CN110222239A - A kind of pattern generation of structurally variable and method using pattern generation browser interface - Google Patents
A kind of pattern generation of structurally variable and method using pattern generation browser interface Download PDFInfo
- Publication number
- CN110222239A CN110222239A CN201910425318.7A CN201910425318A CN110222239A CN 110222239 A CN110222239 A CN 110222239A CN 201910425318 A CN201910425318 A CN 201910425318A CN 110222239 A CN110222239 A CN 110222239A
- Authority
- CN
- China
- Prior art keywords
- pattern
- relationship
- vector
- value
- relational graph
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/901—Indexing; Data structures therefor; Storage structures
- G06F16/9024—Graphs; Linked lists
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T11/00—2D [Two Dimensional] image generation
- G06T11/20—Drawing from basic elements, e.g. lines or circles
- G06T11/206—Drawing of charts or graphs
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Abstract
The invention discloses a kind of pattern generation of structurally variable and utilize the method for pattern generation browser interface, the method for generating pattern, comprising the following steps: S1, construct relationship graph model for the element in pattern;S2, the discrete optimization model using energy equation carry out the matching of element between different topology structure pattern;S3, it is based on RJMCMC algorithm, is sampled between different topology structure pattern, obtains new pattern.The method using pattern generation browser interface, comprising the following steps: 1), obtain the high dimension vector expression characteristic for sampling obtained new pattern, obtain high-dimensional vector space;The high position vector expression characteristic is indicated using the relational graph vector of the pattern;2) high-dimensional vector space, is compressed to by two-dimensional manifold space using GPLVM;3), by the point inverse mapping in two-dimensional manifold space into higher dimensional space, new pattern is obtained, and post-process to it;4) pattern browser interface, is generated.
Description
Technical field
The present invention relates to computer pattern generations and pattern to browse field, raw more particularly, to a kind of pattern of structurally variable
At the method with utilization pattern generation browser interface.
Background technique
The algorithm and system of existing maturation quickly and easily can be gathered around much according to the acquisition of the morphosis of single pattern
There are different element sizes, position, the variant candidate in direction.Such as paper PATEX:exploring pattern variations
In, author is described pattern using the geometrical relationship between patterned member and meets different geometrical relationships respectively by sampling those
The pattern of subset obtain many pattern variants based on source pattern.However such method is only in single pattern variant,
Some features being both able to maintain in some not homologous patterns can not be automatically created between multiple patterns or can bring one
The pattern of a little new changes.
In addition for different patterns, element and arrayed feature between them all may be far from each other.At two
Can also there be countless pattern variants between simple pattern.Have so discrepant new pattern result need as much as possible by
It samples.Finally, lack in current design tool it is some can allow designer in the same set in extremely rapid succession
Browse to all these elements and the different pattern variant of morphosis.
Summary of the invention
The present invention provides a kind of the system of some new patterns is sampled out between two given patterns and provide one
Kind generates the interactive browse interface of pattern.These new patterns be both able to maintain some local features in the pattern of source, can also have
Some variations, even if the number of elements between these source patterns as reference, arrangement of elements feature may have otherness.If
Meter teacher can browse to all these elements and the different pattern result of form in interface.Present invention mainly solves as
The problem of what identifies these respective morphological features of different topology structure pattern and is used and how by arrayed feature,
The problem of being sampled in different spaces set composed by all discrepant new pattern of number of elements.A kind of figure is provided simultaneously
Case interactive browse interface can allow designer optionally to browse all pattern variants that may be searched.
The present invention is realized at least through one of following technical solution.
A kind of method for generating pattern of structurally variable, comprising the following steps:
S1, relationship graph model is constructed for the element in pattern;
S2, the discrete optimization model using energy equation carry out the matching of element between different topology structure pattern;
S3, RJMCMC algorithm (Reverse Jump Markov Chain Monte Carlo, reversible jump Ma Er are based on
Section husband chain Monte-Carlo algorithm), it is sampled between different topology structure pattern, obtains new pattern.
Further, relationship described in step S1 includes the relationship type of element and the relationship type of relationship, wherein element
Relationship type include the distance between element and element relationship, element and element angular separation relationship, element and element
Dimension scale relationship;The relationship type of relationship includes that the distance kept between element is equal difference or waits than certain in relationship, pattern
Angled relationships between one element and the direction of other elements.
Further, the relationship graph model of step S1 building is for describing relationship and relationship between the element in pattern
Between relationship;
Relationship graph model includes relational graph vector, and a relational graph vector represents a pattern;Relational graph vector is one
One-dimensional vector is made of three parts, wherein first part be successively the position (x, y) of each element in pattern, towards θ and
These three information of size s, the second part include the value of relationship between element and element, and third part is relation value r, described
Relation value r includes the relation value of element and element and the relation value of relationship, and relational graph vector μ is as follows:
μ=(x1,y1,θ1,s1,x2,y2,θ2,s2,...,xi,yi,θi,si,r1,r2,r3,...,ri) (1)
Relationship graph model passes through digraphIt is indicated, the node N=E ∪ R of digraph, wherein E is figure
The set of element in case, i-th of element representation are Ei=(xi,yi,θi,si), Ei∈ E, i indicate the quantity of element;R is in pattern
The set of relationship, R=RE∪RR, REIt is element relation set, RRIt is the set of relationship between relationship, each relationship has accordingly
Relation value, wherein the relation value of i-th of relationship is expressed as Ri=(ri), Ri∈R;By node NiCorresponding relationship vector owns
Element information or relation value be denoted as AndWherein Ej∈E;Side the A={ (N of digraphi1,Ri),(Ni2,Ri)}I=1...k+l, wherein k be
The quantity of element relation in relationship graph model, l are the relationship quantity of relationship.Such as Ni1With node Ni2Be with relation value RiIt is related
Element E1With element E2Or relationship R1With relationship R2, wherein RiValue by node Ni1With node Ni2Value be calculated;
Specifically, specifically, the relationship type of element shares 6 kinds, and the relationship type of relationship has 3 kinds, wherein 6 kinds of elements
Relationship type are as follows:
1) a, element E in representative patternaWith another element E in representative patternbThe distance at center be defined
Euclidean distance relationship between element, wherein relation value RiValue are as follows:
Wherein, (xa,ya) it is element EaPosition, (xa,yb) it is element EbPosition;
2) a, element E in representative patternaWith another element E in representative patternbDirection between differential seat angle
The direction difference relationship being defined as between element, relation value RiValue range be [- π, π], clockwise angle change is negative
Number:
Wherein, θaAnd θbRespectively EaAnd EbDirection;
3) a, element E in representative patternaWith another element E in representative patternbSize between difference determined
Size difference relationship of the justice between element, relation value Ri:
Ri=sa-sb
Wherein, SaAnd SbRespectively EaAnd EbSize;
4) a, element E in representative patternaWith another element E in representative patternbCentral point between line
Angle between X-axis is defined as absolute angle difference relationship, relation value RiValue range be [- π, π], clockwise angle
Variation is negative:
5) a, element E in representative patternaWith another element E in representative patternbCentral point between line
With element EaDirection between angle be defined as the relative angular difference relationship between element, relation value RiValue range be
[- π, π], clockwise angle change are negative:
6) a, element E in representative patternaWith another element E in representative patternbCentral point between line
With element EaDirection between the absolute value of angle be defined as the symmetry angle difference relationship between element, relation value RiTake
Being worth range is [0, π], and angle change clockwise and anticlockwise is positive number:
The relationship type of 3 kinds of relationships is respectively as follows:
3-1), two be not angular relationship relationship EvAnd Ed, value RvAnd RdDifference be defined as the relationship of relationship difference,
Relation value RiValue:
Ri=Rv-Rd
3-2), two angular relationship EvAnd EdValue RvAnd RdDifference be defined as the relationship of angular relationship difference, relation value Ri
Value range be [0, π]
3-3), two relationship EvAnd EdValue RvAnd RdQuotient be defined as the relation value R of relationship quotienti:
Ri=Rv/Rd
Further, the discrete optimization model described in step S2 based on energy equation is as follows:
So that
Wherein, m and n is the element number for carrying out the pattern a and pattern b of Match of elemental composition, X respectivelyijIt is for specifying one
Whether i-th of element in pattern be corresponding with j-th of element of another pattern, i ∈ m, j ∈ n;XijFor 1 indicate element i and
There are corresponding relationships by element j, are otherwise not present;VijWork as X for specifiedijWhen being 1, correspondence bring consumption;Element and element
Between corresponding consumption be defined as Euclidean distance between two elements;λ is weight, before balancing in discrete optimization model
OneWith latter
Latter for avoid two in a pattern symmetrical elements respectively correspond in other patterns two it is not right
Therefore the pattern of title establishes a list, each single item in list is that four element (p, q, g, h), wherein p and g indicate figure
Pth and q-th of element in case a are symmetric relations, and q and h indicate that q and h-th of element in pattern a are symmetric relations;If
The length of the list be K, then in latter k-th of list calculating formula SkIs defined as:
Wherein XP, gIt is, X whether corresponding with q-th of element in pattern b for p-th of element in given pattern aQ, hIt is
It is whether corresponding with h-th of element of pattern b for q-th of element in given pattern a,It is xor operation;
The discrete optimization model needs to meet three following constraint conditions simultaneously,Indicate every in pattern a
One element at least will there are corresponding relationships with an element in pattern b;Then indicate each of pattern b member
Element also at least will there are corresponding relationships with an element in pattern a;It then indicates when in pattern a
I-th element and pattern b in j-th of element generate corresponding relationship after, if in i-th of element and pattern b in pattern a
Be not that the other elements of j-th of element generate correspondence, then j-th of element in pattern b cannot again with no in pattern a the
The other elements of i element generate correspondence;If in j-th of the element and pattern a in same pattern b is not i-th of element
Other elements generate correspondence, then i-th of element in pattern a cannot again with other yuan of no in pattern b j-th of element
Element generates correspondence.
Further, step S3 had both been able to maintain between the pattern of different topology structure using RJMCMC sampling algorithm
Can also there are some new patterns additionally changed, in collection process, RJMCMC algorithm while local feature in master pattern
The Markov Chain for maintaining the dimension of a stochastic variable variable, the stochastic variable being continuously generated in the Markov Chain, repeatedly
A fixed probability distribution p is gradually leveled off to during generation up to probability distribution p complete stability, next from the Markov
All stochastic variables sampled in chain are all satisfied the probability distribution of this fixation;
The RJMCMC algorithm is as follows:
Set probability density function:
Wherein, p indicates probability distribution, and Z is to make to be distributed normalized partition function, does not need to calculate in RJMCMC algorithm and match
It is sampled in the case where dividing function, F is energy function;
F=F (μ, μa,μb) (4)
The probability density function of relational graph vector μ indicates are as follows:
Wherein, β is the temperature coefficient by being manually set;
Sampling process specifically: by the relational graph vector μ of pattern aaWith the relational graph vector μ of pattern bbRJMCMC is inputted to calculate
Method, μa∈ μ, μb∈ μ, μaIt will be as markovian initializaing variable with e0It indicates, each variable that Markov Chain obtains
eiRepresent a relational graph vector μiRepresent a new pattern;
RJMCMC algorithm can randomly choose unrestrained shifting during each iteration obtains markovian new variables first
One of operation in operation or skip operation;If obtaining m-th of variable e in Markov ChainmThe m times iteration choosing
Unrestrained shifting operation is selected, then markovian new variables emIt will be by a upper variable em-1The relational graph vector μ's of middle representative
Any one μiIt is upper to add one from normal distributionThe offset of middle sampling obtains, the normal distributionIt will be by artificially specifying;
If obtaining i-th of variable x in Markov ChainiI-th iteration selected skip operation, then Markov Chain
New variables xiCandidate variables xi' will be by a upper variable xi-1In increase or reduce the dimension of vector at random and obtain,
Specifically, in variable xi-1The relational graph vector μ of representativei-1It is middle to be randomly choosed according to the corresponding relationship of the element calculated in step S2
The corresponding relationship of one group of element, and new element is added in the corresponding relationship to the pattern for lacking new element, the four of new element
A information generates at random, while generating new element and relational graph vector μi-1The relationship between element in corresponding pattern,
Or the element having more is subtracted, while eliminating the element and relational graph vector μ having morei-1Element relation in corresponding pattern,
Form new relational graph vector μi;
RJMCMC algorithm chooses whether to receive candidate variables x' according to rejection probability is received as followsiAs new variables xi:
WhereinExpression receives xi+1As the probability of new variables in Markov Chain, value [0,1], if this changes
Generation selection is unrestrained to move operation, then refuses acceptance probability selection formula (6), wherein p (μi+1) it is using relational graph vector μi+1It calculates
The probability density arrived;p(μi) it is using relational graph vector μiThe probability density being calculated;If current iteration selection jump behaviour
Make, then refuses acceptance probability selection formula (7), wherein q (μi|μi+1) it is from relational graph vector μi+1By the dimension for increasing and decreasing vector
Obtain relational graph vector μiProbability;q(μi+1|μi) it is from relational graph vector μiBy increase and decrease vector dimension obtain relational graph to
Measure μi+1Probability;Then RJMCMC algorithm one number t of stochastical sampling from a 0-1 distribution;If number t is less thanThen RJMCMC algorithm receives xi+1As new variables in Markov Chain;If number t is greater thanThen RJMCMC
Algorithm does not receive xi+1As new variables in Markov Chain, x is keptiAs new variables in Markov Chain;
After iteration after a period of time, markovian stochastic variable is calculated according to formula (5) in RJMCMC algorithm
To the number probability that occurs of value will be equal to this value for being calculated, these variables will be sampled as RJMCMC algorithm
As a result.
Further, following energy function is constructed between two patterns, is newly sampled for metrology step S3 new
The quality of pattern:
Wherein, μ is the relational graph vector of element in new pattern, μaAnd μbIt is first in given pattern a and pattern b respectively
The relational graph vector of element;α is the weight of first item in energy function;First item F in energy functionvalid(μ) is defined as:
Fvalid(μ)=μ-σ (μ) (9)
Wherein σ (μ) is the element in relational graph vector and vector set new composed by the actual value of its relationship
Wherein σi(μ) represents i-th in σ (μ) vector, μiI-th in μ vector is represented, if i-th in μ is figure
The information of an element in case, then i-th in σ (μ), which keeps i-th value in μ to remain unchanged, indicates the phase of the element in pattern
Same information;If i-th in μ be a relationship between element and element in pattern value, i-th in σ (μ) if is this
The actual value of relationship;It is then from node NiIt is got required for calculating in several information of representative element
Location information, directional information or angle information, thenIt indicates according to node NiInstitute's generation
The relationship type of table utilizes its relevant two input node Ni1And Ni2The corresponding element information for being included calculates the relationship
Actual value;If i-th in μ is a relation value between relationship and relationship in relational graph vector, i-th in σ (μ)
It is equally the actual value of the relationship, thenIt indicates according to node NiRepresentative relationship type utilizes
Its relevant two input node Ni1And Ni2The relation value for being included calculates the actual value of the relationship;
β is the weight of Section 2 in energy function, and some elements in new pattern that user's constrained sampling obtains are to maintain
The position of some elements in given pattern a and pattern b, size and Orientation, therefore Fconstrain(μ,μa,μb) be defined as measuring
The difference between the information of element and the information of the element in its specified pattern to suffer restraints in the relational graph vector of new pattern
Away from:
Fconstrain(μ,μa,μb)=μ { Eca}-μa{Ec'a}+μ{Ecb}-μb{E'cb} (11)
Wherein EcaIt indicates to suffer restraints in new pattern and needs the element set of the element information in holding pattern a, E'caTable
Show EcaIn element need the element set in the pattern a that keeps;EcbIt indicates to suffer restraints in new pattern and needs holding pattern b
In element information element set, E'cbIndicate EcbIn element need the element set in the pattern b that keeps;μ{EcaTable
Show EcaElement information of the element in the relational graph vector μ of new pattern in set;μa{E'caIndicate E'caElement in set
In the relational graph vector μ of pattern aaIn element information;μb{E'cbIndicate E'cbElement in set pattern b relational graph to
Measure μbIn element information;
γ is the weight of Section 3 in energy function, Fgroup(μ) is defined as measuring between symmetric relation group and its average value
Gap:
Fgroup(μ) is one as composed by the gap between symmetric relation groups all in relational graph vector μ and its average value
Dimensional vector, the element symmetry in pattern show as relation value in relational graph vector, and G indicates the relation value phase of symmetric relation group
Deng set, if the set for having H relation value equal in newly-generated pattern, μ { GhIndicate that the relationship in h-th of set is being closed
It is vector composed by the value in figure vector, H=1~h,Indicate the average value of all relation values in h-th of set.
δ is the weight of Section 4 in energy function, and a pattern conduct is randomly choosed from given pattern a and pattern b
New pattern instructs pattern, this instructs the relational graph vector of pattern to be expressed as τ, therefore Flocal(μ, τ) is defined as measuring new pattern
With instruct the gap between pattern:
Flocal(μ, τ)=μ-τ (13)
Finally, meet from the pattern that one best for energy equation: the actual value of the relationship in pattern is corresponding with pattern
Relation vector figure μ in relation value it is equal;Relation value in the corresponding relation vector figure μ of pattern and the symmetric relation where it
The average value of relation value is equal in group;The information of the element to suffer restraints in pattern is equal with the information for the element that needs are kept;
The corresponding relation vector figure μ of pattern is equal with directive relationship vectogram τ;
It is a kind of to utilize pattern generation browser interface method, comprising the following steps:
1) the high dimension vector expression characteristic for, obtaining the new pattern that sampling obtains, obtains high-dimensional vector space;It is described it is high-order to
Measure expression characteristic is indicated using the relational graph vector of the pattern;
2), using GPLVM (Gaussian Process Latent Variable Model, Gaussian process hidden variable mould
Type) high-dimensional vector space is compressed to two-dimensional manifold space;
3), by the point inverse mapping in two-dimensional manifold space into higher dimensional space, new pattern is obtained, and after carrying out to it
Reason;
4) pattern browser interface, is generated, the pattern browser interface is divided into two parts, the two-dimensional flow that a part is
Shape thermal map, user are slided on two-dimensional manifold thermal map by mouse, and another part will appear corresponding new pattern.
Further, step 2)
Compression is using Gaussian process latent variable model tool box, and the tool box is by high dimension vector expression characteristic dimensionality reduction at two
Dimensional vector, forms two-dimensional manifold space, and all bivectors determine the height of sampled result as corresponding coordinate, the coordinate
The two-dimensional manifold of dimensional vector space.Tool box can also calculate pattern and sampled result all in the high-dimensional vector space simultaneously
Covariance value, covariance value is bigger, and the color shown on two-dimensional manifold is redder;Covariance value is smaller, shows on two-dimensional manifold
Color is more blue, ultimately forms a two-dimensional manifold thermal map.
Further, step 3) is particular by Gaussian process latent variable model tool box by the point in two-dimensional manifold space
Coordinate inverse mapping to obtaining corresponding new relation figure vector.One completely new relational graph vector determines a completely new pattern.
Further, step 3) is post-processed by following pattern repair function, is existed for repairing in pattern
Symmetric relation:
Wherein μ*It is the pattern relationships figure vector for the minimum value for meeting formula (14), μ is the variable of repair function, μ0It is desirable
Pass through the relational graph vector for the pattern that this function is repaired.
The beneficial effects of the present invention are: corresponding to allocation plan the present invention is based on the element of genetic algorithm can effectively record
The specific physical effect such as division and merging caused by being deformed between these different topology structure patterns.It is adopted by RJMCMC
Sample algorithm can stablize acquisition and largely both be able to maintain in multiple source patterns between the source pattern of two different topology structures
Also there can be some new patterns additionally changed while local feature.GPLVM algorithm can learn locating for these sampled results
Higher dimensional space simultaneously expresses this higher dimensional space by the method for low dimensional manifold, can unify rapidly to browse one within this space
A little new patterns for having consecutive variations effect.
Detailed description of the invention
Fig. 1 is a kind of pattern generation of structurally variable and the method flow using pattern generation browser interface in embodiment
Figure;
Fig. 2 is the schematic diagram of the attribute of two levels of one pattern of the present embodiment;
Fig. 3 is the digraph of the relational graph vector of one pattern of the present embodiment;
Fig. 4 is the pattern a and pattern b for needing to carry out element pairing in the present embodiment;
Fig. 5 is the result figure that the present embodiment matches the element of two patterns of Fig. 4;
Fig. 6 is the new pattern that the present embodiment RJMCMC algorithm obtains;
Fig. 7 is another new pattern that the present embodiment RJMCMC algorithm obtains;
Fig. 8 is the generated browser interface schematic diagram of pattern of the present embodiment.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description.
A kind of method for generating pattern of structurally variable as shown in Figure 1, comprising the following steps:
S1, relationship graph model is constructed for the element in master pattern;
Master pattern described in step S1 is all made of element.It is special that there is the element of these patterns strong geometry to arrange
Sign.There is certain relationship between the element and element of pattern, there is also more advanced relationships, the i.e. pass of relationship between these relationships
System, relationship described in step S1 include the relationship type of element and the relationship type of relationship, and wherein the relationship type of element includes
Angular separation relationship, the dimension scale relationship of element and element of the distance between element and element relationship, element and element;It closes
The relationship type of system includes that the distance kept between element is equal difference or waits than some element and other yuan in relationship, pattern
The relationship etc. that angle between the direction of element is consistent.
Pattern in the present embodiment shares the relationship type of six kinds of elements and the relationship type of three kinds of relationships.
The relationship type of six kinds of elements is respectively as follows:
(1) representative pattern 1. in an element E1With representative pattern 2. in another element E2Center distance quilt
The Euclidean distance relationship being defined as between element.(2) an element E in representative pattern1With it is another in representative pattern
A element E2Direction between differential seat angle be defined as the direction difference relationship between element.The value range of relation value be [- π,
π], clockwise angle change is negative.(3) an element E in representative pattern1With another element E in representative pattern2
Size between difference be defined as the size difference relationship between element.(4) an element E in representative pattern1Scheme with representing
Another element E in case2Central point between line and X-axis between angle be defined as absolute angle difference relationship.It closes
The value range of set occurrence is [- π, π], and clockwise angle change is negative.(5) an element E in representative pattern1And representative
Another element E in pattern2Central point between line and element E1Direction between angle be defined as between element
Relative angular difference relationship.The value range of relation value is [- π, π], and clockwise angle change is negative.(6) representative pattern
In an element E1With another element E in representative pattern2Central point between line and element E1Direction between
The absolute value of angle is defined as the symmetry angle difference relationship between element.The value range of relation value be [0, π], clockwise and
Angle change counterclockwise is positive number.(5) between (6) the difference is that either with or without absolute value is sought.
The relationship type of three kinds of relationships is respectively as follows:
1. the difference R of two relationships1-R2It is defined as the relationship of relationship difference.2. the difference of two angular relationships is defined as
The relationship of angular relationship difference, the value range of relation value are [0, π].3. the quotient R of two relationships1/R2It is defined as the pass of relationship quotient
System.
There are two layer attributes for one pattern: first layer attribute is element, and second layer attribute is arrangement, with left frame in Fig. 2
For interior pattern, for the part in pattern dotted line frame there are two the attribute of level, first layer attribute is element, and second layer attribute is row
Column are indicated by the distance between element and element relationship here.
Step S1 uses relationship graph model to describe the relationship and these relationships between element and element in pattern
Between relationship.
Relationship graph model can be indicated by relational graph vector.Specifically set: μ is the relational graph of element in a pattern
Vector.Relational graph vector is an one-dimensional vector, is made of three parts, wherein first part is successively each member in pattern
The position (x, y) of element, towards θ and size s these three information, the second part includes the value of relationship between element and element, the
Three parts are then the values of the relationship of element relation.Relation value (relation value of inclusion relation value and relationship) is in relational graph vector
It is unified to be indicated with r:
μ=(x1,y1,θ1,s1,x2,y2,θ2,s2,...,r1,r2,r3,...)(1)
Relationship graph model passes through digraphIt is indicated, the node N=E ∪ R of digraph, wherein E is figure
The set of element in case, i-th of element representation are Ei=(xi,yi,θi,si), Ei∈ E, i indicate the quantity of element;R is in pattern
The set of relationship, R=RE∪RR, REIt is element relation set, RRIt is the set of relationship between relationship, each relationship has accordingly
Relation value, wherein the relation value of i-th of relationship is expressed as Ri=(ri), Ri∈R;Side the A={ (N of digraphi1,Ri),(Ni2,
Ri)}I=1...k+l, wherein k is the quantity of element relation in relationship graph model, and l is the quantity of the relationship of relationship.According to node Ni1With
Node Ni2It is and relationship RiRelevant two element E1、E2Or relationship R1、R2, wherein RiValue by node Ni1With node Ni2Packet
The value contained is calculated.
By node NiAll element informations or relation value of corresponding relationship vector are denoted as AndWherein Ej∈E。
Specifically, pattern shares the relationship type of 6 kinds of elements and the relationship type of 3 kinds of relationships, wherein 6 kinds of elements
Relationship type are as follows:
1) a, element E in representative patternaWith another element E in representative patternbThe distance at center be defined
Euclidean distance relationship between element, Ea∈ E, Eb∈ E, relation value RiValue are as follows:
Wherein, (xa,ya) it is element EaPosition, (xa,yb) it is element EbPosition;
2) a, element E in representative patternaWith another element E in representative patternbDirection between differential seat angle
The direction difference relationship being defined as between element, relation value RiValue range be [- π, π], clockwise angle change is negative
Number:
Wherein, θaAnd θbRespectively EaAnd EbDirection;
3) a, element E in representative patternaWith another element E in representative patternbSize between difference determined
Size difference relationship of the justice between element, relation value Ri:
Ri=sa-sb
Wherein, SaAnd SbRespectively EaAnd EbSize;
4) a, element E in representative patternaWith another element E in representative patternbCentral point between line
Angle between X-axis is defined as absolute angle difference relationship, relation value RiValue range be [- π, π], clockwise angle
Variation is negative:
5) a, element E in representative patternaWith another element E in representative patternbCentral point between line
With element EaDirection between angle be defined as the relative angular difference relationship between element, relation value RiValue range be
[- π, π], clockwise angle change are negative:
6) a, element E in representative patternaWith another element E in representative patternbCentral point between line
With element EaDirection between the absolute value of angle be defined as the symmetry angle difference relationship between element, relation value RiTake
Being worth range is [0, π], and angle change clockwise and anticlockwise is positive number:
The relationship type of 3 kinds of relationships is respectively as follows:
3-1), two be not angular relationship relationship EvAnd Ed, value RvAnd RdDifference be defined as the relationship of relationship difference,
Relation value RiValue:
Ri=Rv-Rd
3-2), two angular relationship EvAnd EdValue RvAnd RdDifference be defined as the relationship of angular relationship difference, relation value Ri
Value range be [0, π]
3-3), two relationship EvAnd EdValue RvAnd RdQuotient be defined as the relation value R of relationship quotienti:
Ri=Rv/Rd
Fig. 3 is the digraph of the relational graph vector of a pattern, and there are distance differences between element and element in pattern
The relationship i.e. relationship type of the first element;The difference relationship i.e. relationship type of the third relationship between distance.
S2, the matching that element is carried out between different topology structure pattern;
The discrete optimization model based on following energy equation is constructed, for looking for corresponding element between two patterns:
Wherein, m and n is the element number for carrying out the pattern a and pattern b of Match of elemental composition, X respectivelyijIt is for specifying one
Whether i-th of element in pattern be corresponding with j-th of element of another pattern, i ∈ m, j ∈ n;XijFor 1 indicate element i and
There are corresponding relationships by element j, are otherwise not present;VijWork as X for specifiedijWhen being 1, correspondence bring consumption;Element and element
Between corresponding consumption be defined as Euclidean distance between two elements;λ is weight, before balancing in discrete optimization model
OneWith latter
Latter for avoid two in a pattern symmetrical elements respectively correspond in other patterns two it is not right
Therefore the pattern of title establishes a list, each single item in list is that four element (p, q, g, h), wherein p and g indicate figure
Pth and q-th of element in case a are symmetric relations, and q and h indicate that q and h-th of element in pattern a are symmetric relations.If
The length of the list be K, then in latter k-th of list calculating formula SkIs defined as:
Wherein Xp,gIt is, X whether corresponding with q-th of element in pattern b for p-th of element in given pattern aq,h
Be it is whether corresponding with h-th of element of pattern b for q-th of element in given pattern a,It is xor operation;
The discrete optimization model needs to meet three following constraint conditions simultaneously,Indicate every in pattern a
One element at least will there are corresponding relationships with an element in pattern b;Then indicate each of pattern b
Element also at least will there are corresponding relationships with an element in pattern a;It then indicates to work as pattern a
In i-th element and pattern b in j-th of element generate corresponding relationship after, if i-th of element and pattern b in pattern a
In be not j-th of element other elements generate correspondence, then j-th of element in pattern b cannot again with no in pattern a
The other elements of i-th of element generate correspondence;If in j-th of the element and pattern a in same pattern b is not i-th yuan
The other elements of element generate correspondence, then i-th of element in pattern a cannot again with no in pattern b j-th of element other
Element generates correspondence.
Fig. 4 is two patterns for needing to carry out element pairing in the present embodiment, and Fig. 5 is by discrete optimization model to Fig. 4
The element of two patterns carry out element pairing as a result, each of pattern element all has label, the figure of left and right two in figure
There are corresponding relationships for the identical element of label in case, and according to the positional relationship of element in two patterns, discrete optimization model allows
A point in one pattern can correspond to multiple points on another pattern.In Fig. 4, wherein one on left side pattern inner ring
Point can correspond to two points on the pattern inner ring of the right.
S3, it is sampled between different topology structure pattern based on RJMCMC algorithm, obtains new pattern;
Following energy function is constructed between two patterns for measuring the quality for newly sampling obtained pattern:
F(μ,μa,μb)=| | α Fvalid(μ)||2+||βFconstrain(μ,μa,μb)||2+||γFgroup(μ)||2+||δ
Flocal(μ,τ)||2
s.t.μ{Rca}=μa{Rc′a}andμ{Rcb}=μb{Rc′b} (4)
Wherein, μ is the relational graph vector of element in new pattern, μaAnd μbIt is element in two given patterns respectively
Relational graph vector.α is the weight of first item in energy function;First item F in energy functionvalid(μ) is defined as:
Fvalid(μ)=μ-σ (μ) (5)
Wherein σ (μ) is the element and vector new composed by the actual value of its relationship in relational graph vector
Wherein σi(μ) represents i-th in σ (μ) vector, μiI-th in μ vector is represented, if i-th in μ is figure
The information of an element in case, then i-th in σ (μ), which keeps i-th value in μ to remain unchanged, indicates the phase of the element in pattern
Same information;If i-th in μ be a relationship between element and element in pattern value, i-th in σ (μ) if is this
The actual value of relationship;It is then from node NiIt is got required for calculating in several information of representative element
Location information, directional information or angle information, thenIt indicates according to node NiInstitute's generation
The relationship type of table utilizes its relevant two input node Ni1And Ni2The corresponding element information for being included calculates the relationship
Actual value;If i-th in μ is a relation value between relationship and relationship in relational graph vector, i-th in σ (μ)
It is equally the actual value of the relationship, thenIt indicates according to node NiRepresentative relationship type utilizes it
Relevant two input node Ni1And Ni2The relation value for being included calculates the actual value of the relationship;
β is the weight of Section 2 in energy function, and some elements in new pattern that user's constrained sampling obtains are to maintain
The position of some elements in given pattern a and pattern b, size and Orientation, therefore Fconstrain(μ,μa,μb) be defined as measuring
The difference between the information of element and the information of the element in its specified pattern to suffer restraints in the relational graph vector of new pattern
Away from:
Fconstrain(μ,μa,μb)=μ { Eca}-μa{E'ca}+μ{Ecb}-μb{E'cb} (11)
Wherein EcaIt indicates to suffer restraints in new pattern and needs the element set of the element information in holding pattern a, E'caTable
Show EcaIn element need the element set in the pattern a that keeps;EcbIt indicates to suffer restraints in new pattern and needs holding pattern b
In element information element set, E'cbIndicate EcbIn element need the element set in the pattern b that keeps;μ{EcaTable
Show EcaElement information of the element in the relational graph vector μ of new pattern in set;μa{E'caIndicate E'caElement in set
In the relational graph vector μ of pattern aaIn element information;μb{E'cbIndicate E'cbElement in set pattern b relational graph to
Measure μbIn element information;
γ is the weight of Section 3 in energy function, Fgroup(μ) is defined as measuring between symmetric relation group and its average value
Gap:
Fgroup(μ) is one as composed by the gap between symmetric relation groups all in relational graph vector μ and its average value
Dimensional vector, the element symmetry in pattern show as relation value in relational graph vector, and G indicates the relation value phase of symmetric relation group
Deng set, if the set for having H relation value equal in newly-generated pattern, μ { GhIndicate that the relationship in h-th of set is being closed
It is vector composed by the value in figure vector, H=1~h,Indicate the average value of all relation values in h-th of set.
δ is the weight of Section 4 in energy function, and a pattern conduct is randomly choosed from given pattern a and pattern b
New pattern instructs pattern, this instructs the relational graph vector of pattern to be expressed as τ, therefore Flocal(μ, τ) is defined as measuring new pattern
With instruct the gap between pattern:
Flocal(μ, τ)=μ-τ (13)
Finally, meet from the pattern that one best for energy equation: the actual value of the relationship in pattern is corresponding with pattern
Relation vector figure μ in relation value it is equal;Relation value in the corresponding relation vector figure μ of pattern and the symmetric relation where it
The average value of relation value is equal in group;The information of the element to suffer restraints in pattern is equal with the information for the element that needs are kept;
The corresponding relation vector figure μ of pattern is equal with directive relationship vectogram τ;
Further, it is obtained between the pattern of different topology structure using RJMCMC sampling algorithm and had largely both been able to maintain
Can also there are some new patterns additionally changed while local feature in multiple source patterns;RJMCMC algorithm maintenance one is random
The variable Markov Chain of the dimension of variable, the stochastic variable being continuously generated in the Markov Chain gradually become in the process of running
A fixed probability distribution p is bordering on until complete stability.Next sampled from the Markov Chain it is all with
Machine variable is all satisfied the probability distribution of this fixation.Therefore RJMCMC algorithm sets such a probability density function:
Wherein F is energy function:
F=F (μ, μa,μb) (11)
The probability density function of relational graph vector μ can be expressed as
Z is to make to be distributed normalized partition function, typically more complicated, but RJMCMC algorithm can match not needing to calculate
It is sampled in the case where dividing function.β is a temperature coefficient.According to RJMCMC algorithm calculate energy equation difference, β's
Value is from artificial setting.
By the relational graph vector μ of pattern aaWith the relational graph vector μ of pattern bbInput RJMCMC algorithm, μa∈ μ, μb∈ μ, μa
It will be as markovian initializaing variable with x0It indicates, each variable x that Markov Chain obtainsiRepresent a relational graph
Vector μi;RJMCMC algorithm can randomly choose unrestrained shifting during each iteration obtains markovian new variables first
One of operation in operation or skip operation.
Each variable e that Markov Chain obtainsiRepresent a relational graph vector μiAlso represent a new pattern;
RJMCMC algorithm can randomly choose unrestrained shifting during each iteration obtains markovian new variables first
One of operation in operation or skip operation;If obtaining m-th of variable e in Markov ChainmThe m times iteration choosing
Unrestrained shifting operation is selected, then markovian new variables emIt will be by a upper variable em-1The relational graph vector μ's of middle representative
Any one μiIt is upper to add one from normal distributionThe offset of middle sampling obtains.The normal distributionIt will be by artificially specifying.
If obtaining i-th of variable x in Markov ChainiI-th iteration selected skip operation, then Markov Chain
New variables xiCandidate variables xi' will be by a upper variable xi-1In increase or reduce the dimension of vector at random and obtain,
Specifically, in variable xi-1The relational graph vector μ of representativei-1It is middle to be randomly choosed according to the corresponding relationship of the element calculated in step S2
The corresponding relationship of one group of element, and new element is added in the corresponding relationship to the pattern for lacking new element, the four of new element
A information generates at random.New element and relational graph vector μ are generated simultaneouslyi-1The relationship between element in corresponding pattern,
Or the element having more is subtracted, while eliminating the element and relational graph vector μ having morei-1Element relation in corresponding pattern,
Form new relational graph vector μi;
RJMCMC algorithm chooses whether to receive candidate variables x according to rejection probability is received as followsi' become new variables xi:
WhereinExpression receives xi+1As the probability of new variables in Markov Chain, value [0,1], if this changes
Generation selection is unrestrained to move operation, then refuses acceptance probability selection formula (6), wherein p (μi+1) it is using relational graph vector μi+1It calculates
The probability density arrived;p(μi) it is using relational graph vector μiThe probability density being calculated;If current iteration selection jump behaviour
Make, then refuses acceptance probability selection formula (7), wherein q (μi|μi+1) it is from relational graph vector μi+1By the dimension for increasing and decreasing vector
Obtain relational graph vector μiProbability;q(μi+1|μi) it is from relational graph vector μiBy increase and decrease vector dimension obtain relational graph to
Measure μi+1Probability;Then RJMCMC algorithm one number t of stochastical sampling from a 0-1 distribution;If number t is less thanThen RJMCMC algorithm receives xi+1As new variables in Markov Chain;If number t is greater thanThen RJMCMC
Algorithm does not receive xi+1As new variables in Markov Chain, x is keptiAs new variables in Markov Chain.
After iteration after a period of time, markovian stochastic variable is calculated according to formula (5) in RJMCMC algorithm
To the number probability that occurs of value will be equal to this value for being calculated, these variables will be sampled as RJMCMC algorithm
As a result.
RJMCMC algorithm passes through skip operation during iterationObtain the X ' that variable dimension is n1, RJMCMC calculation
Method obtains X ' by unrestrained shifting operation in succession2、X′3、X′4, following skip operationThe X ' that variable dimension is m is obtained5.RJMCMC algorithm using unrestrained shifting operation in succession obtain variable X '6、X′7、X′8, next take skip operationIt obtains another
The variable of outer dimension.The result that these variables will be sampled as RJMCMC algorithm.Fig. 6 and Fig. 7 is two in Fig. 4 respectively
Pattern carries out two new patterns that RJMCMC algorithm obtains, respectively pattern X3 ' and pattern X7 '.
The element as shown, radius of circle represents the direction of the element in figure, in Fig. 6 and Fig. 7 on pattern inner ring
Quantity is different from the element on inner ring in two given patterns in Fig. 4.
A method of utilizing the pattern generation browser interface of the structurally variable, comprising the following steps:
1) the high dimension vector expression characteristic for the new pattern that sampling obtains, is obtained;
Since the new pattern sampled by RJMCMC algorithm all has respective relational graph vector, new pattern
The relational graph vector that the pattern can be directly used in high-order vector expression characteristic indicates.
2), using GPLVM (Gaussian Process Latent Variable Model Gaussian process hidden variable mould
Type) above-mentioned high-dimensional vector space is compressed to two-dimensional manifold space;
Gaussian process latent variable model tool box has been used to carry out the dimensionality reduction in space.The tool can be directly by new pattern
High dimension vector expression characteristic compression dimensionality reduction is indicated at a bivector.All bivectors have determined one as coordinate
The two-dimensional manifold of the high-dimensional vector space of a sampled result.Tool box can also calculate figure all in the high-dimensional vector space simultaneously
The covariance value of case and sampled result, covariance value is bigger, and the color shown on two-dimensional manifold is redder;Covariance value is smaller, and two
The color shown in dimension manifold is more blue, ultimately forms a two-dimensional manifold thermal map.
3), by the point inverse mapping in two-dimensional manifold space into higher dimensional space, new pattern is obtained, and after carrying out to it
Reason;
It constructs following pattern repair function to be post-processed, for repairing symmetric relation present in pattern
Wherein μ is the variable of repair function, μ0It is the relational graph vector for needing the pattern repaired by this function.
S7, generate pattern browser interface: the pattern browser interface is divided into two parts, the two dimension that a part is
Manifold thermal map, user are slided on two-dimensional manifold thermal map by mouse, and another part will appear corresponding new pattern.
As shown in figure 8, the left side shows the heat diagram of two-dimensional manifold, the right shows the new pattern of corresponding higher dimensional space;
The browsing method of user is as follows: user's sliding mouse in the two-dimensional manifold heat diagram on the interface left side, on the right of interface
There is corresponding pattern.As shown in figure 8, the left side shows two-dimensional manifold heat diagram in interface, when user arbitrarily drags mouse on the diagram
It marks, the right will will appear corresponding pattern in interface.
It is provided for the embodiments of the invention technical solution above to be described in detail, specific is applied in the present invention
Principle and implementation of the present invention are described for example, and it is of the invention that the above embodiments are only used to help understand
Method and its core concept;At the same time, for those skilled in the art, according to the thought of the present invention, in specific embodiment party
There will be changes in formula and application range, in conclusion the contents of this specification are not to be construed as limiting the invention.
Claims (10)
1. a kind of method for generating pattern of structurally variable, which comprises the following steps:
S1, relationship graph model is constructed for the element in pattern;
S2, the discrete optimization model using energy equation carry out the matching of element between different topology structure pattern;
S3, it is based on RJMCMC algorithm, is sampled between different topology structure pattern, obtains new pattern.
2. a kind of method for generating pattern of structurally variable according to claim 1, which is characterized in that pass described in step S1
System include element relationship type and relationship relationship type, wherein the relationship type of element include between element and element away from
The dimension scale relationship of angular separation relationship, element and element from relationship, element and element;The relationship type of relationship includes member
The distance kept between element is equal difference or waits than the angle between some element in relationship, pattern and the direction of other elements
Relationship.
3. a kind of method for generating pattern of structurally variable according to claim 1, it is characterised in that: the pass of step S1 building
It is graph model is for describing the relationship between the relationship and relationship between the element in pattern;
Relationship graph model includes relational graph vector, wherein a relational graph vector represents a pattern;The relational graph vector is
One one-dimensional vector is mainly made of three parts, wherein first part is successively the position (x of each element in patterni,yi
Y), towards θiWith size siThese three information, the second part include the value of relationship between element and element, and third part is
Relation value r, the relation value r include the relation value of element and element and the relation value of relationship, and relational graph vector μ is as follows:
μ=(x1,y1,θ1,s1,x2,y2,θ2,s2,...,xi,yi,θi,si,r1,r2,r3,...,ri) (1)
Relationship graph model passes through digraphIt is indicated, the node N=E ∪ R of digraph, wherein E is in pattern
The set of element, i-th of element representation are Ei=(xi,yi,θi,si), Ei∈ E, i indicate the quantity of element;R is relationship in pattern
Set, R=RE∪RR, REIt is element relation set, RRIt is the set of relationship between relationship, each relationship has corresponding relationship
Value, wherein the relation value of i-th of relationship is expressed as Ri=(ri), Ri∈R;Side the A={ (N of digraphi1,Ri),(Ni2,
Ri)}I=1...k+l, wherein k is the quantity of element relation in relationship graph model, and l is the relationship quantity of relationship;By node NiIt is corresponding to close
It is that all element informations or relation value of figure vector are denoted as AndWherein Ej∈E;
Specifically, the relationship type of element shares 6 kinds, and the relationship type of relationship has 3 kinds, wherein the relationship type of 6 kinds of elements
Are as follows:
1) a, element E in representative patternaWith another element E in representative patternbThe distance at center be defined as member
Euclidean distance relationship between element, wherein relation value RiValue are as follows:
Wherein, (xa,ya) it is element EaPosition, (xa,yb) it is element EbPosition;
2) a, element E in representative patternaWith another element E in representative patternbDirection between differential seat angle determined
Direction difference relationship of the justice between element, relation value RiValue range be [- π, π], clockwise angle change be negative:
Wherein, θaAnd θbRespectively EaAnd EbDirection;
3) a, element E in representative patternaWith another element E in representative patternbSize between difference be defined as
Size difference relationship between element, relation value Ri:
Ri=sa-sb
Wherein, SaAnd SbRespectively EaAnd EbSize;
4) a, element E in representative patternaWith another element E in representative patternbCentral point between line and X-axis
Between angle be defined as absolute angle difference relationship, relation value RiValue range be [- π, π], clockwise angle change
For negative:
5) a, element E in representative patternaWith another element E in representative patternbCentral point between line and member
Plain EaDirection between angle be defined as the relative angular difference relationship between element, relation value RiValue range be [- π,
π], clockwise angle change is negative:
6) a, element E in representative patternaWith another element E in representative patternbCentral point between line and member
Plain EaDirection between the absolute value of angle be defined as the symmetry angle difference relationship between element, relation value RiValue model
It encloses for [0, π], angle change clockwise and anticlockwise is positive number:
The relationship type of 3 kinds of relationships is respectively as follows:
3-1), two be not angular relationship relationship EvAnd Ed, value RvAnd RdDifference be defined as the relationship of relationship difference, relation value
RiValue:
Ri=Rv-Rd
3-2), two angular relationship EvAnd EdValue RvAnd RdDifference be defined as the relationship of angular relationship difference, relation value RiTake
Being worth range is [0, π]
3-3), two relationship EvAnd EdValue RvAnd RdQuotient be defined as the relation value R of relationship quotienti:
Ri=Rv/Rd
4. a kind of method for generating pattern of structurally variable according to claim 1, it is characterised in that: base described in step S2
It is as follows in the discrete optimization model of energy equation:
So that
Wherein, m and n is the element number for carrying out the pattern a and pattern b of Match of elemental composition, X respectivelyijIt is for specifying a pattern
In i-th of element it is whether corresponding with j-th of element of another pattern, i ∈ m, j ∈ n;XijElement i and element are indicated for 1
There are corresponding relationships by j, are otherwise not present;VijWork as X for specifiedijWhen being 1, correspondence bring consumption;Between element and element
Corresponding consumption is defined as the Euclidean distance between two elements;λ is weight, for balancing the previous item in discrete optimization modelWith latter
Latter for avoid two in a pattern symmetrical elements respectively correspond in other patterns two it is asymmetric
Therefore pattern establishes a list, each single item in list is that four element (p, q, g, h), wherein p and g indicate pattern a
In pth and q-th of element be symmetric relation, q and h indicate that q and h-th of element in pattern a are symmetric relations;If the column
The length of table be K, then in latter k-th of list calculating formula SkIs defined as:
Wherein Xp,gIt is, X whether corresponding with q-th of element in pattern b for p-th of element in given pattern aq,hIt is to be used for
Whether q-th of element in given pattern a be corresponding with h-th of element of pattern b,It is xor operation;
The discrete optimization model needs to meet three following constraint conditions simultaneously,Indicate each of pattern a
Element at least will there are corresponding relationships with an element in pattern b;Then indicate each of pattern b element
It also at least will there are corresponding relationships with an element in pattern a;It then indicates when in pattern a
After j-th of element in i-th element and pattern b generates corresponding relationship, if in i-th of element and pattern b in pattern a
That the other elements of j-th of element generate correspondence, then j-th of element in pattern b cannot again with no in pattern a i-th
The other elements of a element generate correspondence;If in j-th of the element and pattern a in same pattern b is not i-th of element
Other elements generate correspondence, then i-th of element in pattern a cannot again with the other elements of no in pattern b j-th of element
Generate correspondence.
5. a kind of method for generating pattern of structurally variable according to claim 1, which is characterized in that described in step S3
RJMCMC algorithm is as follows:
Set probability density function:
Wherein, p indicates probability distribution, and Z is to make to be distributed normalized partition function, does not need to calculate partition letter in RJMCMC algorithm
It is sampled in the case where number, F is energy function;
F=F (μ, μa,μb) (4)
The probability density function of relational graph vector μ indicates are as follows:
Wherein, β is the temperature coefficient by being manually set;
Sampling process specifically: by the relational graph vector μ of pattern aaWith the relational graph vector μ of pattern bbInput RJMCMC algorithm, μa
∈ μ, μb∈ μ, μaIt will be as markovian initializaing variable with e0It indicates, each variable e that Markov Chain obtainsiGeneration
One relational graph vector μ of tableiRepresent a new pattern;
RJMCMC algorithm can randomly choose unrestrained move first and operate during each iteration obtains markovian new variables
Or one of operation in skip operation;If obtaining m-th of variable e in Markov ChainmThe m times iteration select
Unrestrained move operates, then markovian new variables emIt will be by a upper variable em-1The relational graph vector μ's of middle representative is any
One μiIt is upper to add one from normal distributionThe offset of middle sampling obtains, the normal distributionIt will be by artificially specifying;
If obtaining i-th of variable x in Markov ChainiI-th iteration selected skip operation, then it is markovian new
Variable xiCandidate variables xi' will be by a upper variable xi-1In increase or reduce the dimension of vector at random and obtain, specifically
, in variable xi-1The relational graph vector μ of representativei-1It is middle to randomly choose one group according to the corresponding relationship of the element calculated in step S2
The corresponding relationship of element, and new element is added in the corresponding relationship to the pattern for lacking new element, four letters of new element
Breath is random to be generated, while generating new element and relational graph vector μi-1The relationship between element in corresponding pattern, or
The element having more is subtracted, while eliminating the element and relational graph vector μ having morei-1Element relation in corresponding pattern is formed
New relational graph vector μi;
RJMCMC algorithm chooses whether to receive candidate variables x according to rejection probability is received as followsi' become new variables xi:
WhereinExpression receives xi+1As the probability of new variables in Markov Chain, value [0,1], if current iteration selects
It selects unrestrained move to operate, then refuses acceptance probability selection formula (6), wherein p (μi+1) it is using relational graph vector μi+1It is calculated
Probability density;p(μi) it is using relational graph vector μiThe probability density being calculated;If current iteration selects skip operation,
Refuse acceptance probability selection formula (7), wherein q (μi|μi+1) it is from relational graph vector μi+1Dimension by increasing and decreasing vector obtains
Relational graph vector μiProbability;q(μi+1|μi) it is from relational graph vector μiDimension by increasing and decreasing vector obtains relational graph vector
μi+1Probability;Then RJMCMC algorithm one number t of stochastical sampling from a 0-1 distribution;If number t is less than
Then RJMCMC algorithm receives xi+1As new variables in Markov Chain;If number t is greater thanThen RJMCMC algorithm is not
Receive xi+1As new variables in Markov Chain, x is keptiAs new variables in Markov Chain;
After iteration after a period of time, markovian stochastic variable is calculated according to formula (5) in RJMCMC algorithm
The number probability that value occurs will be equal to this value being calculated, the knot that these variables will be sampled as RJMCMC algorithm
Fruit.
6. a kind of method for generating pattern of structurally variable according to claim 1, it is characterised in that: between two patterns
Following energy function is constructed, the quality for the metrology step S3 new pattern newly sampled:
Wherein, μ is the relational graph vector of element in new pattern, μaAnd μbIt is element in given pattern a and pattern b respectively
Relational graph vector;α is the weight of first item in energy function;First item F in energy functionvalid(μ) is defined as:
Fvalid(μ)=μ-σ (μ) (9)
Wherein σ (μ) is the element in relational graph vector and vector set new composed by the actual value of its relationship
Wherein σi(μ) represents i-th in σ (μ) vector, μiI-th in μ vector is represented, if i-th in μ is in pattern
The information of one element, then i-th in σ (μ), which keeps i-th value in μ to remain unchanged, indicates the identical letter of the element in pattern
Breath;If i-th in μ be a relationship between element and element in pattern value, i-th in σ (μ) if is the relationship
Actual value;It is then from node NiPosition required for calculating is got in several information of representative element
Confidence breath, directional information or angle information, thenIt indicates according to node NiRepresentative
Relationship type utilizes its relevant two input node Ni1And Ni2The corresponding element information for being included calculates the reality of the relationship
Actual value;If i-th in μ is a relation value between relationship and relationship in relational graph vector, i-th in σ (μ) is same
It is the actual value of the relationship, thenIt indicates according to node NiRepresentative relationship type utilizes its phase
The two input node N closedi1And Ni2The relation value for being included calculates the actual value of the relationship;
β is the weight of Section 2 in energy function, and some elements in new pattern that user's constrained sampling obtains are to maintain given
Pattern a and pattern b in the positions of some elements, size and Orientation, therefore Fconstrain(μ,μa,μb) be defined as measuring new figure
Gap between the information of the element to suffer restraints in the relational graph vector of case and the information of the element in its specified pattern:
Fconstrain(μ,μa,μb)=μ { Eca}-μa{E'ca}+μ{Ecb}-μb{E'cb} (11)
Wherein EcaIt indicates to suffer restraints in new pattern and needs the element set of the element information in holding pattern a, E'caIndicate Eca
In element need the element set in the pattern a that keeps;EcbIndicate the member needed in holding pattern b that suffers restraints in new pattern
The element set of prime information, E'cbIndicate EcbIn element need the element set in the pattern b that keeps;μ{EcaIndicate EcaCollection
Element information of the element in the relational graph vector μ of new pattern in conjunction;μa{E'caIndicate E'caElement in set is in pattern a
Relational graph vector μaIn element information;μb{E’cbIndicate E'cbThe relational graph vector μ of element in set in pattern bbIn
Element information;
γ is the weight of Section 3 in energy function, Fgroup(μ) is defined as measuring the difference between symmetric relation group and its average value
Away from:
Fgroup(μ) be composed by the gap in relational graph vector μ between all symmetric relation groups and its average value it is one-dimensional to
It measures, the element symmetry in pattern shows as relation value in relational graph vector, and G indicates that the relation value of symmetric relation group is equal
Set, if the set for having H relation value equal in newly-generated pattern, μ { GhIndicate h-th set in relationship in relational graph
Vector composed by value in vector, H=1~h,Indicate the average value of all relation values in h-th of set;
δ is the weight of Section 4 in energy function, and a pattern is randomly choosed from given pattern a and pattern b as new figure
Case instructs pattern, this instructs the relational graph vector of pattern to be expressed as τ, therefore Flocal(μ, τ) is defined as measuring new pattern and refer to
Lead the gap between pattern:
Flocal(μ, τ)=μ-τ (13).
7. a kind of method of the pattern generation browser interface using structurally variable described in claim 1, which is characterized in that including with
Lower step:
1) the high dimension vector expression characteristic for, obtaining the new pattern that sampling obtains, obtains high-dimensional vector space;The high position vector table
It is indicated up to feature using the relational graph vector of the pattern;
2) high-dimensional vector space, is compressed to by two-dimensional manifold space using Gaussian process latent variable model;
3), by the point inverse mapping in two-dimensional manifold space into higher dimensional space, new pattern is obtained, and post-process to it;
4) pattern browser interface, is generated, the pattern browser interface is divided into two parts, the two-dimensional manifold heat that a part is
Figure, user are slided on two-dimensional manifold thermal map by mouse, and another part will appear corresponding new pattern.
8. the method for pattern generation browser interface according to claim 7, which is characterized in that the compression of step 2) be using
High dimension vector expression characteristic dimensionality reduction at bivector, is formed two-dimensional flow by Gaussian process latent variable model tool box, the tool box
Shape space, all bivectors determine the two dimension of the high-dimensional vector space of sampled result as corresponding coordinate, the coordinate
Manifold.
9. the method for pattern generation browser interface according to claim 7, which is characterized in that step 3) is particular by height
This process latent variable model tool box by the coordinate inverse mapping of the point in two-dimensional manifold space to obtain corresponding new relation figure to
Amount, a completely new relational graph vector determine a completely new pattern.
10. the method for pattern generation browser interface according to claim 7, which is characterized in that step 3) is by as follows
Pattern repair function post-processed, for repairing symmetric relation present in pattern:
Wherein μ*It is the pattern relationships figure vector for the minimum value for meeting formula (14), μ is the variable of repair function, μ0It is to need to pass through
The relational graph vector for the pattern that this function is repaired.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910425318.7A CN110222239B (en) | 2019-05-21 | 2019-05-21 | Method for generating pattern with variable structure and generating browsing interface by using pattern |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910425318.7A CN110222239B (en) | 2019-05-21 | 2019-05-21 | Method for generating pattern with variable structure and generating browsing interface by using pattern |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110222239A true CN110222239A (en) | 2019-09-10 |
CN110222239B CN110222239B (en) | 2022-12-16 |
Family
ID=67821561
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910425318.7A Active CN110222239B (en) | 2019-05-21 | 2019-05-21 | Method for generating pattern with variable structure and generating browsing interface by using pattern |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110222239B (en) |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20030210210A1 (en) * | 2001-04-19 | 2003-11-13 | Tsuyoshi Ide | Discrete pattern, apparatus, method, and program storage device for generating and implementing the discrete pattern |
-
2019
- 2019-05-21 CN CN201910425318.7A patent/CN110222239B/en active Active
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20030210210A1 (en) * | 2001-04-19 | 2003-11-13 | Tsuyoshi Ide | Discrete pattern, apparatus, method, and program storage device for generating and implementing the discrete pattern |
Non-Patent Citations (1)
Title |
---|
GUERRERO P等: "PATEX:exploring pattern variations", 《ACM TRANSACTIONS ON GRAPHICS》 * |
Also Published As
Publication number | Publication date |
---|---|
CN110222239B (en) | 2022-12-16 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
McLoughlin et al. | Similarity measures for enhancing interactive streamline seeding | |
Hogräfer et al. | The state of the art in map‐like visualization | |
CN104346481B (en) | A kind of community detection method based on dynamic synchronization model | |
Golpîra | A novel Multiple Attribute Decision Making approach based on interval data using U2P-Miner algorithm | |
CN106156067B (en) | For creating the method and system of data model for relation data | |
Nayyeri et al. | 5* knowledge graph embeddings with projective transformations | |
CN110097609A (en) | A kind of fining embroidery texture moving method based on sample territory | |
CN107045519B (en) | A kind of visualization system of analysis crowd movement law | |
CN105354335B (en) | For the time-varying tree graph layout method of time-varying hierarchical data and application | |
Talagala et al. | FFORMPP: Feature-based forecast model performance prediction | |
Mandal et al. | Introduction of control points in splines for synthesis of optimized cam motion program | |
CN107146241B (en) | A kind of point cloud registration method based on differential evolution algorithm and TrimmedICP algorithm | |
CN108776835A (en) | A kind of deep neural network training method | |
CN100454341C (en) | Process type ground fast drawing method based on fractal hierarchical tree | |
CN103810388A (en) | Large-scale ontology mapping method based on partitioning technology oriented towards mapping | |
Desjardins et al. | Interactive visual clustering | |
Liu et al. | Illustration design model with clustering optimization genetic algorithm | |
CN112381148B (en) | Semi-supervised image classification method based on random regional interpolation | |
CN104537254A (en) | Fine drawing method based on social statistical data | |
CN105426601A (en) | Diversity design scheme reporting method and system based on BIM (Building Information Modeling) | |
Feng et al. | An optimised cellular automata model based on adaptive genetic algorithm for urban growth simulation | |
CN110222239A (en) | A kind of pattern generation of structurally variable and method using pattern generation browser interface | |
CN111353525A (en) | Modeling and missing value filling method for unbalanced incomplete data set | |
CN107066926A (en) | Positioned using the 3D objects of descriptor | |
CN109948055A (en) | A kind of recommended method based on group's preference |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |