readme

来自「一个功能强大的神经网络分析程序」· 代码 · 共 28 行

Programs used to generate benchmarks.

performance - Program to benchmark the performance of running the libraries
quality - Program to benchmark the quality of the training algorithms
benchmark.sh - Shell file that runs a series of benchmarks
gnuplot - File containing gnuplot commands for displaying graphs

In order to run benchmarks, run:
make
./benchmarks.sh
gnuplot < gnuplot

A file named allplots.ps will display the benchmark graphs.

In order to generate benchmarks against other libraries (jneural and
lwneuralnet), these libraries need to be installed (they can be
downloaded from http://voltar-confed.org/jneural/ and
http://sourceforge.net/projects/lwneuralnet/). The Makefile need to be
altered to point to these libraries and the performance.cc and
quality.cc files need to be altered in order to use the libraries
(just uncomment a few lines in the top).

Pregenerated benchmarks are available in the benchmark.pdf file.
These benchmarks are generated on a AMD Athlon XP 1600+ machine 
(actually only 1400 MHz) with a 256 KB L2 cache and a 128 KB L1 cache.
An earlier version of the library is benchmarked and commented in the 
full report (this benchmark also include benchmarks on an iPAQ).