[Shootout-list] re: data.csv format
Bengt Kleberg
bengt.kleberg@ericsson.com
Wed, 08 Jun 2005 07:26:18 +0200
On 2005-06-07 20:51, Brent Fulgham wrote:
...deleted
> web site faster to display. Languages that do not
> make it through all values of 'n' are shown as
> 'failed'
> in the summary view, so this behavior is what we
> want.
i would like to suggest that this is not 100% correct. if we have more
''n'' and many of these ''n'' values will make slow languages time out,
then we would only have the very fastest languages in the summary view.
as i have tried to explain, i do not mind if this happens since i think
the summary is uninteresting compared to the (potential of the) graphs,
but i have been repeatedly told that the summary view is an important
part of the narrow focus and it is not to be harmed by any new ideas.
so, my plan for adding ''n'' is to make sure that the current summary
view is not changed in any significant way, while still generating more
metric values.
to do this i think it is neccessary to generate data.csv not from the
highest ''n'', but from the ''n'' that has the largest amount of usefull
run times. usefull run time in this case is:
0.01 < usefull run time < time out
note that i am also advocating making the lower bound for usefull
higher, but that is a seperate discussion. such a change would make a
significant change to the summary view, even if we only have a single
''n'' (or 3 very close ''n'').
> Anyone trying to analyze data should be looking at
> 'ndata.csv', since it has the full set of data.
i agree. my primary goal is to make it possible to have more data in the
full set.
bengt