File	Date	Author	Commit
cuda	2012-11-06	Fabrizio	[2073d6] Second Version Multiple GPUs
README	2013-10-24	Fabrizio	[b5aa49] new files
binario.txt	2013-10-24	Fabrizio	[b5aa49] new files
conf.cu	2013-10-24	Fabrizio	[b5aa49] new files
output.txt	2013-10-24	Fabrizio	[b5aa49] new files
saida.txt	2013-10-24	Fabrizio	[b5aa49] new files
saida_1.1.0.txt	2013-10-24	Fabrizio	[b5aa49] new files
ternario.txt	2013-10-24	Fabrizio	[b5aa49] new files

Read Me

This file describe informations about mgpuinference's project.

INTRODUCTION

This software allows to infer large Gene Regulatory Netowrks (GRN). Such software use feature selection methods, 
more specifically, a parallel exhaustive search based on GPU architecture.

The cadinality of the predictor subset for each target must be equal to 2. This version allows
two or more quantization values in the input data.

CONFIGURATION PARAMETERS

There are parameters to configure the software in file conf.cu:

	-BINS - Number of BINS (states) : 2, 3, etc

	-TARGETS_PER_BLOCK - Number os target genes per block. We recommend to use 32, 64 or 128.

	-SAMPLES - Number of experimental samples (rows) of the problem.

	-COLS - Number of target genes (columns) of the GRN (number of genes, if you consider all genes as targets) - 
	      - shall be multiply of 32	(dummy genes are added).

	-N_GPUS - Number of GPUs in the computer. We recommend to use 1, 2 or 4. 

We suggest to execute this method on tty1 or disable driver graphics of LINUX. We did not test in Microsoft Windows.

On line 259 of the file conf.cu, the user has to set the corresponding quantization (number of possible values) present in the
input file.

There are two files (binario.txt and ternario.txt) in the folder of the code. The first file corresponds to the 
binary quantization, and the second to the ternary quantization. 

COMPILATION

Too compile, open the command line (console) and type this command: nvcc conf.cu -arch sm_xx, where xx is the compute 
capability of the graphic cards. We assume that NVIDIA compiler is configured in environment variables.

To configure NVIDIA compiler, please access http://askubuntu.com/questions/231503/nvcc-compiler-setup-ubuntu-12-04

RUNNING ON CONSOLE

To execute the software, write ./a.out  on command line (Linux console).

FORMAT INPUT DATA

Shall be an input file containing gene expression data samples in rows and the samples belonging to the GRN in columns. 
The input data must be quantized (discretized). Such file cannot have label in rows and columns.

The binary data samples used in the experiments were generated using the
Artificial Gene Network (AGN) simulator (https://code.google.com/p/jagn/), 
considering the Erdös-Rényi Boolean network model. Such
simulator allows to control the number of genes present in the network and the
number of samples.

MINIMUM REQUERIMENTS

NVIDIA compilier version 4.2 or higher and gcc version 4.4 or higher. We recommend to install CUDA Toolkit.
We tested this software on ubuntu 12.04 LTS.