Commit f61b1766a7f08159397f6f34ebfce943f8071e1a

Added Timings table to the article
article/article.pdf
(512 / 377)
Binary files differ
article/article.tex
(18 / 0)
  
163163\item sparklinet kaikkien tapausten tuloksista
164164\item Ajo-ajat eri vehkeillä jossain yksittäisessä tapauksessa. Esim. Criterionilla tehtynä?
165165\item Tilastolliset mittarit (4)
166
167\begin{figure}
168\centerline{\includegraphics[width=9.5cm]{../results/timings.pdf}}
169\caption{Kernel density estimate of runtimes of various PRNGs.}\label{fig:runtimes}
170\end{figure}
171
172Runtimes of various PRNGs are plotted in the Figure~\ref{fig:runtimes}. The figure shows a
173gaussian Kernel Density estimate based on a bootstrapped (cite: Davison, A.C;
174Hinkley, D.V. (1997) Bootstrap methods and their application.
175http://statwww.epfl.ch/davison/BMA/) runtime sample. The runtimes range from
176$\approx 152ns$ for simple LCG to $\approx 177ns$ for the slowest generator, the MWC$_{4096}$, yielding
177performance difference of approximately $15\%$. Considering the shear number of pseudorandom numbers
178generated in a single run, or worse, a batch of runs by DE, this is quite significant amount. In
179simple cases, such as in this article this means that using a simple PRNG the algorithm can
180calculate $10\%$ longer runs, which gives a more significant impact than that of the difference in
181the PRNG quality.
182
183
166184\end{itemize}
1671853-5 pages
168186