6:22 p.m.
Dear all,
Today's git help entry for max
does not mention matrix argument
But
set seed 13
matrix a_matrix = mnormal(10^4,10^4)
set stopwatch
ma1 = max(vec(a_matrix))
eval $stopwatch
set stopwatch
ma2 = maxc(maxr(a_matrix))
eval $stopwatch
I have 1.6487076 sec for max(vec..
and 7.9388368 sec for the second version
And the first version seems to be doomed
for deprecation
Oleh
6:34 p.m.
On Thu, 15 Mar 2018, oleg_komashko(a)ukr.net wrote:
Dear all,
Today's git help entry for max
does not mention matrix argument
But
set seed 13
matrix a_matrix = mnormal(10^4,10^4)
set stopwatch
ma1 = max(vec(a_matrix))
eval $stopwatch
set stopwatch
ma2 = maxc(maxr(a_matrix))
eval $stopwatch
I have 1.6487076 sec for max(vec..
and 7.9388368 sec for the second version
And the first version seems to be doomed
for deprecation
Not necessarily. We have in mind (well, not fully discussed yet) that
max(a_matrix) as revised would return the largest element in a_matrix
(regardless of rows and columns). It would therefore be equivalent to
but faster than max(vec(a_matrix)).
Allin
6:47 p.m.
As for personally, the last
proposition about max(matrix)->global max
is very attractive
I also think another point worth considering
is non-uniformity between series and matrix
argumented statistical functions:
e.g. sum is global sum for a matrix as expected
median is for series only
quantile is for matrices
May be, func and funr may be extended for median
and non-suffixed functions for matrices will be global as sum?
And what about quantile for series
Oleh
15 березня 2018, 20:34:35, від "Allin Cottrell" <cottrell(a)wfu.edu>:
On Thu, 15 Mar 2018, > oleg_komashko(a)ukr.net wrote:
> Dear all,
> Today's git help entry for max
> does not mention matrix argument
> But
>
> set seed 13
> matrix a_matrix = mnormal(10^4,10^4)
>
> set stopwatch
> ma1 = max(vec(a_matrix))
> eval $stopwatch
>
>
> set stopwatch
> ma2 = maxc(maxr(a_matrix))
> eval $stopwatch
>
> I have 1.6487076 sec for max(vec..
>
> and 7.9388368 sec for the second version
> And the first version seems to be doomed
> for deprecation
Not necessarily. We have in mind (well, not fully discussed yet) that
max(a_matrix) as revised would return the largest element in a_matrix
(regardless of rows and columns). It would therefore be equivalent to
but faster than max(vec(a_matrix)).
Allin
_______________________________________________
Gretl-devel mailing list
Gretl-devel@lists.wfu.eduhttp://lists.wfu.edu/mailman/listinfo/gretl-devel
6:50 p.m.
Am 15.03.2018 um 19:47 schrieb oleg_komashko(a)ukr.net:
median is for series only
quantile is for matrices
Yeah, but you have quantile(m, 0.5) which gives you the median for a
matrix as well.
May be, func and funr may be extended for median
and non-suffixed functions for matrices will be global as sum?
And what about quantile for series
For any series x you can always do {x} to get the matrix version (as you
probably know), so why is it a problem?
-sven
8:34 p.m.
so why is it a problem?
Of course, everything is computable
but why not to have more easy and
intuitive syntax for a set of
"obligatory" selection of statistical
functions?
15 березня 2018, 20:50:19, від "Sven Schreiber" <svetosch(a)gmx.net>:
Am 15.03.2018 um 19:47 schrieb > oleg_komashko(a)ukr.net:
> median is for series only
> quantile is for matrices
Yeah, but you have quantile(m, 0.5) which gives you the median for a
matrix as well.
> May be, func and funr may be extended for median
> and non-suffixed functions for matrices will be global as sum?
> And what about quantile for series
For any series x you can always do {x} to get the matrix version (as you
probably know), so why is it a problem?
-sven
_______________________________________________
Gretl-devel mailing list
Gretl-devel@lists.wfu.eduhttp://lists.wfu.edu/mailman/listinfo/gretl-devel
2787
days inactive
2787
days old
4 comments
3 participants
participants (3)
-
Allin Cottrell -
oleg_komashko＠ukr.net -
Sven Schreiber