Re: [PATCH v6 13/15] scripts/simplebench: improve view of ascii table

[Vladimir Sementsov-Ogievskiy]

25.09.2020 12:31, Max Reitz wrote:
> On 18.09.20 20:19, Vladimir Sementsov-Ogievskiy wrote:
> > Introduce dynamic float precision and use percentage to show delta.
> >
> > Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
> > ---
> >   scripts/simplebench/simplebench.py | 26 +++++++++++++++++++++++++-
> >   1 file changed, 25 insertions(+), 1 deletion(-)
> >
> > diff --git a/scripts/simplebench/simplebench.py 
> > b/scripts/simplebench/simplebench.py
> > index 716d7fe9b2..56d3a91ea2 100644
> > --- a/scripts/simplebench/simplebench.py
> > +++ b/scripts/simplebench/simplebench.py
> > @@ -79,10 +79,34 @@ def bench_one(test_func, test_env, test_case, count=5, 
> > initial_run=True):
> >       return result
> >   
> >   
> > +def format_float(x):
> > +    res = round(x)
> > +    if res >= 100:
> > +        return str(res)
> > +
> > +    res = f'{x:.1f}'
> > +    if len(res) >= 4:
> > +        return res
> > +
> > +    return f'{x:.2f}'
>
> This itches me to ask for some log() calculation.
>
> Like:
>
> %.*f' % (math.ceil(math.log10(99.95 / x)), x)

Oh yes, that's cool.

> > +def format_percent(x):
> > +    x *= 100
> > +
> > +    res = round(x)
> > +    if res >= 10:
> > +        return str(res)
> > +
> > +    return f'{x:.1f}' if res >= 1 else f'{x:.2f}'
>
> Same here.  (Also, why not append a % sign in this function?)

OK

> >   def ascii_one(result):
> >       """Return ASCII representation of bench_one() returned dict."""
> >       if 'average' in result:
> > -        s = '{:.2f} +- {:.2f}'.format(result['average'], result['delta'])
> > +        avg = result['average']
> > +        delta_pr = result['delta'] / avg
> > +        s = f'{format_float(avg)}±{format_percent(delta_pr)}%'
>
> Pre-existing, but isn’t the ± range generally assumed to be the standard
> deviation?

Hmm. Actually, why not, let's just use standard deviation. I wanted to show 
maximum deviation, not mean, to not miss some bugs in experiment (big deviation 
of one test run). Still, seems standard deviation is good enough in it.

> >           if 'n-failed' in result:
> >               s += '\n({} failed)'.format(result['n-failed'])
> >           return s


--
Best regards,
Vladimir