CN106095968A - The R tree-like position multiple target node split method of n dimension massive point cloud - Google Patents
The R tree-like position multiple target node split method of n dimension massive point cloud Download PDFInfo
- Publication number
- CN106095968A CN106095968A CN201610437506.8A CN201610437506A CN106095968A CN 106095968 A CN106095968 A CN 106095968A CN 201610437506 A CN201610437506 A CN 201610437506A CN 106095968 A CN106095968 A CN 106095968A
- Authority
- CN
- China
- Prior art keywords
- node
- tree
- split
- bounding box
- dimension
- 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.)
- Pending
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
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/22—Indexing; Data structures therefor; Storage structures
- G06F16/2228—Indexing structures
- G06F16/2264—Multidimensional index structures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/22—Indexing; Data structures therefor; Storage structures
- G06F16/2228—Indexing structures
- G06F16/2246—Trees, e.g. B+trees
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Software Systems (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention provides a kind of n R tree-like position multiple target node split method of dimension massive point cloud, belong to product reverse-engineering field, for solving the Construct question of n dimension massive point cloud R tree index, it is characterised in that: the point data in a cloud file is read in linear list storage organization；Point data in linear list is inserted one by one in R tree index；If node generation overflow, then overflow node is divided；It is split axis that chosen position distribution function and distribution of shapes function are the axle minimized；Overflow node child node is being divided axial coordinate components ascending sort by its bounding box central point, and is producing candidate with the minimum node number allowed in non-root node for restrictive condition and divide disaggregationQ；ObtainQPareto optimal solution setP, and willPThe value of middle Silhouette value maximum is as node split result.The method can be that in reverse-engineering, surface in kind magnanimity sampled data builds R tree index, and constructed R tree index has the features such as structure efficiency is fast, efficiency data query is fast.
Description
Technical field
The present invention provides a kind of n R tree-like position multiple target node split method of dimension massive point cloud, belongs to product reverse-engineering
Field.
Background technology
R tree index structure can meet the numbers such as the spatial point during curve reestablishing, tri patch, patch surface well
According to dynamic insertion, delete, the operational requirements such as inquiry, it is widely used in CAD/CAM, GIS-Geographic Information System, medical image divide
The fields such as analysis, rehabilitating historic building.
For the literature search discovery of at present relevant R tree node splitting algorithm, Sellis etc. academic journal "Computer Science DepartmentScientific paper " the The R+-tree:A dynamic index for multi-delivered on "
Dimensional objects " in propose R+ tree a certain certain data objects is stored in multiple leaf index node, it is to avoid
Overlap between sibling, improves the recall precision of R tree.Beckmann etc. are at collection of thesis " Proceedings of
The 2009 ACM SIGMOD International Conference on Management of data " 2009:799-
Scientific paper " the A revised r*-tree in comparison with related index delivered on 812
Structures " in propose the division of RR* tree node time choose optimum division according to node bounding box initial center weighted deviation
Solve so that data are inserted and significantly improved with space querying efficiency.Theodoridis etc. are at collection of thesis " Advances in
Databases and Information Systems " scientific paper " the Revisiting r-that delivers on 2002:149-162
Tree construction principles " in propose CR tree be many bunches of divisions by two bunches of traditional split transforamtions, utilize k
Means clustering algorithm obtains division and solves, and reduces the calculation cost in fission process, improves achievement efficiency, it is not necessary to force weight
Insert and wait complex technology.Sleit etc. are at academic journal " Journal of Information Science " 2014,40 (2): 222-
Scientific paper " the Corner-based splitting:An improved node splitting delivered on 236
Algorithm for R-tree " in propose CBS tree division node time with each node bounding box summit as classification center, sharp
Realize node split with diagonal angle Vertex Clustering, decrease candidate solution, improve achievement efficiency.
The optimisation strategy that R tree construction is used by above-mentioned R tree variant when node split is substantially multiple-objection optimization plan
Slightly. for dividing two or more new nodes of gained, at least need the weight between their bounding box space and bounding box
Folded space carries out minimizing optimization.Practice have shown that, this multiple-objection optimization strategy can improve R tree construction to a certain extent,
Therefore, node split problem can be considered as a multi-objective optimization question to solve.Existing main flow R tree variant is by this
Multi-objective optimization question is converted to what the single-object problem of a series of cascade solved, and i.e. subjective gives multiple targets
Primary and secondary divides, and optimizes one by one.Owing to each optimization aim involved during node split is not independent of one another, between them
There is certain dependency, use the solving result of the single object optimization method of cascade that optimality may be caused only to meet certain
Main target, and show very poor in other targets.
In sum, those skilled in the art have been become for multiple-objection optimization strategy during node split urgently to be resolved hurrily
Technical problem.
Summary of the invention
It is an object of the invention to provide the R tree-like position multiple target node split method of a kind of n dimension massive point cloud, the method
Utilize Pareto Optimization Method node split multi-objective optimization question, and carry out with shape information according to the position of child node
Choosing split axis preamble optimization, its technical scheme is:
The R tree-like position multiple target node split method of a kind of n dimension massive point cloud, it is characterised in that step is followed successively by: one, will some cloud
Data point in data file is added in linear list storage organization；Two, the data point in linear list is inserted one by one R tree index
In, if node generation overflow, then select axle to divide overflow node: a) by child node by its bounding box central point at each axle
To coordinate components ascending sort, and with underflow parameter for restrictive condition produce each axial candidate divide disaggregation；B) time is chosen
Occupied by choosing division disaggregation, the axle of space minimum is split axis；C) candidate obtaining split axis axial divides disaggregation；D) from c)
The candidate obtained divides to solve to concentrate and chooses the candidate that node bounding box area of space and overlapping space region all minimize and divide solution
For node split result.
For realizing goal of the invention, the R tree-like position multiple target node split method of described n dimension massive point cloud, its feature exists
In step b) in step 2, the position considering overflow node child node realizes choosing of split axis with shape information,
Its step particularly as follows: (1) set overflow node child node bounding box center point setc i , initialize coordinate axesaIt is 1；(2) meter
Calculate, , Forc i ?aThe projection coordinate of axle,ForIn minima,E (d a ) representAverage,MFor the child node number in overflow node；(3)
Calculate,It isiIndividual child node bounding boxaThe length of side of dimension；(4) calculate (p(a
) , s(a)) and (0,0) distanced a , and willd a Add to setd a In }；(5) makeaIncrease 1, return (2), untila
= nTill, n is the dimension of node bounding box；(6) choosed a In }d a Value maximumaAxle is split axis.
For realizing goal of the invention, the R tree-like position multiple target node split method of described n dimension massive point cloud, its feature exists
In step d) in step 2, use multiple-objection optimization strategy to divide solution from candidate and concentrate acquisition node split result, its step
Rapid particularly as follows: (1) set candidate divide disaggregation asQ {(L i , R i ),QMiddle element numberk, initializing set T is empty set
Close,iIt is 1；(2) calculateQ i Middle candidate divides the girth sum of solutionWith overlapping
Degree,Expression node bounding box girth, and will (p i , o i ) add toT
In；(3) makeiIncrease 1, continue executing with (2), untili =kTill；(4) Pareto sort method is utilized to obtainTPareto
Optimal solution setT * : a) initializemIt is 1；If b), andSo thatT m >T n Set up, then perform c),
Otherwise, willT m Add set toT * In, whereinT m >T n Representp m >p n Ando m >o n ；C) makemIncrease 1, continue executing with
B), untilm =kTill；(5) basisTWithQOne-to-one relationship, obtainQPareto optimal solution setQ * ；(6)
FromQ * In choose the maximum division solution of Silhouette value (L i ,R i ) as node split result, Silhouette value is counted
Calculation formula is:
In formulaMBy the maximum child node number allowed in non-root node,s (f j ) it is overflow node child nodef j 's
Silhouette index, if, thens(f j ) computing formula is:
In formulaa (f j ) representf j Respectively withL i In other node constitute the average of girth sum of bounding box,b (f j ) representf j
Respectively withR i Middle node constitutes the average of the girth sum of bounding box.
The present invention compared with prior art, has the advantage that
(1) morpheme multiple target is chosen split axis and is made the distribution of spatial distribution of nodes and data more with consistency, and candidate is divided disaggregation
Simplified, improve R tree and build efficiency, and the situation that there is long and narrow bounding box can be processed, rely on less structure and join
Number, therefore node split algorithm has stronger data adaptability herein, it is easy to accomplish；
(2) based on Pareto Optimization Method node split optimal solution On The Choice so that by girth sum, degree of overlapping energy
Enough parallel optimum, improve R tree performance, improve spatial neighbors search efficiency；
(3) the morpheme multiple target node split algorithm using R tree builds R tree so that node split is the most reasonable, node overlapping
Degree significantly reduces, k NN Query efficiency increases, to spatial point, tri patch, patch surface during raising curve reestablishing
Have great importance etc. the treatment effeciency of data.
Accompanying drawing explanation
Fig. 1 is to utilize the inventive method to build the program flow diagram of R tree for three-dimensional massive point cloud；
Fig. 2 is one group of nodal set schematic diagram containing long and narrow node bounding box；
Fig. 3-Fig. 5 is to use the result figure of nodal set in RR* tree algorithm, CR tree algorithm, the inventive method split graph 2 respectively；
Fig. 6 is to implement node split to test the surface in kind sampling point figure of buddha point cloud model used；
Fig. 7-Fig. 9 is each internal index node that the figure of buddha point cloud model using RR* tree algorithm to be different scales builds R tree respectively
Layer and the node bounding box distribution results figure of leaf index node layer；
Figure 10-Figure 12 is each internal index knot that the figure of buddha point cloud model using CR tree algorithm to be different scales builds R tree respectively
Point layer and the node bounding box distribution results figure of leaf index node layer；
Figure 13-Figure 15 is each internal index that the figure of buddha point cloud model using the inventive method to be different scales builds R tree respectively
The node bounding box distribution results figure of node layer and leaf index node layer；
Figure 16-18 is the figure of buddha point cloud model structure using RR* tree algorithm, CR tree algorithm, the inventive method to be different scales respectively
Build the degree of overlapping result figure of the inside index node of R tree；
Figure 19 is that the figure of buddha point cloud model using RR* tree algorithm, CR tree algorithm, the inventive method to be different scales builds R respectively
Tree required time comparing result figure；
Figure 20 is that the figure of buddha point cloud model using RR* tree algorithm, CR tree algorithm, the inventive method to be different scales builds R respectively
The k NN Query time comparing result figure of tree.
Detailed description of the invention
Below in conjunction with the accompanying drawings and embodiment the invention will be further described.
Utilize the inventive method to build the program flow diagram of R tree as shown in Figure 1 for three-dimensional massive point cloud, wherein realize journey
Language used by sequence is C.This program main modular includes being stored in by cloud data in linear memory structure, each according to overflow node
Axial morpheme distribution function choose split axis, according to child node bounding box central point in split axis coordinate components ascending order row
Sequence be division candidate's disaggregation that restrictive condition obtains overflow node with the minimum node number allowed in non-root nodeQ, obtain
TakeQPareto optimal solution setPAnd choosePThe division solution of middle Silhouette value maximum is as node split result etc..
As in figure 2 it is shown, be one group of nodal set schematic diagram containing long and narrow node bounding box, it is respectively adopted RR* tree algorithm, CR tree
The nodal set in node split method split graph 2 in algorithm, the inventive method, the minimum node that the most non-root node is allowed
Number m takes 0.2M, and result figure is respectively shown in Fig. 3, Fig. 4, Fig. 5.Comparison diagram 3-Fig. 5 finds, relative to RR* tree algorithm, CR tree
Algorithm, utilizes the bounding box shape size the most uniform of the division result that the nodal set in the inventive method split graph 2 obtains
Cause.
As shown in Figure 6, it is to implement node split to test the surface in kind sampling point figure of buddha point cloud model used, uses
Optical grating projection formula 3 D measuring instrument obtains, and some cloud point number is 54, and 298, and it is carried out simplifying in various degree, simplify Factor minute
It is not 0.2,0.4,0.6,0.8,1.0, obtains 5 point setsA 、 B 、 C 、 D 、 EIf,S ={A,B,C,D,E }。
Be respectively adopted RR* tree algorithm, CR tree algorithm, the inventive method are that S builds R tree and indexes, and the most non-root node is permitted
Maximum node number M permitted takes 30, and minimum node number m that non-root node is allowed takes 0.2M, Fig. 7-Fig. 9 for utilizing RR* tree
Algorithm builds inside index node layer and the node bounding box distribution results figure of leaf index node layer of R tree, and Figure 10-Figure 12 is profit
Inside index node layer and the node bounding box distribution results figure of leaf index node layer, the Qi Zhongtu of R tree is built by CR tree algorithm
13-Figure 15 is to utilize the inventive method to build the inside index node layer of R tree and the node bounding box distribution of leaf index node layer
Result figure.Comparison diagram 7-Figure 15 finds, for RR* tree algorithm, CR tree algorithm, at figure of buddha dot cloud hole, utilizes this
The R tree that inventive method builds occurs without bounding box transboundary and big bounding box at leaf node layer, and the bounding box of each node index level
Size is more consistent, and this just explanation, compared with CR tree, RR* tree algorithm, the spatial distribution of nodes of the inventive method is distributed more with data
Concordance.
As shown in figs. 16-18, it is to use RR* tree algorithm, CR tree algorithm, the inventive method to be respectivelySBuild R tree index
Internal node degree of overlapping result figure, maximum node number M that the most non-root node is allowed takes 30, and non-root node is allowed
Minimum node number m takes 0.2M.Comparison diagram 16-Figure 18 finds, relative to RR* tree algorithm, CR tree algorithm, the weight of the inventive method
Folded angle value is minimum.
Be utilized respectively RR* tree algorithm, CR tree algorithm, the inventive method are the R tree index structure constructed by S, and use c language
The R tree structure time of time function three kinds of methods of statistics of speech, then the time comparing result figure of these three method is as shown in figure 19,
Maximum node number M that the most non-root node is allowed takes 30, and minimum node number m that non-root node is allowed takes 0.2M.By
Figure 19 understands, and relative to RR* tree algorithm, CR tree algorithm, the inventive method achievement minimal time, illustrates that the inventive method can be effective
Improve the structure efficiency of R tree.
Be utilized respectively RR* tree algorithm, CR tree algorithm, the inventive method are the R tree index structure constructed by S, carry out k for S
NN Query also uses the time function of c language to add up the k NN Query time of three kinds of methods, then the k neighbour of these three method
As shown in figure 20, maximum node number M that the most non-root node is allowed takes 30 to query time comparing result figure, non-root node institute
Minimum node number m allowed takes 0.2M, k and takes 20.As shown in Figure 20, relative to RR* tree algorithm, CR tree algorithm, side of the present invention
The k NN Query minimal time of method, illustrates that the inventive method can be effectively improved the spatial neighbors search efficiency of R tree.
The above, be only presently preferred embodiments of the present invention, is not the restriction that the present invention makees other form, appoints
What those skilled in the art changed possibly also with the technology contents of the disclosure above or be modified as equivalent variations etc.
Effect embodiment.But every without departing from technical solution of the present invention content, the technical spirit of the foundation present invention is to above example institute
Any simple modification, equivalent variations and the remodeling made, still falls within the protection domain of technical solution of the present invention.
Claims (3)
1. the R tree-like position multiple target node split method of a n dimension massive point cloud, it is characterised in that step is followed successively by:, by point
Data point in cloud data file is added in linear list storage organization；Two, the data point in linear list is inserted one by one R tree rope
In drawing, if node generation overflow, then select axle to divide overflow node: a) by child node by its bounding box central point respectively
Axial coordinate components ascending sort, and divide disaggregation with underflow parameter for the restrictive condition each axial candidate of generation；B) choose
The axle that candidate divides space occupied by disaggregation minimum is split axis；C) candidate obtaining split axis axial divides disaggregation；D) from
C) candidate obtained divides solution concentration and chooses candidate's division that node bounding box area of space all minimizes with overlapping space region
Solve as node split result.
The R tree-like position multiple target node split method of n the most according to claim 1 dimension massive point cloud, it is characterised in that step
In step b) in rapid two, the position considering overflow node child node realizes choosing of split axis with shape information, its step
Rapid particularly as follows: (1) set overflow node child node bounding box center point setc i , initialize coordinate axesaIt is 1；(2) calculate, , Forc i ?aThe projection coordinate of axle,ForIn minima,E (d a ) representAverage,MFor the child node number in overflow node；(3) calculate,It isiIndividual child node bounding boxaThe length of side of dimension；(4) calculate (p (a ) ,s (a)) and (0,0) distanced a , and willd a Add to setd a In }；(5) makeaIncrease 1, return (2), untila =n
Till, n is the dimension of node bounding box；(6) choosed a In }d a Value maximumaAxle is split axis.
The R tree-like position multiple target node split method of n the most according to claim 1 dimension massive point cloud, it is characterised in that step 2
In step d) in, use multiple-objection optimization strategy to divide solution from candidate and concentrate and obtain node split result, its step is particularly as follows: (1)
If candidate divides disaggregationQ {(L i , R i ),QMiddle element numberk, initializing set T is null set,iIt is 1；(2) meter
CalculateQ i Middle candidate divides the girth sum of solutionWith degree of overlapping,Expression node bounding box girth, and will (p i , o i ) add toTIn；(3) makeiIncrease 1, continue executing with (2), directly
Extremelyi =kTill；(4) Pareto sort method is utilized to obtainTPareto optimal solution setT * : a) initializemIt is 1；b)
If, andSo thatT m >T n Set up, then perform c), otherwise, willT m Add set toT * In, whereinT m >T n Representp m > p n Ando m >o n ；C) makemIncrease 1, continue executing with b), untilm =kTill；(5) basisTWithQ
One-to-one relationship, obtainQPareto optimal solution setQ * ；(6) fromQ * In choose the division that Silhouette value is maximum
Solve (L i , R i ) as node split result, Silhouette value computing formula is:
In formulaMBy the maximum child node number allowed in non-root node,s (f j ) it is overflow node child nodef j 's
Silhouette index, if, thens (f j ) computing formula is:
In formulaa (f j ) representf j Respectively withL i In other node constitute the average of girth sum of bounding box,b (f j ) representf j
Respectively withR i Middle node constitutes the average of the girth sum of bounding box.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610437506.8A CN106095968A (en) | 2016-06-20 | 2016-06-20 | The R tree-like position multiple target node split method of n dimension massive point cloud |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610437506.8A CN106095968A (en) | 2016-06-20 | 2016-06-20 | The R tree-like position multiple target node split method of n dimension massive point cloud |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106095968A true CN106095968A (en) | 2016-11-09 |
Family
ID=57236998
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610437506.8A Pending CN106095968A (en) | 2016-06-20 | 2016-06-20 | The R tree-like position multiple target node split method of n dimension massive point cloud |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106095968A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108509438A (en) * | 2017-02-24 | 2018-09-07 | 南京烽火星空通信发展有限公司 | A kind of ElasticSearch fragments extended method |
| CN109697733A (en) * | 2018-12-26 | 2019-04-30 | 广州文远知行科技有限公司 | Point searching method and device in point cloud space, computer equipment and storage medium |
| CN111615792A (en) * | 2018-01-18 | 2020-09-01 | 黑莓有限公司 | Method and apparatus for entropy coding of point clouds |
| CN112765405A (en) * | 2019-10-21 | 2021-05-07 | 千寻位置网络有限公司 | Method and system for clustering and inquiring spatial data search results |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5987468A (en) * | 1997-12-12 | 1999-11-16 | Hitachi America Ltd. | Structure and method for efficient parallel high-dimensional similarity join |
| CN102880675A (en) * | 2012-09-11 | 2013-01-16 | 山东理工大学 | Dynamic index self-adaptive construction method of product reverse engineering data based on mean value shift |
| CN104731984A (en) * | 2015-04-22 | 2015-06-24 | 山东理工大学 | Incremental clustering optimization solution method for splitting problems of overflow nodes of R trees |
-
2016
- 2016-06-20 CN CN201610437506.8A patent/CN106095968A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5987468A (en) * | 1997-12-12 | 1999-11-16 | Hitachi America Ltd. | Structure and method for efficient parallel high-dimensional similarity join |
| CN102880675A (en) * | 2012-09-11 | 2013-01-16 | 山东理工大学 | Dynamic index self-adaptive construction method of product reverse engineering data based on mean value shift |
| CN104731984A (en) * | 2015-04-22 | 2015-06-24 | 山东理工大学 | Incremental clustering optimization solution method for splitting problems of overflow nodes of R trees |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108509438A (en) * | 2017-02-24 | 2018-09-07 | 南京烽火星空通信发展有限公司 | A kind of ElasticSearch fragments extended method |
| CN108509438B (en) * | 2017-02-24 | 2021-08-31 | 南京烽火星空通信发展有限公司 | ElasticSearch fragment expansion method |
| CN111615792A (en) * | 2018-01-18 | 2020-09-01 | 黑莓有限公司 | Method and apparatus for entropy coding of point clouds |
| CN111615792B (en) * | 2018-01-18 | 2024-05-28 | 黑莓有限公司 | Method and apparatus for entropy encoding a point cloud |
| CN109697733A (en) * | 2018-12-26 | 2019-04-30 | 广州文远知行科技有限公司 | Point searching method and device in point cloud space, computer equipment and storage medium |
| CN112765405A (en) * | 2019-10-21 | 2021-05-07 | 千寻位置网络有限公司 | Method and system for clustering and inquiring spatial data search results |
| CN112765405B (en) * | 2019-10-21 | 2022-11-25 | 千寻位置网络有限公司 | Method and system for clustering and inquiring spatial data search results |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Malkov et al. | Efficient and robust approximate nearest neighbor search using hierarchical navigable small world graphs | |
| Zhang et al. | An end-to-end deep learning architecture for graph classification | |
| CN106095968A (en) | The R tree-like position multiple target node split method of n dimension massive point cloud | |
| Asudeh et al. | Efficient computation of regret-ratio minimizing set: A compact maxima representative | |
| CN103116625A (en) | Volume radio direction finde (RDF) data distribution type query processing method based on Hadoop | |
| Pulgar-Rubio et al. | MEFASD-BD: multi-objective evolutionary fuzzy algorithm for subgroup discovery in big data environments-a mapreduce solution | |
| CN107292186A (en) | A kind of model training method and device based on random forest | |
| CN107291847A (en) | A kind of large-scale data Distributed Cluster processing method based on MapReduce | |
| CN104699698A (en) | Graph query processing method based on massive data | |
| CN102819569A (en) | Matching method for data in distributed interactive simulation system | |
| CN109840551B (en) | Method for optimizing random forest parameters for machine learning model training | |
| CN104158182A (en) | Large-scale power grid flow correction equation parallel solving method | |
| CN104731984B (en) | Automobile wheel hub surface sampling point R tree overflow node incremental clustering optimization method | |
| CN108833302A (en) | Resource allocation methods under cloud environment based on fuzzy clustering and stringent bipartite matching | |
| CN103049555A (en) | Dynamic hierarchical integrated data accessing method capable of guaranteeing semantic correctness | |
| Wang et al. | An efficient algorithm for distributed outlier detection in large multi-dimensional datasets | |
| Liang et al. | Scalable 3d spatial queries for analytical pathology imaging with mapreduce | |
| Reis et al. | An evaluation of data model for NoSQL document-based databases | |
| Yu et al. | DBWGIE-MR: A density-based clustering algorithm by using the weighted grid and information entropy based on MapReduce | |
| Mežnar et al. | Snore: Scalable unsupervised learning of symbolic node representations | |
| Prokopenko et al. | A single-tree algorithm to compute the Euclidean minimum spanning tree on GPUs | |
| CN108596390A (en) | A method of solving Vehicle Routing Problems | |
| Galicia et al. | Rdfpartsuite: bridging physical and logical RDF partitioning | |
| CN104111947A (en) | Retrieval method of remote sensing images | |
| Yue et al. | Time-based trajectory data partitioning for efficient range query |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20161109 |
|
| RJ01 | Rejection of invention patent application after publication |