Documentation for BigTensor (Single)

Installation

Environment

BIGtensor (Single) runs on single machine.
The shell scripts and code packaging scripts works easily in Linux or Unix machines.
BIGtensor (Single) needs the following softwares to be installed in the system:

  • g++ version 5.4.0 or greater
  • OpenCL version 1.2 or greater
  • GPU enviroments (like a NVIDIA CUDA) for running OpenCL

Download and Install

Installation is as simple as downloading and extracting the installation file.

  1. Download the file BIGtensor_single-1.0.tar.gz.
  2. Extract the file by running the following command: tar xvzf BIGtensor_single-1.0.tar.gz.

Complie

BIGtensor (Single) Executable

  1. Use the Makefile scripts (e.g. 'make tf') to compile each executable for the operations.
  2. Execute the executable file (e.g. './bin/demo_tf') to perform tensor operation you want!

Detailed explanations about how to use this library is on USER Guide and API Doc below.

Code example

Here is a demo code that decomposes a tensor by PARAFAC. Note that you can use PARAFAC using 13 lines of code!


int main(int argc, char* argv[]) {
	char InputPath = "path/to/input/tensor", ResultPath = "path/to/output/tensor";
	int gpu_mode = 2;	// 0: cpu, 1: single-gpu, 2: multi-gpu
	int nnz_mode = 1;	// 0: fully observable, 1: partially observable
	int fact_mode = 1;	// 0: Tucker, 1: PARAFAC
	int rrank = 10;		// rank of Core tensor

	Tensor A = Tensor(InputPath);
	A.GPU_Mode_Change(gpu_mode);
	A.Observation(nnz_mode);
	A.Factorization_Mode_Change(fact_mode);
	Tensor Core;
	Tensor::tensor_factorization(A, Core, rrank, ResultPath);
}	

                    

User Guide

The User Guide provides information on the general overview, installation, and supported functions of BIGtensor.


User Guide

API Doc

The API Doc provides detailed information on classes and functions in BIGtensor.


API Doc