CN106557581A - A kind of hypergraph division methods migrated based on multi-level framework and super side - Google Patents
A kind of hypergraph division methods migrated based on multi-level framework and super side Download PDFInfo
- Publication number
- CN106557581A CN106557581A CN201611072775.5A CN201611072775A CN106557581A CN 106557581 A CN106557581 A CN 106557581A CN 201611072775 A CN201611072775 A CN 201611072775A CN 106557581 A CN106557581 A CN 106557581A
- Authority
- CN
- China
- Prior art keywords
- hypergraph
- super
- super side
- node
- subset
- 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
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/51—Indexing; Data structures therefor; Storage structures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/56—Information retrieval; Database structures therefor; File system structures therefor of still image data having vectorial format
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Complex Calculations (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The present invention relates to a kind of hypergraph division methods migrated based on multi-level framework and super side, its step includes：Step 1, is generated the undirected hypergraph file of the power of tax, is preserved with predetermined file format；Step 2, reads the undirected hypergraph file of the power of tax, and the undirected hypergraph of power is assigned in the storage in internal memory；Step 3, is divided the undirected hypergraph of the power of tax, starts the multi-level hypergraph partiting step migrated based on super side, and the tax to being stored in internal memory is weighed undirected hypergraph and divided, and the division result for finally giving is stored in the undirected hypergraph of tax power and is divided in file.By the method for the present invention, solve the problems, such as to be currently based on that super subset uniformity coefficient is not good enough while the node quantity for dividing is higher and super in the hypergraph division methods for dividing, make the sub- hypergraph after division more uniform, improve the treatment effeciency based on hypergraph division methods class application.
Description
Technical field
The present invention relates to a kind of division methods of figure, more particularly to a kind of be based on multi-level framework and the migration of super side
Hypergraph division methods.
Background technology
Hypergraph is except the definition with general graph theory, moreover it is possible to solve the insurmountable problem of some general graph theorys.Reality should
With such as Workflow Management, task scheduling, data mining etc. is all more suitable for hypergraph being modeled.
Existing figure dividing system can be divided into the method divided based on node and the method divided based on side.Currently for
The achievement in research of the existing above two division methods of division of general figure.And for hypergraph, it is mainly based upon the hypergraph of node division
Division methods, it is rarely seen to have the hypergraph division methods divided based on super side.In existing hypergraph division methods, divided based on node
Hypergraph division methods only nodal set is evenly dividing, this is not meant to that sub- hypergraph is uniform, because each sub- hypergraph
Interior super side size and number may differ by very remote.When group hypergraph divides uneven, will reduce based on hypergraph division methods
Using treatment effeciency.It is based on the hypergraph division methods that super side divides, only a kind of distributed based on super edge flip at present
Method, the method lack global planning, and with the basic operation that is exchanged on super side, motility is not enough, causes the node split
Quantity is higher, super side subset uniformity coefficient is poor, divides quality defect.
The content of the invention
It is contemplated that at least solving one of technical problem present in prior art.
For this purpose, it is an object of the present invention to provide a kind of hypergraph division methods migrated based on multi-level framework and super side,
By designing a kind of hypergraph division methods migrated based on multi-level framework and super side, solve and be currently based on the super of super side division
Across the problem that the higher and super side subset uniformity coefficient of node quantity for dividing is not good enough in figure division methods, the sub- hypergraph after division is made
It is more uniform, improve the treatment effeciency based on hypergraph division methods class application.
For achieving the above object, the invention provides a kind of hypergraph division side migrated based on multi-level framework and super side
Method, including：
Step 1, is generated the undirected hypergraph file of the power of tax, is preserved with predetermined file format；
Step 2, reads the undirected hypergraph file of the power of tax, and the undirected hypergraph of power is assigned in the storage in internal memory；
Step 3, is divided the undirected hypergraph of the power of tax, starts the multi-level hypergraph partiting step migrated based on super side, to being stored in
Tax in internal memory is weighed undirected hypergraph and is divided, and the division result for finally giving is stored in the undirected hypergraph of tax power and divides file
In.
More specifically, the step 1 is specifically included：
Step 1.1, the 1st parameter of the 1st row of the file format of preservation represent the number m for assigning Quan Chao sides, the 2nd parameter
The number n for assigning power node is represent, the 3rd parameter represents hypergraph and assign power information type fmt, and the fmt=1 represents only super
While there is weights, fmt=10 represents that only node has weights, and fmt=11 represents super side and node has weights, and fmt is omitted and represented
Non- empowerment hypergraph；
Step 1.2, the 2nd row of file format start to the often row of m+1 rows the relevant information that represent one article of super side, the
1 numerical value is the value information on super side, the node numbering that its remainder values is connected by super side；
Step 1.3, the m+2 rows of file format start the often capable node that represent respective sequence to m+n+1 rows
Value information.
More specifically, the step 3 is specifically included：
Step 3.1, enters into the roughening stage of Hierarchy Method, and the roughening of present level layer is assigned power using side roughening algorithm
Some nodes of undirected hypergraph are combined together, and related super side is combined, and obtains the roughening empowerment hypergraph of each level course,
Repeat this process till the undirected hypergraph of roughening tax power is sufficiently small, that is, obtain a minimum and assign the undirected hypergraph of power；
Step 3.2, enters into the initial division stage of Hierarchy Method, runs the random super side with Constraints of Equilibrium and divides journey
Sequence, obtains the minimum initial division for assigning the undirected hypergraph of power；
Step 3.3, the hypergraph program initialization divided based on super side of Hierarchy Method optimizing phase initialize super side point
Cloth, spatial distribution of nodes, the migration income vector on super side, and super side migration Priority Queues；
Step 3.4, enters into the optimization elaboration phase of Hierarchy Method, is projected back in initial tax from minimum empowerment hypergraph and weighs nothing
To hypergraph, assign in the undirected hypergraph of power, using the hypergraph partition program migrated based on super side to refinement in the refinement of each level course
The projection division for assigning the undirected hypergraph of power is optimized, and the near-optimization for obtaining each level course refinement empowerment hypergraph divides collection, directly
To the undirected hypergraph of initial tax power is returned to；
The empowerment hypergraph division result for finally giving is stored in the undirected hypergraph of tax power and is divided in file by step 3.5.
More specifically, the step 3.2 is specifically included：
Step 3.2.1, reads Constraints of Equilibrium and requires；
Step 3.2.2, travels through per bar super side, is that super side generates a purpose subset numbering per bar with random process, judges
If the super side is put in the purpose subset, if Constraints of Equilibrium can be broken, it is then to generate new purpose with random process again
Subset is numbered, then is balanced the judgement of constraint, until finding the purpose subset for not breaking Constraints of Equilibrium；If all subsets are all
It is undesirable, then select to deviate the minimum subset of Constraints of Equilibrium to deposit；If the super side is put in the purpose subset, do not have
Break Constraints of Equilibrium, then super side is positioned over selected purpose subset, and execution step 3.2.3；
Step 3.2.3, travels through each node on the super side, if the node is present in other subsets, and shape before
State is non-cutting state, then the node is labeled as cutting state, and split node counter CUTV increases 1, the copy of the node
Enumerator REPLICA increases 1；If the node is present in other subsets, and state before is cutting state, then the node
Copy counter REPLICA increases 1.
More specifically, the step 3.3 is specifically included：
Step 3.3.1, it is super while concentrate take one it is super while, the super subset when being located super is recorded as subset P；
Step 3.3.2, selects a super side subset in addition to P, is designated as subset Q；
Step 3.3.3, travels through each node for connecting the super side, if number of copies of the node in P is more than 1 and in Q
In number of copies be equal to 0, then what migration income of the super side from P to Q will deduct the node cuts off cost value；If the node exists
Number of copies in P is equal to 1 and the number of copies in Q is more than 0, then migration income of the super side from P to Q will increase cutting for the node
Cut cost value；
Step 3.3.4, selects another subset in addition to P, repeat step 3.3.3, until traversal it is all of except P with
Outer subset；
Step 3.3.5, selects an other super side, repeat step 3.3.1,3.3.2,3.3.3,3.3.4, until having traveled through
It is super while concentrate it is all super while.
More specifically, the step 3.4 is specifically included：
Step 3.4.1, loop initialization, initialization iterative cycles enumerator COUNT are 0；
Step 3.4.2, iterative cycles enumerator COUNT increase 1, and all super sides that thaw；
Step 3.4.3, it is super while migrate Priority Queues in it is super while, select highest priority transportable super side, choosing
Select the super side subset of move target of not destruction of balance constraints；
Step 3.4.4, the super side is migrated to purpose subset, the change of value is cut, and freeze the super side after record migration；
Step 3.4.5, updates the super migration income when all not frozen neighbours are super, updates super side distribution, more
The copy distribution of the node of the new super side connection；
Step 3.4.6, from it is super while migrate Priority Queues in delete this it is super while；
Step 3.4.7, repeat step 3.4.3,3.4.4,3.4.5,3.4.6, until not having in super side migration Priority Queues
Transportable super side；
Step 3.4.8, is rolled back to the division distribution for occurring that minimum cuts off value；
Step 3.4.9, repeat step 3.4.1 to 3.4.8, until iterative cycles enumerator COUNT reaches the given upper limit.
More specifically, step 3.4.5 is specifically included：
Step 3.4.5.1, is designated as P the subset being located before the super side n* migrations of migration, and after migration, place subset is designated as Q；
Step 3.4.5.2, selects a node of connecting hyper-edge n*；
Step 3.4.5.3, the number of copies by the node in P reduce 1；
Step 3.4.5.4, if number of copies of the node in P is 0, travel through that the node connected all is not frozen
This super side place subset is designated as S, if S is not equal to P, this super side is subtracted from the migration income that S moves to P by the super side of knot
Remove node；Otherwise, execution step 3.4.5.5；
Step 3.4.5.5, if number of copies of the node in P is equal to 1, travel through that the node connected it is all not by
This super side place subset is designated as S by the super side freezed, if S is equal to P, this super side is moved to non-S subsets from subset P
Migration income cuts off cost value plus node；
Step 3.4.5.6, the number of copies by the node in Q increase by 1；
Step 3.4.5.7, if number of copies of the node in Q is equal to 1, travel through that the node connected it is all not by
This super side place subset is designated as S by the super side freezed, if S is not equal to Q, this super side is moved to the migration income of Q from S
Cost value is cut off plus node；Otherwise, execution step 3.4.5.8；
Step 3.4.5.8, if number of copies of the node in Q is equal to 2, travel through that the node connected it is all not by
This super side place subset is designated as S by the super side freezed, if S is equal to Q, this super side is moved to non-S subsets from subset Q
What migration income deducted node cuts off cost value；
Step 3.4.5.9, another node of selection connecting hyper-edge n*, repeat step 3.4.5.3～3.4.5.8, until
All nodes of connecting hyper-edge n* are traveled through.
By designing a kind of hypergraph division methods migrated based on multi-level framework and super side, solve at present across division
The problem that the higher and super side subset uniformity coefficient of node quantity is not good enough, makes the sub- hypergraph after division more uniform, improves based on super
The treatment effeciency of figure division methods class application.
The method of the present invention is quickly surpassed side migration income meter by node is key under given Constraints of Equilibrium
Calculate, while being split to node and being replicated, effectively reduce the super side contact across dividing subset.The method efficiently finds ratio
The more excellent empowerment hypergraph division result of prior art, reduces space complexity and time complexity, finally considerably improves
Assign the performance that the undirected hypergraph of power is divided.
The additional aspect and advantage of the present invention will be given in following description section, partly will be become from the following description
Obtain substantially, or recognized by the practice of the present invention.
Description of the drawings
The above-mentioned and/or additional aspect and advantage of the present invention will become from the description with reference to accompanying drawings below to embodiment
It is substantially and easy to understand, wherein：
Fig. 1 is shown according to a kind of flow process of the hypergraph division methods migrated based on multi-level framework and super side of the present invention
Figure；
Fig. 2 shows the hypergraph division methods flow process migrated based on multi-level framework and super side of one embodiment of the invention
Figure；
Fig. 3 shows the division methods flow chart migrated based on super side of optimizing phase of the invention.
Specific embodiment
It is in order to be more clearly understood that the above objects, features and advantages of the present invention, below in conjunction with the accompanying drawings and concrete real
Apply mode to be further described in detail the present invention.It should be noted that in the case where not conflicting, the enforcement of the application
Feature in example and embodiment can be mutually combined.
Many details are elaborated in the following description in order to fully understand the present invention, but, the present invention may be used also
Implemented with being different from mode described here using other, therefore, protection scope of the present invention does not receive following public tool
The restriction of body embodiment.
To realize above-mentioned goal of the invention, the invention of the present invention is mainly by following some realization：
First, based on the multi-level framework for dividing, in the multi-level roughening stage for dividing, the algorithm being roughened using side by some
Node is combined together, and obtains the roughening picture of next level course；In the initial division stage, the random super side with Constraints of Equilibrium is adopted
Division methods carry out direct K point to minimal graph；It is in the multi-level optimization for dividing and elaboration phase, super using what is migrated based on super side
Figure division methods, the division to the projection of each level course are optimized and refine, till original hypergraph is returned to.
2nd, the above-mentioned hypergraph division methods migrated based on super side, realize that by way of iteration direct K point of hypergraph is drawn
Point.All thaw when iteration starts every time all super sides, and travels through super side by the sequence of priority migration queue, allows super side to select migration
The purpose subset of Income Maximum.Once super side is migrated, then frozen, no longer selected, until all possible super side
After all passing through migration, roll back to hypergraph and cut off the maximum state of value, i.e., it is overall to cut off the minimum division shape of node copy amount
State, then state is divided with this carry out next round iteration as original state.
3rd, above-mentioned super side moving method, is to migrate benefit and Constraints of Equilibrium as standard to instruct migration with super side, and
Migration income is that the migration based on node is key quickly calculating, and per bar, super side has in each migratory direction accordingly
Migration income.Migration income is associated with value is cut off, that is, after migrating, divided node copy amount is fewer, and migration income is got over
Greatly.In order to better illustrate the solution of the present invention, illustrate below in conjunction with Figure of description.
Fig. 1 is shown according to a kind of flow process of the hypergraph division methods migrated based on multi-level framework and super side of the present invention
Figure.
As shown in figure 1, a kind of hypergraph division methods migrated based on multi-level framework and super side of the invention, bag
Include：
Step 1, is generated the undirected hypergraph file of the power of tax, is preserved with predetermined file format；
Step 2, reads the undirected hypergraph file of the power of tax, and the undirected hypergraph of power is assigned in the storage in internal memory；
Step 3, is divided the undirected hypergraph of the power of tax, starts the multi-level hypergraph partiting step migrated based on super side, to being stored in
Tax in internal memory is weighed undirected hypergraph and is divided, and the division result for finally giving is stored in the undirected hypergraph of tax power and divides file
In.
More specifically, the step 1 is specifically included：
Step 1.1, the 1st parameter of the 1st row of the file format of preservation represent the number m for assigning Quan Chao sides, the 2nd parameter
The number n for assigning power node is represent, the 3rd parameter represents hypergraph and assign power information type fmt, and the fmt=1 represents only super
While there is weights, fmt=10 represents that only node has weights, and fmt=11 represents super side and node has weights, and fmt is omitted and represented
Non- empowerment hypergraph；
Step 1.2, the 2nd row of file format start to the often row of m+1 rows the relevant information that represent one article of super side, the
1 numerical value is the value information on super side, the node numbering that its remainder values is connected by super side；
Step 1.3, the m+2 rows of file format start the often capable node that represent respective sequence to m+n+1 rows
Value information.
More specifically, the step 3 is specifically included：
Step 3.1, enters into the roughening stage of Hierarchy Method, and the roughening of present level layer is assigned power using side roughening algorithm
Some nodes of undirected hypergraph are combined together, and related super side is combined, and obtains the roughening empowerment hypergraph of each level course,
Repeat this process till the undirected hypergraph of roughening tax power is sufficiently small, that is, obtain a minimum and assign the undirected hypergraph of power；
Step 3.2, enters into the initial division stage of Hierarchy Method, runs the random super side with Constraints of Equilibrium and divides journey
Sequence, obtains the minimum initial division for assigning the undirected hypergraph of power；
Step 3.3, the hypergraph program initialization divided based on super side of Hierarchy Method optimizing phase initialize super side point
Cloth, spatial distribution of nodes, the migration income vector on super side, and super side migration Priority Queues；
Step 3.4, enters into the optimization elaboration phase of Hierarchy Method, is projected back in initial tax from minimum empowerment hypergraph and weighs nothing
To hypergraph, assign in the undirected hypergraph of power, using the hypergraph partition program migrated based on super side to refinement in the refinement of each level course
The projection division for assigning the undirected hypergraph of power is optimized, and the near-optimization for obtaining each level course refinement empowerment hypergraph divides collection, directly
To the undirected hypergraph of initial tax power is returned to；
The empowerment hypergraph division result for finally giving is stored in the undirected hypergraph of tax power and is divided in file by step 3.5.
More specifically, the step 3.2 is specifically included：
Step 3.2.1, reads Constraints of Equilibrium and requires；
Step 3.2.2, travels through per bar super side, is that super side generates a purpose subset numbering per bar with random process, judges
If the super side is put in the purpose subset, if Constraints of Equilibrium can be broken, it is then to generate new purpose with random process again
Subset is numbered, then is balanced the judgement of constraint, until finding the purpose subset for not breaking Constraints of Equilibrium；If all subsets are all
It is undesirable, then select to deviate the minimum subset of Constraints of Equilibrium to deposit；If the super side is put in the purpose subset, do not have
Break Constraints of Equilibrium, then super side is positioned over selected purpose subset, and execution step 3.2.3；
Step 3.2.3, travels through each node on the super side, if the node is present in other subsets, and shape before
State is non-cutting state, then the node is labeled as cutting state, and split node counter CUTV increases 1, the copy of the node
Enumerator REPLICA increases 1；If the node is present in other subsets, and state before is cutting state, then the node
Copy counter REPLICA increases 1.
More specifically, the step 3.3 is specifically included：
Step 3.3.1, it is super while concentrate take one it is super while, the super subset when being located super is recorded as subset P；
Step 3.3.2, selects a super side subset in addition to P, is designated as subset Q；
Step 3.3.3, travels through each node for connecting the super side, if number of copies of the node in P is more than 1 and in Q
In number of copies be equal to 0, then what migration income of the super side from P to Q will deduct the node cuts off cost value；If the node exists
Number of copies in P is equal to 1 and the number of copies in Q is more than 0, then migration income of the super side from P to Q will increase cutting for the node
Cut cost value；
Step 3.3.4, selects another subset in addition to P, repeat step 3.3.3, until traversal it is all of except P with
Outer subset；
Step 3.3.5, selects an other super side, repeat step 3.3.1,3.3.2,3.3.3,3.3.4, until having traveled through
It is super while concentrate it is all super while.
More specifically, the step 3.4 is specifically included：
Step 3.4.1, loop initialization, initialization iterative cycles enumerator COUNT are 0；
Step 3.4.2, iterative cycles enumerator COUNT increase 1, and all super sides that thaw；
Step 3.4.3, it is super while migrate Priority Queues in it is super while, select highest priority transportable super side, choosing
Select the super side subset of move target of not destruction of balance constraints；
Step 3.4.4, the super side is migrated to purpose subset, the change of value is cut, and freeze the super side after record migration；
Step 3.4.5, updates the super migration income when all not frozen neighbours are super, updates super side distribution, more
The copy distribution of the node of the new super side connection；
Step 3.4.6, from it is super while migrate Priority Queues in delete this it is super while；
Step 3.4.7, repeat step 3.4.3,3.4.4,3.4.5,3.4.6, until not having in super side migration Priority Queues
Transportable super side；
Step 3.4.8, is rolled back to the division distribution for occurring that minimum cuts off value；
Step 3.4.9, repeat step 3.4.1 to 3.4.8, until iterative cycles enumerator COUNT reaches the given upper limit.
More specifically, step 3.4.5 is specifically included：
Step 3.4.5.1, is designated as P the subset being located before the super side n* migrations of migration, and after migration, place subset is designated as Q；
Step 3.4.5.2, selects a node of connecting hyper-edge n*；
Step 3.4.5.3, the number of copies by the node in P reduce 1；
Step 3.4.5.4, if number of copies of the node in P is 0, travel through that the node connected all is not frozen
This super side place subset is designated as S, if S is not equal to P, this super side is subtracted from the migration income that S moves to P by the super side of knot
Remove node；Otherwise, execution step 3.4.5.5；
Step 3.4.5.5, if number of copies of the node in P is equal to 1, travel through that the node connected it is all not by
This super side place subset is designated as S by the super side freezed, if S is equal to P, this super side is moved to non-S subsets from subset P
Migration income cuts off cost value plus node；
Step 3.4.5.6, the number of copies by the node in Q increase by 1；
Step 3.4.5.7, if number of copies of the node in Q is equal to 1, travel through that the node connected it is all not by
This super side place subset is designated as S by the super side freezed, if S is not equal to Q, this super side is moved to the migration income of Q from S
Cost value is cut off plus node；Otherwise, execution step 3.4.5.8；
Step 3.4.5.8, if number of copies of the node in Q is equal to 2, travel through that the node connected it is all not by
This super side place subset is designated as S by the super side freezed, if S is equal to Q, this super side is moved to non-S subsets from subset Q
What migration income deducted node cuts off cost value；
Step 3.4.5.9, another node of selection connecting hyper-edge n*, repeat step 3.4.5.3～3.4.5.8, until
All nodes of connecting hyper-edge n* are traveled through.
Fig. 2 shows the hypergraph division methods flow process migrated based on multi-level framework and super side of one embodiment of the invention
Figure.
As shown in Fig. 2 the undirected hypergraph file of power is assigned in input；Read the hypergraph file and storage storage tax power is super in internal memory
Figure；Start the tax migrated based on multi-level framework and super side and weigh undirected hypergraph optimization partition program, enter into Hierarchy Method
In the roughening stage, some nodes that present level layer is roughened the undirected hypergraph of the power of tax are combined together using side method of roughening, are obtained
The roughening empowerment hypergraph of each level course；Repeat this process until be roughened empowerment hypergraph it is sufficiently small till, that is, obtain one it is minimum
The undirected hypergraph of power is assigned, then into the initial division stage of Hierarchy Method, the random division program with Constraints of Equilibrium is run,
Obtain the initial division of minimum empowerment hypergraph；The optimizing phase of Hierarchy Method is entered into, is projected back in just from minimum empowerment hypergraph
Beginning empowerment hypergraph, in the refinement empowerment hypergraph of each level course, is weighed undirected hypergraph optimization using the tax migrated based on super side and is drawn
Branch is divided to the projection for refining empowerment hypergraph and is optimized, and obtains the division that the undirected hypergraph of power is assigned in each level course refinement
Collection, until returning to original tax weighs undirected hypergraph；Obtain assigning the division result of the undirected hypergraph partition problem of power, and store it in tax
Weigh the division file of undirected hypergraph.
Fig. 3 shows the division methods flow chart migrated based on super side of optimizing phase of the invention.
As shown in figure 3, step is as follows：
1：Undirected hypergraph is weighed in the tax of roughening and projects to a upper level course.
2：The super side distribution of initialization and spatial distribution of nodes, count divided node number and copy amount.
3：The migration income on the super side of initialization, it is key according to the migration of node, each super side is rapidly calculated to each mesh
The migration of super side subset income.
4：The super side of initialization migrates Priority Queues, by the order that the migration income on super side is descending, to super side and migration
Rank in direction.
5：Loop initialization, is set to 0 iteration round enumerator COUNT.
6：Judge whether iteration round enumerator reaches the upper limit that will terminate circulating, if it is, jumping to the 15th step；Otherwise
The 7th step is continued executing with, enters excess of imports side transition process.
7：Thaw all super sides.
8：Judge whether Priority Queues also has transportable super side, if there is transportable super side in queue, perform
9th step, otherwise performs the 13rd step.
9：Select super side and its migratory direction.Select the super side and its migration side of highest priority in migration Priority Queues
To, judge super side after migration whether can destruction of balance constraint, select if destruction of balance is understood lesser priority super side or/
And migratory direction, super side and the migratory direction until finding not destruction of balance constraint.
10：Perform it is super while migration and when freezing super.The node connected by super side is copied in purpose division, in original
Node deletion without other super side connections in division.
11：Update the migration income on the super side do not freezed and cut off value.Key according to the migration of node, quick renewal is moved
Move it is super while neighbours it is super while in not frozen super side migration income, further according to the node segmentation situation calculating hypergraph of hypergraph
Cut off value.
12：The super side for just having completed migration operation is deleted in Priority Queues.Repeat step 8-12, until Priority Queues does not have
There is transportable super side, then perform the 13rd step.
13：It is rolled back to the division state for occurring that hypergraph minimum cuts off value.Minimum is cut off the super side migration after value occurs
Do reverse migration operation.
14：Iteration round enumerator COUNT values Jia 1.Repeat step 6-14, until meet iterative cycles end condition, then
Perform the 15th step.
15：Judge that current tax weighs whether undirected hypergraph returns to the initial level for assigning the undirected hypergraph of power, if it is not, then repeating
Step 1-15, until completing the initial Partitioning optimization for assigning the undirected hypergraph of power；If having returned to the initial level for assigning the undirected hypergraph of power,
Then obtain assigning the division result of the undirected hypergraph partition problem of power.
The present invention weighs undirected hypergraph based on the tax that multi-level framework and super side migrate and optimizes division methods, in given balance
Under constraint, quickly surpassed side migration income calculation by node is key, while node is split and is replicated, effectively
Reduce the super side contact across dividing subset.The method efficiently finds the empowerment hypergraph division result more excellent than prior art,
Space complexity and time complexity are reduced, the performance that the undirected hypergraph of the power of tax is divided finally is considerably improved.
This patent can apply the data processing field of various complex networks, need using hypergraph to be modeled simultaneously any
Carry out to use under the application scenarios that data division is processed.
In the description of this specification, the term such as term " installation ", " being connected ", " connection " all should be interpreted broadly, for example,
" connection " can be fixedly connected, or be detachably connected, or be integrally connected；" being connected " can be joined directly together,
Can be indirectly connected to by intermediary.For the ordinary skill in the art, on can understanding as the case may be
State term concrete meaning in the present invention.
In the description of this specification, the description of term " one embodiment ", " some embodiments ", " specific embodiment " etc.
Mean that the specific features, structure, material or the feature that describe with reference to the embodiment or example are contained in of the invention at least one real
Apply in example or example.In this manual, identical embodiment or reality are not necessarily referring to the schematic representation of above-mentioned term
Example.And, the specific features of description, structure, material or feature can in one or more any embodiment or example with
Suitable mode is combined.
The preferred embodiments of the present invention are the foregoing is only, the present invention is not limited to, for the skill of this area
For art personnel, the present invention can have various modifications and variations.It is all within the spirit and principles in the present invention, made any repair
Change, equivalent, improvement etc., should be included within the scope of the present invention.
Claims (7)
1. a kind of hypergraph division methods migrated based on multi-level framework and super side, it is characterised in that comprise the steps：
Step 1, is generated the undirected hypergraph file of the power of tax, is preserved with predetermined file format；
Step 2, reads the undirected hypergraph file of the power of tax, and the undirected hypergraph of power is assigned in the storage in internal memory；
Step 3, is divided the undirected hypergraph of the power of tax, starts the multi-level hypergraph partiting step migrated based on super side, to being stored in internal memory
In tax weigh undirected hypergraph and divided, the division result for finally giving is stored in into the undirected hypergraph of tax power and is divided in file.
2. a kind of hypergraph division methods migrated based on multi-level framework and super side according to claim 1, its feature are existed
In the step 1 is specifically included：
Step 1.1, the 1st parameter of the 1st row of the file format of preservation represent the number m for assigning Quan Chao sides, and the 2nd parameter is represented
The number n for assigning power node, the 3rd parameter represents hypergraph and assign power information type fmt, and the fmt=1 represents that only super side has
Weights, fmt=10 represent that only node has weights, and fmt=11 represents super side and node has weights, and fmt is omitted and represented non-tax
Power hypergraph；
Step 1.2, the 2nd row of file format start to the often row of m+1 rows the relevant information that represent one article of super side, the 1st
Numerical value is the value information on super side, the node numbering that its remainder values is connected by super side；
Step 1.3, the m+2 rows of file format start to the often row of m+n+1 rows the weights of the node that represent respective sequence
Information.
3. a kind of hypergraph division methods migrated based on multi-level framework and super side according to claim 1, its feature are existed
In the step 3 is specifically included：
Step 3.1, enters into the roughening stage of Hierarchy Method, the roughening of present level layer is assigned power using side roughening algorithm undirected
Some nodes of hypergraph are combined together, and related super side is combined, and obtains the roughening empowerment hypergraph of each level course, repeats
This process till the undirected hypergraph of roughening tax power is sufficiently small obtains a minimum and assigns the undirected hypergraph of power；
Step 3.2, enters into the initial division stage of Hierarchy Method, runs the random super side partition program with Constraints of Equilibrium,
Obtain the minimum initial division for assigning the undirected hypergraph of power；
Step 3.3, the hypergraph program initialization divided based on super side of Hierarchy Method optimizing phase, the super side distribution of initialization,
Spatial distribution of nodes, the migration income vector on super side, and super side migration Priority Queues；
Step 3.4, enters into the optimization elaboration phase of Hierarchy Method, is projected back in initial tax from minimum empowerment hypergraph and weighs undirected surpassing
Figure, is assigned in the undirected hypergraph of power in the refinement of each level course, assigns power to refinement using the hypergraph partition program migrated based on super side
The projection of undirected hypergraph is divided and is optimized, and the near-optimization for obtaining each level course refinement empowerment hypergraph divides collection, until returning
Assign weighing undirected hypergraph to initial；
The empowerment hypergraph division result for finally giving is stored in the undirected hypergraph of tax power and is divided in file by step 3.5.
4. a kind of hypergraph division methods migrated based on multi-level framework and super side according to claim 3, its feature are existed
In the step 3.2 is specifically included：
Step 3.2.1, reads Constraints of Equilibrium and requires；
Step 3.2.2, travels through per bar super side, is that super side generates a purpose subset numbering per bar with random process, if judged
The super side is put in the purpose subset, if Constraints of Equilibrium can be broken, be then new purpose subset to be generated with random process again
Numbering, then the judgement of constraint is balanced, until finding the purpose subset for not breaking Constraints of Equilibrium；If all subsets are not all inconsistent
Close and require, then select to deviate the minimum subset of Constraints of Equilibrium to deposit；If the super side is put in the purpose subset, do not break
Constraints of Equilibrium, then be positioned over selected purpose subset, and execution step 3.2.3 super side；
Step 3.2.3, travels through each node on the super side, if the node is present in other subsets, and state before is
Non- cutting state, then be labeled as cutting state the node, and split node counter CUTV increases 1, and the copy of the node is counted
Device REPLICA increases 1；If the node is present in other subsets, and state before is cutting state, then the copy of the node
Enumerator REPLICA increases 1.
5. a kind of hypergraph division methods migrated based on multi-level framework and super side according to claim 3, its feature are existed
In the step 3.3 is specifically included：
Step 3.3.1, it is super while concentrate take one it is super while, the super subset when being located super is recorded as subset P；
Step 3.3.2, selects a super side subset in addition to P, is designated as subset Q；
Step 3.3.3, travels through each node for connecting the super side, if number of copies of the node in P is more than 1 and in Q
Number of copies is equal to 0, then what migration income of the super side from P to Q will deduct the node cuts off cost value；If the node is in P
Number of copies be equal to 1 and number of copies in Q is more than 0, then migration income of the super side from P to Q will increase cuing off for the node
Cost value；
Step 3.3.4, selects another subset in addition to P, repeat step 3.3.3, until traversal is all of in addition to P
Subset；
Step 3.3.5, selects an other super side, repeat step 3.3.1,3.3.2,3.3.3,3.3.4, until having traveled through super side
The all super side concentrated.
6. a kind of hypergraph division methods migrated based on multi-level framework and super side according to claim 3, its feature are existed
In the step 3.4 is specifically included：
Step 3.4.1, loop initialization, initialization iterative cycles enumerator COUNT are 0；
Step 3.4.2, iterative cycles enumerator COUNT increase 1, and all super sides that thaw；
Step 3.4.3, it is super while migrate Priority Queues in it is super while, select highest priority transportable super side, select not
The super side subset of move target of destruction of balance constraints；
Step 3.4.4, the super side is migrated to purpose subset, the change of value is cut, and freeze the super side after record migration；
Step 3.4.5, updates the super migration income when all not frozen neighbours are super, updates super side distribution, and updating should
The copy distribution of the node of super side connection；
Step 3.4.6, from it is super while migrate Priority Queues in delete this it is super while；
Step 3.4.7, repeat step 3.4.3,3.4.4,3.4.5,3.4.6, until not having to move in super side migration Priority Queues
The super side moved；
Step 3.4.8, is rolled back to the division distribution for occurring that minimum cuts off value；
Step 3.4.9, repeat step 3.4.1 to 3.4.8, until iterative cycles enumerator COUNT reaches the given upper limit.
7. a kind of hypergraph division methods migrated based on multi-level framework and super side according to claim 6, its feature are existed
In step 3.4.5 is specifically included：
Step 3.4.5.1, is designated as P the subset being located before the super side n* migrations of migration, and after migration, place subset is designated as Q；
Step 3.4.5.2, selects a node of connecting hyper-edge n*；
Step 3.4.5.3, the number of copies by the node in P reduce 1；
Step 3.4.5.4, if number of copies of the node in P is 0, travel through that the node connected is all not frozen
This super side place subset is designated as S by super side, if S is not equal to P, this super side is deducted knot from the migration income that S moves to P
Point；Otherwise, execution step 3.4.5.5；
Step 3.4.5.5, if number of copies of the node in P is equal to 1, travel through that the node connected is all not frozen
Super side, this super side place subset is designated as into S, if S be equal to P, this super side is moved to the migration of non-S subsets from subset P
Income cuts off cost value plus node；
Step 3.4.5.6, the number of copies by the node in Q increase by 1；
Step 3.4.5.7, if number of copies of the node in Q is equal to 1, travel through that the node connected is all not frozen
Super side, this super side place subset is designated as into S, if S is not equal to Q, this super side is added from the migration income that S moves to Q
Node cuts off cost value；Otherwise, execution step 3.4.5.8；
Step 3.4.5.8, if number of copies of the node in Q is equal to 2, travel through that the node connected is all not frozen
Super side, this super side place subset is designated as into S, if S be equal to Q, this super side is moved to the migration of non-S subsets from subset Q
What income deducted node cuts off cost value；
Step 3.4.5.9, selects another node of connecting hyper-edge n*, repeat step 3.4.5.3～3.4.5.8, until traversal
All nodes of complete connecting hyper-edge n*.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611072775.5A CN106557581B (en) | 2016-11-29 | 2016-11-29 | Hypergraph division method based on multi-level framework and hyperedge migration |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611072775.5A CN106557581B (en) | 2016-11-29 | 2016-11-29 | Hypergraph division method based on multi-level framework and hyperedge migration |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106557581A true CN106557581A (en) | 2017-04-05 |
| CN106557581B CN106557581B (en) | 2021-02-12 |
Family
ID=58445555
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611072775.5A Active CN106557581B (en) | 2016-11-29 | 2016-11-29 | Hypergraph division method based on multi-level framework and hyperedge migration |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106557581B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112183001A (en) * | 2020-10-10 | 2021-01-05 | 上海国微思尔芯技术股份有限公司 | Hypergraph-based multistage clustering method |
| CN112183000A (en) * | 2020-10-10 | 2021-01-05 | 上海国微思尔芯技术股份有限公司 | Hypergraph partitioning method supporting interconnection constraint |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103885839A (en) * | 2014-04-06 | 2014-06-25 | 孙凌宇 | Cloud computing task scheduling method based on multilevel division method and empowerment directed hypergraphs |
| US20150023491A1 (en) * | 2012-09-07 | 2015-01-22 | Genesys Telecommunications Laboratories, Inc. | Method of distributed aggregation in a call center |
| CN104679966A (en) * | 2015-03-26 | 2015-06-03 | 孙凌宇 | Empowerment hypergraph optimized partitioning method based on multilayer method and discrete particle swarm |
| WO2015169029A1 (en) * | 2014-05-05 | 2015-11-12 | 中兴通讯股份有限公司 | Graph data partitioning method and device |
-
2016
- 2016-11-29 CN CN201611072775.5A patent/CN106557581B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150023491A1 (en) * | 2012-09-07 | 2015-01-22 | Genesys Telecommunications Laboratories, Inc. | Method of distributed aggregation in a call center |
| CN103885839A (en) * | 2014-04-06 | 2014-06-25 | 孙凌宇 | Cloud computing task scheduling method based on multilevel division method and empowerment directed hypergraphs |
| WO2015169029A1 (en) * | 2014-05-05 | 2015-11-12 | 中兴通讯股份有限公司 | Graph data partitioning method and device |
| CN104679966A (en) * | 2015-03-26 | 2015-06-03 | 孙凌宇 | Empowerment hypergraph optimized partitioning method based on multilayer method and discrete particle swarm |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112183001A (en) * | 2020-10-10 | 2021-01-05 | 上海国微思尔芯技术股份有限公司 | Hypergraph-based multistage clustering method |
| CN112183000A (en) * | 2020-10-10 | 2021-01-05 | 上海国微思尔芯技术股份有限公司 | Hypergraph partitioning method supporting interconnection constraint |
| CN112183000B (en) * | 2020-10-10 | 2023-03-07 | 上海思尔芯技术股份有限公司 | Hypergraph partitioning method supporting interconnection constraint |
| CN112183001B (en) * | 2020-10-10 | 2023-07-04 | 上海思尔芯技术股份有限公司 | Hypergraph-based multistage clustering method for integrated circuits |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106557581B (en) | 2021-02-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10171284B2 (en) | Reachability-based coordination for cyclic dataflow | |
| CN104636204B (en) | A kind of method for scheduling task and device | |
| US11294958B2 (en) | Managing a distributed knowledge graph | |
| CN103765381A (en) | Parallel operation on B+ trees | |
| CN110222029A (en) | A kind of big data multidimensional analysis computational efficiency method for improving and system | |
| JP2002517833A (en) | Computer implemented scheduling system and method using abstract local search method | |
| CN105808754A (en) | Method for rapidly discovering accumulation mode from movement trajectory data | |
| CN110134668A (en) | Data migration method, device and equipment applied to block chain | |
| CN106557581A (en) | A kind of hypergraph division methods migrated based on multi-level framework and super side | |
| Pérez-Arteaga et al. | Cost comparison of lambda architecture implementations for transportation analytics using public cloud software as a service | |
| Xie et al. | Origin-based algorithms for traffic assignment: algorithmic structure, complexity analysis, and convergence performance | |
| CN108415912A (en) | Data processing method based on MapReduce model and equipment | |
| Liu et al. | Hierarchical spark: A multi-cluster big data computing framework | |
| CN113157694A (en) | Database index generation method based on reinforcement learning | |
| CN106250101A (en) | Migration before stack method for parallel processing based on MapReduce and device | |
| Engelhardt | Comparison of agile and traditional project management: Simulation of process models | |
| Li et al. | Time‐cost efficient memory configuration for serverless workflow applications | |
| CN107357691A (en) | The processing method and processing device of image file | |
| CN104598600B (en) | A kind of parallel analysis of digital terrain optimization method based on distributed memory | |
| CN104572802B (en) | For the method and system for the data for being loaded with complex relationship | |
| CN115952008A (en) | Unified scheduling method and device for server cluster resources | |
| Serpa et al. | Flexible use of temporal and spatial reasoning for fast and scalable CPU broad‐phase collision detection using KD‐Trees | |
| CN104598567A (en) | Data statistics and de-duplication method based on Hadoop MapReduce programming frame | |
| CN114153395B (en) | Object storage data life cycle management method, device and equipment | |
| CN110333860A (en) | A kind of method and relevant apparatus removing view object |
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 |