scalable.

7. the visual method for analyzing performance of Games Software as claimed in claim 1, its feature It is：In step C, low frame rate value is represented according to warm colour and cool colour represents high frame rate value Mode map corresponding color.

8. the visual method for analyzing performance of Games Software as claimed in claim 1, its feature It is：The step of visual analysis method also includes qualitative assessment Games Software scene performance, First, the size and sampled point threshold value of the grid are determined, calculating falls into the average of the grid Frame per second and hits, then calculate according to formula (3) and obtain scene of game performance scores score,

$score=\underset{i=0}{\overset{n}{\&Sigma;}}{\mathrm{fps}}_i*(sample\_{\mathrm{num}}_i/total\_sample)---\left(3\right)$

Wherein, n is all number of grids for meeting condition, fps_iFor the average frame per second of the grid, Sample_num is the number of samples of the grid, and total_sample is that the grid for meeting condition owns Hits summation.