CUDT: A CUDA Based Decision Tree Algorithm

Lo, Win Tsung; Chang, Yu Fan; Sheu, Ruey-Kai; Chiu, Chun Chieh; Yuan, Shyan-Ming

doi:10.1155/2014/745640

Cited by 37 publications

(17 citation statements)

References 16 publications

(15 reference statements)

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…No performance results are reported. Another decision tree construction algorithm is described in Lo et al (2014). They report speedups of 5-55Â over WEKA's Java-based implementation of C4.5 (Quinlan, 2014), called J48, and 18Â over SPRINT.…”

Section: Building Tree Classifiers On Gpusmentioning

confidence: 99%

Accelerating the XGBoost algorithm using GPU computing

Mitchell¹,

Frank²

2017

PeerJ Computer Science

207

View full text Add to dashboard Cite

We present a CUDA-based implementation of a decision tree construction algorithm within the gradient boosting library XGBoost. The tree construction algorithm is executed entirely on the graphics processing unit (GPU) and shows high performance with a variety of datasets and settings, including sparse input matrices. Individual boosting iterations are parallelised, combining two approaches. An interleaved approach is used for shallow trees, switching to a more conventional radix sort-based approach for larger depths. We show speedups of between 3Â and 6Â using a Titan X compared to a 4 core i7 CPU, and 1.2Â using a Titan X compared to 2Â Xeon CPUs (24 cores). We show that it is possible to process the Higgs dataset (10 million instances, 28 features) entirely within GPU memory. The algorithm is made available as a plug-in within the XGBoost library and fully supports all XGBoost features including classification, regression and ranking tasks.

show abstract

Section: Building Tree Classifiers On Gpusmentioning

confidence: 99%

Accelerating the XGBoost algorithm using GPU computing

Mitchell¹,

Frank²

2017

PeerJ Computer Science

207

View full text Add to dashboard Cite

show abstract

“…As for the GPU platform, we used the NVidia Geforce Titan (876 MHz, 2,496 CUDA cores, and 6GB DDR5 off-chip memory) with the same processor and the main memory running on the Ubuntu 14.04 LTS. To generate the executable code, first, we generated the RF by the scikit-learn, then, we used the CUDA Tree (CU-DAT) [13] to generate the executable code. To measure the LPS, we used 1,000 random test vectors, while to measure the power consumption excluding the idle power, we inserted the power measurement between the host PC and the power source.…”

Section: Compared With Other Platformsmentioning

confidence: 99%

An FPGA Realization of a Random Forest with <i>k</i>-Means Clustering Using a High-Level Synthesis Design

Jinguji

Sato

Nakahara

2018

IEICE Trans. Inf. & Syst.

View full text Add to dashboard Cite

Akira JINGUJI†a) , Nonmember, Shimpei SATO †b) , and Hiroki NAKAHARA †c) , Members SUMMARY A random forest (RF) is a kind of ensemble machine learning algorithm used for a classification and a regression. It consists of multiple decision trees that are built from randomly sampled data. The RF has a simple, fast learning, and identification capability compared with other machine learning algorithms. It is widely used for application to various recognition systems. Since it is necessary to un-balanced trace for each tree and requires communication for all the ones, the random forest is not suitable in SIMD architectures such as GPUs. Although the accelerators using the FPGA have been proposed, such implementations were based on HDL design. Thus, they required longer design time than the soft-ware based realizations. In the previous work, we showed the high-level synthesis design of the RF including the fully pipelined architecture and the all-toall communication. In this paper, to further reduce the amount of hardware, we use k-means clustering to share comparators of the branch nodes on the decision tree. Also, we develop the krange tool flow, which generates the bitstream with a few number of hyper parameters. Since the proposed tool flow is based on the high-level synthesis design, we can obtain the high performance RF with short design time compared with the conventional HDL design. We implemented the RF on the Xilinx Inc. ZC702 evaluation board. Compared with the CPU (Intel Xeon (R) E5607 Processor) and the GPU (NVidia Geforce Titan) implementations, as for the performance, the FPGA realization was 8.4 times faster than the CPU one, and it was 62.8 times faster than the GPU one. As for the power consumption efficiency, the FPGA realization was 7.8 times better than the CPU one, and it was 385.9 times better than the GPU one. key words: machine learning, random forest, k-means clustering, FPGA

show abstract

“…Win-Tsung Lo [5] et al, have designed and implement a new parallelized decision tree algorithm on a CUDA (compute unified device architecture), which is a GPGPU solution provided by NVIDIA where CPU is responsible for flow control while the GPU is responsible for computation. Andrea Dal Pozzolo [6] et al, have shown how Hellinger Distance Decision Trees can be successfully applied in unbalanced and evolving stream data by removing instance propagations between batches.…”

Section: Current Approaches In Decision Treesmentioning

confidence: 99%

An improved approach on class imbalance data using within-class minority oversampling technique

2016

IJLTET

View full text Add to dashboard Cite

CUDT: A CUDA Based Decision Tree Algorithm

Cited by 37 publications

References 16 publications

Accelerating the XGBoost algorithm using GPU computing

Accelerating the XGBoost algorithm using GPU computing

An FPGA Realization of a Random Forest with <i>k</i>-Means Clustering Using a High-Level Synthesis Design

An improved approach on class imbalance data using within-class minority oversampling technique

Contact Info

Product

Resources

About