metrics: Fix percentage difference row

[GabyCT]

Fix the percentage difference calculation for the metrics report generation.

Fixes #709

Signed-off-by: Gabriela Cervantes gabriela.cervantes.tellez@intel.com

[GabyCT]

@grahamwhaley currently, I am fixing memory-footprint.R, if you are agree with this approach, I will do the changes in the rest of the diff calculations, thanks

[grahamwhaley]

Hi @GabyCT .
I see the formula you used gives the % diff between the two numbers using the mid-point (average). Whilst that makes sense if you want to contemplate the % diff from pov of both numbers (iyswim), I think it makes more sense for us to look at the diff from the pov of the second number - that is, what is the % difference between the second number wrt the first - not the midpoint.

To show what I mean, I threw some obvious numbers into the .json files (100/80/110).
Here are the results:

From my original bad % code:

Results	noKSM	KSM	Units
x	100.0	100.0	Kb
y	110.0	80.0	Kb
diff	-9.09	25	%

From your PR:

Results	noKSM	KSM	Units
x	100.0	100.0	Kb
y	110.0	80.0	Kb
diff	9.52	22.22	%

That's not quite what I had in mind though. I did a mod (for ref: using a formula like this, dropping the average and the abs to:

if (length(resultdirs) == 2) {
        # This is a touch hard wired - but we *know* we only have two
        # datasets...
        diff=c("diff")
        diference = as.double(rstats[2,2]) - as.double(rstats[1,2])
#       average = (as.double(rstats[1,2]) + as.double(rstats[2,2])) / 2
        val = 100 * (diference/as.double(rstats[1,2]))
        diff[2] = round(val, digits=2)
        diference = as.double(rstats[2,3]) - as.double(rstats[1,3])
#       average = (as.double(rstats[1,3]) + as.double(rstats[2,3])) / 2
        val = 100 * (diference/as.double(rstats[1,3]))
        diff[3] = round(val, digits=2)
        rstats=rbind(rstats, diff)

        unts[3]="%"
}

which gets me:

Results	noKSM	KSM	Units
x	100.0	100.0	Kb
y	110.0	80.0	Kb
diff	10	-20	%

which to my eyes makes a lot more sense. wdyt?

Note: oh, also, whilst there - s/diference/difference/ ;-)

[GabyCT]

@grahamwhaley , yes I saw that formula too, I got confused if we really wanted to see the + and - signs but for me that looks better, I will work on the changes and apply them in the rest of diffs that we have in the dockerfiles thanks

[GabyCT]

@grahamwhaley changes applied thanks

[grahamwhaley]

yay, this is looking good to me, and generated good output in my test. Just one tiny spelling niggle....
s/diference/difference/ throughout :-)

[GabyCT]

@grahamwhaley changes applied

[grahamwhaley]

lgtm
thanks for fixing my original wonky math ;-)

[grahamwhaley]

/test

[devimc]

woow! good catch @GabyCT
lgtm