CN105046382A - Heterogeneous system parallel random forest optimization method and system - Google Patents
Heterogeneous system parallel random forest optimization method and system Download PDFInfo
- Publication number
- CN105046382A CN105046382A CN201510591067.1A CN201510591067A CN105046382A CN 105046382 A CN105046382 A CN 105046382A CN 201510591067 A CN201510591067 A CN 201510591067A CN 105046382 A CN105046382 A CN 105046382A
- Authority
- CN
- China
- Prior art keywords
- node
- data fragmentation
- coprocessor
- host node
- data
- 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
Abstract
The invention discloses a heterogeneous system parallel random forest optimization system and method. The heterogeneous system parallel random forest optimization system is applied to a central processing unit and a coprocessor mixed heterogeneous cluster and comprises one master node and multiple slave nodes, wherein the main master node is used for dividing a data file to be calculated into multiple data fragments and transmitting the data fragments to all the slave nodes respectively; all the slave nodes are used for receiving the data fragments distributed by the master node to calculate the data fragments and transmitting a calculated optimal solution building decision-making tree to the master node to generate a random forest. As the data fragments are calculated by the slave nodes in parallel, the time for searching for the optimal solution is shortened, efficiency of the whole system is substantially improved, the system is unnecessarily limited by network bandwidth deficiency, small memory capacity and other conditions, and the requirement for processing large-scale data of high-performance application is met.
Description
Technical field
The present invention relates to machine learning field, espespecially a kind of heterogeneous system walks abreast random forest optimization system and method.
Background technology
In recent years, along with the development that society economy science and technology is maked rapid progress, many applications a large amount of data in run-up, to the information that these data analysis contain to excavate data, become the joint demand in nearly all field, and in actual applications, machine learning is day by day important in the effect of data mining analysis technology, receives and pay close attention to widely.
In prior art, sorting technique the most frequently used in machine learning adopts random forests algorithm exactly, and random forest is a kind of integrated study sorting technique having supervision, is to gather by a large amount of classification tree the precision of prediction improving model.This design just because of it makes random forests algorithm have good tolerance to the abnormal data in sample data and noise etc., and the classification problem for data more complicated has good concurrency and extendability.
But adopt prior art, find the optimum solution time in data-optimized process long, in this data volume current with the exponential epoch increased, the process obviously for super large data seems unable to do what one wishes.
Summary of the invention
In order to solve the problems of the technologies described above, the invention provides a kind of heterogeneous system to walk abreast random forest optimization system and method, can by multiple from node to data fragmentation parallel computation, thus accelerate the time finding optimum solution, whole system efficiency is significantly promoted, do not need to be limited to the situation such as network bandwidth deficiency, memory size be little, meet the requirement that performance application carries out for large-scale data processing.
First aspect, the invention provides a kind of heterogeneous system and to walk abreast random forest optimization system, be applied to central processing unit and coprocessor mixing isomeric group, comprise: a host node and multiple from node;
Described host node is used for data file to be calculated to be divided into multiple data fragmentation, sends data fragmentation respectively to each described from node, receives each described decision tree built from node and generates random forest;
Described from node for receive described host node distribute described data fragmentation calculate, by calculate after optimum solution build decision tree be sent to described host node.
Second aspect, the invention provides a kind of heterogeneous system and to walk abreast random forest optimization method, and be applied to heterogeneous system and walk abreast random forest optimization system, it is characterized in that, described system comprises: a host node and multiple from node;
Described host node calls and data file to be calculated is divided into multiple data fragmentation, sends data fragmentation respectively to each described from node, receives each described decision tree built from node and generates random forest;
Described from node receive described host node distribute described data fragmentation calculate, by calculate after optimum solution build decision tree be sent to described host node.
Compared with prior art, the invention provides a kind of heterogeneous system and to walk abreast random forest optimization system and method, comprising: a host node and multiple from node, wherein, host node is used for data file to be calculated to be divided into multiple data fragmentation, send data fragmentation respectively to each from node, respectively calculate from node for the data fragmentation receiving host node distribution, optimum solution structure decision tree after calculating is sent to host node, thus generation random forest, by multiple from node to data fragmentation parallel computation, thus accelerate the time finding optimum solution, whole system efficiency is significantly promoted, do not need to be limited to network bandwidth deficiency, the situations such as memory size is little, meet the requirement that performance application carries out for large-scale data processing.
Other features and advantages of the present invention will be set forth in the following description, and, partly become apparent from instructions, or understand by implementing the present invention.Object of the present invention and other advantages realize by structure specifically noted in instructions, claims and accompanying drawing and obtain.
Accompanying drawing explanation
Accompanying drawing is used to provide the further understanding to technical solution of the present invention, and forms a part for instructions, is used from and explains technical scheme of the present invention, do not form the restriction to technical solution of the present invention with the embodiment one of the application.
Fig. 1 to walk abreast the structural representation of random forest optimization system embodiment one for heterogeneous system that the embodiment of the present invention provides;
Fig. 2 to walk abreast the schematic flow sheet of random forest optimization method embodiment one for heterogeneous system that the embodiment of the present invention provides.
Embodiment
For making the object, technical solutions and advantages of the present invention clearly understand, hereinafter will be described in detail to embodiments of the invention by reference to the accompanying drawings.It should be noted that, when not conflicting, the embodiment in the application and the feature in embodiment can combination in any mutually.
Can perform in the computer system of such as one group of computer executable instructions in the step shown in the process flow diagram of accompanying drawing.Further, although show logical order in flow charts, in some cases, can be different from the step shown or described by order execution herein.
The heterogeneous system that the embodiment of the present invention relates to walks abreast random forest optimization system, be applied to central processing unit and coprocessor mixing isomeric group, can be specifically on computer cluster or server cluster, wherein coprocessor can be graphic process unit (GraphicsProcessingUnit, GPU) or many-core processor (IntelManyIntegratedCore be called for short:, be called for short: mic card), but not as limit.
The system and method that the embodiment of the present invention relates to, is intended to adopt the sorting technique data of random forest in optimizing process, find the optimum solution time in solution prior art long, cannot meets the technical matters of ultra-large data processing.
With embodiment particularly, technical scheme of the present invention is described in detail below.These specific embodiments can be combined with each other below, may repeat no more for same or analogous concept or process in some embodiment.
Fig. 1 to walk abreast the structural representation of random forest optimization system embodiment one for heterogeneous system that the embodiment of the present invention provides, as shown in Figure 1, this heterogeneous system random forest optimization system that walks abreast is applied to central processing unit and coprocessor mixing isomeric group, comprising: a host node 10 and multiple from node 20;
Wherein, described host node 10, for data file to be calculated is divided into multiple data fragmentation, sends data fragmentation respectively to each described from node 20, receives each described decision tree built from node 20 and generates random forest.
Described from node 20 for receive described host node distribute described data fragmentation calculate, by calculate after optimum solution build decision tree be sent to described host node 10.
Concrete, data file to be calculated is divided into multiple data fragmentation by host node 10, GDF general data file is exponential ultra-large data, can Using Call Library Function divide data fragmentation size be 16M to 64M, specifically can determine according to the computing power of child node 20, can be numbered each data fragmentation, by the mode of broadcast, ready-portioned data fragmentation is distributed to according to number order each from node 20, but be not limited to this.
In general arrange more from node 20, then process large-scale data more accurate, the size of the concrete data file that can process as required arranges the quantity from node 20, after respectively receiving from node 20 data fragmentation that host node 10 distributes corresponding, respectively from the data fragmentation that node 20 can be distributed by the parallel computation of Bagging integrated study technology, optimum solution in result after seletion calculation builds decision tree respectively, wherein, this optimum solution can be minimize, the data fragmentation corresponding according to this minimum value builds decision tree, and send to host node 10 to carry out the integration of random forest.
The heterogeneous system that the embodiment of the present invention provides walks abreast random forest optimization system, be applied to central processing unit and coprocessor mixing isomeric group, comprise at least one host node and multiple from node, wherein, host node is used for data file to be calculated to be divided into multiple data fragmentation, send data fragmentation respectively to each from node, respectively calculate from node for the data fragmentation receiving host node distribution, optimum solution structure decision tree after calculating is sent to host node, thus generation random forest, the embodiment of the present invention by multiple from node to data fragmentation parallel computation, thus accelerate the time finding optimum solution, whole system efficiency is significantly promoted, do not need to be limited to network bandwidth deficiency, the situations such as memory size is little, meet the requirement that performance application carries out for large-scale data processing.
Further, from node 20 comprise at least one central processing unit 210 (CentralProcessingUnit, be called for short: CPU) and multiple coprocessor 220 (ManyIntegratedCore, abbreviation: MIC),
The described described data fragmentation distributed for receiving described host node 10 from node 20, comprise: described central processing unit 210 receives described data fragmentation, described data fragmentation is divided into multiple data fragmentation subset, distribute corresponding described data fragmentation subset to each described thread, dispatch thread gives each described coprocessor 220;
Concrete, in order to better the present embodiment is described, suppose that this comprises a central processing unit 210 and 4 pieces of coprocessors 220 from node 20, the data fragmentation of reception is divided into multiple data fragmentation subset by this central processing unit 210, can divide according to the computing power of these 4 pieces of coprocessors 220, distribute corresponding data and distribute subset to each thread, and dispatch thread gives each coprocessor 220, thread and coprocessor 220 are one to one, if there are 4 threads, then distribute 1 thread to a coprocessor 220.
Described from node 20 for calculate described data fragmentation and by calculate after optimum solution build decision tree be sent to described host node 10, comprise: described coprocessor 220 receives the corresponding corresponding described data fragmentation subset sums initial value of thread acquisition and calculates, obtain optimum Split Attribute structure decision tree according to the result after calculating and send to described host node 10.
Concrete, coprocessor 220 receives corresponding thread and obtains corresponding described data fragmentation subset sums initial value, this data fragmentation subset is the data acquisition needing to carry out computing, this initial value refers to the data building initial decision tree and need, these data can produce at random, also can fix, computation process can adopt geometry estimation algorithm, curve fitting method and distribution function method to obtain the measuring and calculating of Geordie (Gini) value, according to the Gini value after calculating, the general conduct selecting Gini value minimum optimum Split Attribute structure decision tree.
Further, described coprocessor 220 also comprises before calculating for the corresponding described data fragmentation subset sums initial value of thread acquisition receiving correspondence:
Described host node 10 is to each described from node 20 distribution process, and described process sends from all devices in node the thread that call request carries out calculating to described, receives the thread that call request that each described equipment returns carries out calculating; Wherein, a central processing unit is as an equipment, and one piece of coprocessor is as an equipment.
Concrete, before the data fragmentation subset sums initial value obtaining corresponding thread at coprocessor 220 calculates, host node 20 is to each from node 20 distribution process, this process can to the thread sending call request from all devices in node 20, this call request comprises the request of the information such as resource, address, and receive the thread that each described equipment returns call request, according to the thread of this call request, described coprocessor 220 can get data fragmentation and initial value.The corresponding thread of each equipment, wherein, central processing unit can as an equipment, and one piece of coprocessor can as an equipment, the quantity of all devices be all processor devices and coprocessor device and.
Further, described central processing unit 210 is multiple data fragmentation subsets for dividing data burst, refers to:
Data fragmentation is on average divided into multiple data fragmentation subset according to the quantity of described coprocessor 220 by described central processing unit 210.
Concrete, in order to better the present embodiment is described, suppose there be M thread, the data fragmentation that central processing unit 210 can will receive host node 10 and distributes, described data fragmentation is on average divided into data fragmentation subset, described data fragmentation subset is then the data fragmentation of 1/M, each data fragmentation subset is sent to from each coprocessor 220 in node 20 respectively, when being the optimum Split Attribute of last selection like this, select from M attribute with regard to only needing, the time so calculating Gini value can greatly reduce, and can reduce to 1/M in optimal situation.
Further, described coprocessor 220 sends to described host node 10 for obtaining optimum Split Attribute structure decision tree according to the result after calculating, refers to:
Result after described coprocessor 220 section of sentencing calculates is optimum solution, is if so, then that the data fragmentation subset of the correspondence of optimum solution builds decision tree as optimum Split Attribute using described result of calculation, the decision tree of structure is sent to described host node; If not, then continue described other data fragmentations from node of process, until the result after calculating is that optimum solution or all data fragmentations all process.
Concrete, result after each described coprocessor 220 calculates can gather, result after all calculating is judged, such as can judge that whether minimum the Geordie value after calculating is, if, then the data fragmentation subset of correspondence minimum for Geordie value is built decision tree as optimum Split Attribute, and the decision tree of structure is sent to described host node 10; If not, then continue process other data fragmentations from node 20, until the Geordie value after calculating is minimum or all data fragmentations are all processed.
Fig. 2 to walk abreast the schematic flow sheet of random forest optimization method embodiment one for heterogeneous system that the embodiment of the present invention provides, as shown in Figure 2, the method is applied to heterogeneous system and walks abreast random forest optimization system, and it is characterized in that, described system comprises: a host node and multiple from node;
Data file to be calculated is divided into multiple data fragmentation by S101, described host node, sends data fragmentation respectively to each described from node, receives each described decision tree built from node and generates random forest;
S102, described from node receive described host node distribute described data fragmentation calculate, by calculate after optimum solution build decision tree be sent to described host node.
The heterogeneous system that the embodiment of the present invention provides walks abreast random forest optimization method, be applied to heterogeneous system to walk abreast random forest optimization system, comprise at least one host node and multiple from node, by host node, data file to be calculated is divided into multiple data fragmentation, send data fragmentation respectively to each from node, respectively calculate from node for the data fragmentation receiving host node distribution, optimum solution structure decision tree after calculating is sent to host node, thus generation random forest, achieve the parallel computation to ultra-large data file, thus accelerate the time finding optimum solution, whole system efficiency is significantly promoted, do not need to be limited to network bandwidth deficiency, the situations such as memory size is little, meet the requirement that performance application carries out for large-scale data processing.
Further, describedly at least one central processing unit and multiple coprocessor is comprised from node;
The described described data fragmentation receiving the distribution of described host node from node, comprise: described central processing unit receives described data fragmentation, described data fragmentation is divided into multiple data fragmentation subset, distribute corresponding described data fragmentation subset to each described thread, dispatch thread gives each described coprocessor;
Described from data fragmentation described in node calculate and by calculate after optimum solution build decision tree be sent to described host node, comprise: described coprocessor receives the corresponding corresponding described data fragmentation subset sums initial value of thread acquisition and calculates, obtain optimum Split Attribute structure decision tree according to the result after calculating and send to described host node.
The heterogeneous system that the embodiment of the present invention provides walks abreast random forest optimization method, and can perform said system embodiment, it realizes principle and technique effect is similar, does not repeat them here.
Further, described coprocessor receives corresponding thread and obtains before corresponding described data fragmentation subset sums initial value calculates, and also comprises:
Described host node is to each described from peer distribution process, and described process, to the described thread sending call request from all devices in node, receives the thread of the call request that each described equipment returns; Wherein, a central processing unit is as an equipment, and one piece of coprocessor is as an equipment.
The large heterogeneous system that the embodiment of the present invention provides walks abreast random forest optimization method, and can perform said system embodiment, it realizes principle and technique effect is similar, does not repeat them here.
Further, described central processing unit dividing data burst is multiple data fragmentation subsets, comprising:
Data fragmentation is on average divided into multiple data fragmentation subset according to the quantity of described coprocessor by described central processing unit.
The heterogeneous system that the embodiment of the present invention provides walks abreast random forest optimization method, and can perform said system embodiment, it realizes principle and technique effect is similar, does not repeat them here.
Further, described coprocessor obtains optimum Split Attribute structure decision tree according to the result after calculating and sends to described host node, comprising:
Described coprocessor judges that the result after calculating is optimum solution, if so, then the data fragmentation subset of the correspondence of the optimum solution after described calculating is built decision tree as optimum Split Attribute, the decision tree of structure is sent to described host node; If not, then continue to calculate described other data fragmentations from node, until the result after calculating is that optimum solution or all data fragmentations all process.
The heterogeneous system that the embodiment of the present invention provides walks abreast random forest optimization method, and can perform said system embodiment, it realizes principle and technique effect is similar, does not repeat them here.
One of ordinary skill in the art will appreciate that: all or part of step realizing said method embodiment can have been come by the hardware that programmed instruction is relevant, aforesaid program can be stored in a computer read/write memory medium, this program, when performing, performs the step comprising said method embodiment; And aforesaid storage medium comprises: ROM, RAM, magnetic disc or CD etc. various can be program code stored medium.
Although the embodiment disclosed by the present invention is as above, the embodiment that described content only adopts for ease of understanding the present invention, and be not used to limit the present invention.Those of skill in the art belonging to any the present invention; under the prerequisite not departing from the spirit and scope disclosed by the present invention; any amendment and change can be carried out in the form implemented and details; but scope of patent protection of the present invention, the scope that still must define with appending claims is as the criterion.
Claims (10)
1. heterogeneous system walks abreast a random forest optimization system, it is characterized in that, is applied to central processing unit and coprocessor mixing isomeric group, comprises: a host node and multiple from node;
Described host node is used for data file to be calculated to be divided into multiple data fragmentation, sends data fragmentation respectively to each described from node, receives each described decision tree built from node and generates random forest;
Described from node for receive described host node distribute described data fragmentation calculate, by calculate after optimum solution build decision tree be sent to described host node.
2. system according to claim 1, is characterized in that, describedly comprises at least one central processing unit and multiple coprocessor from node;
The described described data fragmentation distributed for receiving described host node from node, comprise: described central processing unit receives described data fragmentation, described data fragmentation is divided into multiple data fragmentation subset, distribute corresponding described data fragmentation subset to each described thread, dispatch thread gives each described coprocessor;
Described from node for calculate described data fragmentation and by calculate after optimum solution build decision tree be sent to described host node, comprise: described coprocessor receives the corresponding corresponding described data fragmentation subset sums initial value of thread acquisition and calculates, obtain optimum Split Attribute structure decision tree according to the result after calculating and send to described host node.
3. system according to claim 2, is characterized in that, described coprocessor also comprises before calculating for the corresponding described data fragmentation subset sums initial value of thread acquisition receiving correspondence:
Described host node is to each described from peer distribution process, and described process, to the described thread sending call request from all devices in node, receives the thread of the call request that each described equipment returns; Wherein, a central processing unit is as an equipment, and one piece of coprocessor is as an equipment.
4. system according to claim 2, is characterized in that, it is multiple data fragmentation subsets that described central processing unit is used for dividing data burst, refers to:
Data fragmentation is on average divided into multiple data fragmentation subset according to the quantity of described coprocessor by described central processing unit.
5. system according to claim 2, is characterized in that, described coprocessor sends to described host node for obtaining optimum Split Attribute structure decision tree according to the result after calculating, refers to:
The data fragmentation subset of the correspondence of the optimum solution after described calculating, for judging that the result after calculating is optimum solution, is if so, then built decision tree as optimum Split Attribute, the decision tree of structure is sent to described host node by described coprocessor; If not, then continue to calculate described other data fragmentations from node, until the result after calculating is that optimum solution or all data fragmentations all process.
6. heterogeneous system walks abreast a random forest optimization method, and be applied to heterogeneous system and walk abreast random forest optimization system, it is characterized in that, described system comprises: a host node and multiple from node;
Described host node calls and data file to be calculated is divided into multiple data fragmentation, sends data fragmentation respectively to each described from node, receives each described decision tree built from node and generates random forest;
Described from node receive described host node distribute described data fragmentation calculate, by calculate after optimum solution build decision tree be sent to described host node.
7. method according to claim 6, is characterized in that, describedly comprises at least one central processing unit and multiple coprocessor from node;
The described described data fragmentation receiving the distribution of described host node from node, comprise: described central processing unit receives described data fragmentation, described data fragmentation is divided into multiple data fragmentation subset, distribute corresponding described data fragmentation subset to each described thread, dispatch thread gives each described coprocessor;
Described from data fragmentation described in node calculate and by calculate after optimum solution build decision tree be sent to described host node, comprise: described coprocessor receives the corresponding corresponding described data fragmentation subset sums initial value of thread acquisition and calculates, obtain optimum Split Attribute structure decision tree according to the result after calculating and send to described host node.
8. method according to claim 7, is characterized in that, described coprocessor receives corresponding thread and obtains before corresponding described data fragmentation subset sums initial value calculates, and also comprises:
Described host node is to each described from peer distribution process, and described process, to the described thread sending call request from all devices in node, receives the thread of the call request that each described equipment returns; Wherein, a central processing unit is as an equipment, and one piece of coprocessor is as an equipment.
9. method according to claim 7, is characterized in that, described central processing unit dividing data burst is multiple data fragmentation subsets, comprising:
Data fragmentation is on average divided into multiple data fragmentation subset according to the quantity of described coprocessor by described central processing unit.
10. method according to claim 7, is characterized in that, described coprocessor obtains optimum Split Attribute structure decision tree according to the result after calculating and sends to described host node, comprising:
Described coprocessor judges that the result after calculating is optimum solution, if so, then the data fragmentation subset of the correspondence of the optimum solution after described calculating is built decision tree as optimum Split Attribute, the decision tree of structure is sent to described host node; If not, then continue to calculate described other data fragmentations from node, until the result after calculating is that optimum solution or all data fragmentations all process.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510591067.1A CN105046382A (en) | 2015-09-16 | 2015-09-16 | Heterogeneous system parallel random forest optimization method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510591067.1A CN105046382A (en) | 2015-09-16 | 2015-09-16 | Heterogeneous system parallel random forest optimization method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN105046382A true CN105046382A (en) | 2015-11-11 |
Family
ID=54452909
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510591067.1A Pending CN105046382A (en) | 2015-09-16 | 2015-09-16 | Heterogeneous system parallel random forest optimization method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105046382A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106339351A (en) * | 2016-08-30 | 2017-01-18 | 浪潮(北京)电子信息产业有限公司 | SGD (Stochastic Gradient Descent) algorithm optimization system and method |
| CN107508694A (en) * | 2016-06-14 | 2017-12-22 | 中兴通讯股份有限公司 | Node administration method and node device in a kind of cluster |
| CN109614978A (en) * | 2018-09-29 | 2019-04-12 | 阿里巴巴集团控股有限公司 | Data processing method, device, equipment and computer readable storage medium |
| CN110110764A (en) * | 2019-04-22 | 2019-08-09 | 福建天晴数码有限公司 | Random forest policy optimization method, storage medium based on hybrid network |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110199911A1 (en) * | 2010-02-18 | 2011-08-18 | Oki Electric Industry Co., Ltd. | Network fault detection system |
| CN102831618A (en) * | 2012-07-20 | 2012-12-19 | 西安电子科技大学 | Hough forest-based video target tracking method |
| CN104391970A (en) * | 2014-12-04 | 2015-03-04 | 深圳先进技术研究院 | Attribute subspace weighted random forest data processing method |
| CN104392250A (en) * | 2014-11-21 | 2015-03-04 | 浪潮电子信息产业股份有限公司 | Image classification method based on MapReduce |
| CN104536937A (en) * | 2014-12-30 | 2015-04-22 | 深圳先进技术研究院 | Big data appliance realizing method based on CPU-GPU heterogeneous cluster |
-
2015
- 2015-09-16 CN CN201510591067.1A patent/CN105046382A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110199911A1 (en) * | 2010-02-18 | 2011-08-18 | Oki Electric Industry Co., Ltd. | Network fault detection system |
| CN102831618A (en) * | 2012-07-20 | 2012-12-19 | 西安电子科技大学 | Hough forest-based video target tracking method |
| CN104392250A (en) * | 2014-11-21 | 2015-03-04 | 浪潮电子信息产业股份有限公司 | Image classification method based on MapReduce |
| CN104391970A (en) * | 2014-12-04 | 2015-03-04 | 深圳先进技术研究院 | Attribute subspace weighted random forest data processing method |
| CN104536937A (en) * | 2014-12-30 | 2015-04-22 | 深圳先进技术研究院 | Big data appliance realizing method based on CPU-GPU heterogeneous cluster |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107508694A (en) * | 2016-06-14 | 2017-12-22 | 中兴通讯股份有限公司 | Node administration method and node device in a kind of cluster |
| CN107508694B (en) * | 2016-06-14 | 2021-11-16 | 中兴通讯股份有限公司 | Node management method and node equipment in cluster |
| CN106339351A (en) * | 2016-08-30 | 2017-01-18 | 浪潮(北京)电子信息产业有限公司 | SGD (Stochastic Gradient Descent) algorithm optimization system and method |
| CN106339351B (en) * | 2016-08-30 | 2019-05-10 | 浪潮(北京)电子信息产业有限公司 | A kind of SGD algorithm optimization system and method |
| CN109614978A (en) * | 2018-09-29 | 2019-04-12 | 阿里巴巴集团控股有限公司 | Data processing method, device, equipment and computer readable storage medium |
| CN110110764A (en) * | 2019-04-22 | 2019-08-09 | 福建天晴数码有限公司 | Random forest policy optimization method, storage medium based on hybrid network |
| CN110110764B (en) * | 2019-04-22 | 2021-01-26 | 福建天晴数码有限公司 | Random forest strategy optimization method based on hybrid network and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US8959138B2 (en) | Distributed data scalable adaptive map-reduce framework | |
| Yang et al. | A hybrid computing method of SpMV on CPU–GPU heterogeneous computing systems | |
| CN106156159B (en) | A kind of table connection processing method, device and cloud computing system | |
| CN105022670A (en) | Heterogeneous distributed task processing system and processing method in cloud computing platform | |
| CN105046382A (en) | Heterogeneous system parallel random forest optimization method and system | |
| CN110347515B (en) | Resource optimization allocation method suitable for edge computing environment | |
| CN114418127B (en) | Machine learning calculation optimization method and platform | |
| CN106776455B (en) | Single-machine multi-GPU communication method and device | |
| CN110659278A (en) | Graph data distributed processing system based on CPU-GPU heterogeneous architecture | |
| JP2014525640A (en) | Expansion of parallel processing development environment | |
| Jain et al. | An adaptive parallel algorithm for computing connected components | |
| CN111798316A (en) | Parallel chain consensus method, device and storage medium | |
| CN113177062B (en) | Data query method and device | |
| CN104754008A (en) | Network storage node, network storage system and device and method for network storage node | |
| CN113515672A (en) | Data processing method and device, computer readable medium and electronic equipment | |
| CN105094981A (en) | Method and device for processing data | |
| CN104699946A (en) | Game scene management method and device | |
| Elnashar | Parallel performance of MPI sorting algorithms on dual-core processor windows-based systems | |
| CN108712486B (en) | Workload proving method and device | |
| CN108334532B (en) | Spark-based Eclat parallelization method, system and device | |
| CN112699134A (en) | Distributed graph database storage and query method based on graph subdivision | |
| CN109254844B (en) | Triangle calculation method of large-scale graph | |
| KR101013784B1 (en) | Hybrid parrallel continuous collision detection using cpu and gpu | |
| CN112000478A (en) | Job operation resource allocation method and device | |
| Zhang et al. | Parallel execution of stochastic search procedures on reduced SAT instances |
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: 20151111 |
|
| RJ01 | Rejection of invention patent application after publication |