Open vrivas opened 7 years ago
:+1:
Surprisingly (or not?) best results are achieved with latest versions of Firefox and Chromium (both of them running under Linux), followed in any case by the same navigators under Windows. "Worst" browsers seem to be IE and Edge and... Safari!!!!
You should run node too to compare.
OK... but the library is intended to allow running EA in browsers :/
But, what's the point of making the comparison? What you want to prove is that browsers are fast enough, right? So reviewer number 3 will say that browsers are slower than desktop applications, so it's better to pre-empt that objection by testing it also for node.
Results analysed by @eparrasg for MSE (the measure used by De Jhon in his initial paper)
Rejected
means there exist NO signficant differences wrt control browser.
Non rejected
means just the opposite.
Control Browser: firefox-50-1-0-macs ρ α/i Hypothesis
firefox-50-1-0-androids 0,00003 0,00714 Rejected
firefox-47-0-linuxes 0,00140 0,00833 Rejected
chromium-51-0-2704-79-linux 0,00259 0,01000 Rejected
chrome-55-0-2883-87-windows 0,00811 0,01250 Rejected
edge-14-14393 0,01992 0,01667 Non Rejected
safari-10-0-2-macs 0,71500 0,02500 Non Rejected
chrome-55-0-2883-95-macs 0,71500 0,05000 Non Rejected
We have similar analysis for every measure (13 measures) . The control browsers is always Firefox 50 in Mac, and the Rejected and non rejected appear more or less in the same browsers.
Great. Add that to the paper.
There was an error in the data sent by @eparrasg: the null hypothesis was just the inverse, and the real results are:
Control Browser: firefox-50-1-0-macs ρ α/i Hypothesis
chrome-55-0-2883-95-macs 0,00003 0,00714 Rejected
safari-10-0-2-macs 0,00140 0,00833 Rejected
edge-14-14393 0,00259 0,01000 Rejected
chrome-55-0-2883-87-windows 0,00811 0,01250 Rejected
chromium-51-0-2704-79-linux 0,01992 0,01667 Non Rejected
firefox-47-0-linux 0,71500 0,02500 Non Rejected
firefox-50-1-0-android 0,71500 0,05000 Non Rejected
I.e., the same but now Rejected
means worst than the control browser, while Non rejected
means no differences wrt control browser.
By the way, I can't imagine way chrome in Mac is so bad :/
Experiments using different navigators running individually
(Included in file: https://github.com/geneura-papers/2016-DCAI/blob/master/jsRBFNN/resultados_isafm_2017/resultados_por_navegador_ejecutado_individualmente.ods)