Find the median value of an array.

Hello,

Given an array with 10,000 real values, I want to find the median. And
I want to do this about 100,000 times. My current plan is to implement
Quicksort or Merge Sort in Fortran and pick the middle point of the
array to get the median. I was just wondering Fortran already has a
way to do this. I am sure I am not the first person in history to want
the median of a group of values in Fortran. But I cannot find an
intrinsic function called "median".

Any thoughts?

Thanks.
Daniel.
0
Daniel
11/11/2010 4:49:36 PM
comp.lang.fortran 11606 articles. 1 followers. Post Follow

22 Replies
3797 Views

Similar Articles

[PageSpeed] 15

"Daniel Carrera" <dcarrera@gmail.com> wrote in message 
news:d510fdac-68cd-4637-811d-985d58ec0541@40g2000vbn.googlegroups.com...

> Given an array with 10,000 real values, I want to find the median. And
> I want to do this about 100,000 times. My current plan is to implement
> Quicksort or Merge Sort in Fortran and pick the middle point of the
> array to get the median. I was just wondering Fortran already has a
> way to do this. I am sure I am not the first person in history to want
> the median of a group of values in Fortran. But I cannot find an
> intrinsic function called "median".

I wrote up some stuff a while back:

http://home.comcast.net/~kmbtib/Fortran_stuff/order_stat.i90
http://home.comcast.net/~kmbtib/Fortran_stuff/order_stat_test.f90

Can't recall whether it's exactly what you need, but it's faster
than sorting, anyhow.  Write back if you encounter problems.

-- 
write(*,*) transfer((/17.392111325966148d0,6.5794487871554595D-85, &
6.0134700243160014d-154/),(/'x'/)); end


0
James
11/11/2010 4:56:53 PM
In article <ibh78q$l28$1@news.eternal-september.org>,
James Van Buskirk <not_valid@comcast.net> wrote:
>"Daniel Carrera" <dcarrera@gmail.com> wrote in message 
>news:d510fdac-68cd-4637-811d-985d58ec0541@40g2000vbn.googlegroups.com...
>
>> Given an array with 10,000 real values, I want to find the median. And
>> I want to do this about 100,000 times. My current plan is to implement
>> Quicksort or Merge Sort in Fortran and pick the middle point of the
>> array to get the median. I was just wondering Fortran already has a
>> way to do this. I am sure I am not the first person in history to want
>> the median of a group of values in Fortran. But I cannot find an
>> intrinsic function called "median".
>
>I wrote up some stuff a while back:
>
>http://home.comcast.net/~kmbtib/Fortran_stuff/order_stat.i90
>http://home.comcast.net/~kmbtib/Fortran_stuff/order_stat_test.f90
>
>Can't recall whether it's exactly what you need, but it's faster
>than sorting, anyhow.  Write back if you encounter problems.

There is a much faster algorithm, incidentally.

1) Estimate a range around its value using a small, fixed sample.

2) Pass through the array, copying the values within the range,
and counting those to the left and right.

3) Repeat on the subset if the counts were good, and using the
fact that you now know which side it was if not.

4) Repeat until the sample is small enough to sort.

You can use a limited size array for the subset, incidentally,
and do a modification of the above if you run out of space.

The theory of this is standard statistics.

I have the code somewhere, but wouldn't be able to find it,
because I can't even rememember which decade I did it in and
scanning through all my archives (in half a dozen formats) is
too time-consuming.  Sorry.


Regards,
Nick Maclaren.
0
nmm1
11/11/2010 5:40:48 PM
nmm1@cam.ac.uk wrote:
(snip)
>>"Daniel Carrera" <dcarrera@gmail.com> wrote in message 
>>news:d510fdac-68cd-4637-811d-985d58ec0541@40g2000vbn.googlegroups.com...

>>> Given an array with 10,000 real values, I want to find the median. 
(snip)

> There is a much faster algorithm, incidentally.

Is that the one in Knuth, "The Art of Computer Programming", my
guess is volume 3?   It doesn't sound familiar, but I could have
forgotten how it went.
 
> 1) Estimate a range around its value using a small, fixed sample.
 
> 2) Pass through the array, copying the values within the range,
> and counting those to the left and right.
 
> 3) Repeat on the subset if the counts were good, and using the
> fact that you now know which side it was if not.
 
> 4) Repeat until the sample is small enough to sort.
 
> You can use a limited size array for the subset, incidentally,
> and do a modification of the above if you run out of space.
 
> The theory of this is standard statistics.
 
> I have the code somewhere, but wouldn't be able to find it,
> because I can't even rememember which decade I did it in and
> scanning through all my archives (in half a dozen formats) is
> too time-consuming.  Sorry.

-- glen
0
glen
11/11/2010 9:17:12 PM
In article <ibhmgo$b1q$1@news.eternal-september.org>,
glen herrmannsfeldt  <gah@ugcs.caltech.edu> wrote:
>
>>>> Given an array with 10,000 real values, I want to find the median. 
>
>> There is a much faster algorithm, incidentally.
>
>Is that the one in Knuth, "The Art of Computer Programming", my
>guess is volume 3?   It doesn't sound familiar, but I could have
>forgotten how it went.

No.  Knuth is not strong on statistical algorithms.  It includes
the one that James Van Buskirk posted.


Regards,
Nick Maclaren.
0
nmm1
11/11/2010 9:28:11 PM
nmm1@cam.ac.uk wrote:
> In article <ibhmgo$b1q$1@news.eternal-september.org>,
> glen herrmannsfeldt  <gah@ugcs.caltech.edu> wrote:

>>>>> Given an array with 10,000 real values, I want to find the median. 

>>> There is a much faster algorithm, incidentally.

>>Is that the one in Knuth, "The Art of Computer Programming", my
>>guess is volume 3?   It doesn't sound familiar, but I could have
>>forgotten how it went.
 
> No.  Knuth is not strong on statistical algorithms.  

The one in Knuth is supposed to be O(N), which doesn't mean
that it is fastest for any, or even most, sets of input data.

Some of this seems to be in the wikipedia page Selection_algorithm.

The wikipedia page for Median_graph has a reference to Knuth's
volume 4, I believe available on the web, but not in print.

> It includes the one that James Van Buskirk posted.

I don't always read posted links.

-- glen

0
glen
11/11/2010 9:43:13 PM
In article <ibho1g$b1q$3@news.eternal-september.org>,
glen herrmannsfeldt  <gah@ugcs.caltech.edu> wrote:
>
>>>>>> Given an array with 10,000 real values, I want to find the median. 
>
>>>> There is a much faster algorithm, incidentally.
>
>>>Is that the one in Knuth, "The Art of Computer Programming", my
>>>guess is volume 3?   It doesn't sound familiar, but I could have
>>>forgotten how it went.
> 
>> No.  Knuth is not strong on statistical algorithms.  
>
>The one in Knuth is supposed to be O(N), which doesn't mean
>that it is fastest for any, or even most, sets of input data.

The one I posted is also O(N) with probability one, but with a
much lower factor.  It is very common that replacing a pure
mathematical worst case with a probability zero one produces
a much faster algorithm, that is every bit as reliable in real
use.


Regards,
Nick Maclaren.
0
nmm1
11/11/2010 9:47:00 PM
Thanks for the help.

Now I know that I need to write my own function. Thanks for the
suggestions for algorithms. I'll try to pick something that I
understand, but is efficient. I have an idea for an algorithm based on
the quicksort / mergesort idea. Probably not as efficient as James'
code, but it should be good enough, and it'll be something I
understand.

Thanks for the help.

Cheers,
Daniel.


0
Daniel
11/12/2010 9:52:08 AM
On 12.11.2010. 10:52, Daniel Carrera wrote:
> Thanks for the help.
>
> Now I know that I need to write my own function. Thanks for the
> suggestions for algorithms. I'll try to pick something that I
> understand, but is efficient. I have an idea for an algorithm based on
> the quicksort / mergesort idea. Probably not as efficient as James'
> code, but it should be good enough, and it'll be something I
> understand.
>
> Thanks for the help.

Take a look at Michel Olagnon's (who used to frequent clf):

http://www.fortran-2000.com/rank/

In particular, INDMED and VALMED are based on Knuth. I used some
of other Michel's routines, but not these ones.

--
Jugoslav
www.xeffort.com
0
Jugoslav
11/12/2010 3:02:51 PM
On Nov 12, 4:02=A0pm, Jugoslav Dujic <jdu...@yahoo.com> wrote:
> Take a look at Michel Olagnon's (who used to frequent clf):
>
> http://www.fortran-2000.com/rank/
>
> In particular, INDMED and VALMED are based on Knuth. I used some
> of other Michel's routines, but not these ones.

Thanks. I'll take a look at VALMED. In fact, I have a question
already. VALMED starts with:

Recursive Function D_valmed (XDONT) Result (res_med)

I thought that function definitions always needed a return type like
"real function foo()". When I compile my own code with gfortran it
always complains if I don't include a return type, but Michel's module
compiles just fine the way it is. What's Michel doing that allows him
to get away with a function without a return type?

Anyway, I'll try to understand how Michel's code works. I did come up
with a nice, simple algorithm to get the median in O(n log n), and is
fast enough for my purposes (will add ~2.7 mins to a program that runs
8 hours). But I'll probably learn something interesting by looking at
Michel's code.

For reference, my code is below. It sorts an array of real numbers and
is inspired quicksort / mergesort:


real function qmedian(list)
    real, dimension(:), intent(in) :: list
    integer :: n
    real :: dummy

    n =3D size(list)
    dummy =3D rand(1)
    if (mod(n,2) =3D=3D 0) then
        qmedian =3D qmedian_helper(list, n/2)
    else
        qmedian =3D qmedian_helper(list, n/2 + 1)
    end if
end function

real recursive function qmedian_helper(list, q) result(res)
    ! Parameters
    real, dimension(:), intent(in) :: list
    integer, intent(in) :: q

    ! Internal variables
    real, allocatable, dimension(:) :: left, right
    real :: pivot, lmax, lmin
    integer :: n, j

    ! Algorithm
    n =3D size(list)
    call bounds(list, lmin, lmax)
    if (q =3D=3D 1) then
        res =3D lmin
    else if (q =3D=3D n) then
        res =3D lmax
    else
        !!! Split 'list' into two parts and recurse.

        ! Pick a random element from the list.
        pivot =3D list(int(rand(0)*n)+1)

        j =3D count(list <=3D pivot)

        ! The previous conditions should remove corner cases.
        allocate(left(j))
        allocate(right(n-j))

        left =3D pack(list, list <=3D pivot)
        right=3D pack(list, list  > pivot)

        if (j >=3D q) then
            ! Left side contains kth element.
            res =3D qmedian_helper(left, q)
        else
            ! Right side contains kth element.
            res =3D qmedian_helper(right, q-j)
        end if
    end if
end function

subroutine bounds(list, lmin, lmax)
    real, dimension(:), intent(in) :: list
    real, intent(out) :: lmin, lmax
    integer :: n, i

    n =3D size(list)
    lmin =3D list(1)
    lmax =3D list(1)
    do i =3D 2,n
        if (list(i) < lmin) lmin =3D list(i)
        if (list(i) > lmax) lmax =3D list(i)
    end do
end subroutine


Cheers,
Daniel.
0
Daniel
11/12/2010 7:47:41 PM
Daniel Carrera wrote:
> On Nov 12, 4:02 pm, Jugoslav Dujic <jdu...@yahoo.com> wrote:
>> Take a look at Michel Olagnon's (who used to frequent clf):
>>
>> http://www.fortran-2000.com/rank/
>>
>> In particular, INDMED and VALMED are based on Knuth. I used some
>> of other Michel's routines, but not these ones.
> 
> Thanks. I'll take a look at VALMED. In fact, I have a question
> already. VALMED starts with:
> 
> Recursive Function D_valmed (XDONT) Result (res_med)
> 
> I thought that function definitions always needed a return type like
> "real function foo()". 

I suspect the answer is made clear if you look at the first point under
"programming style" in the above link,

Ian
0
Ian
11/12/2010 7:57:41 PM
On Nov 12, 11:47=A0am, Daniel Carrera <dcarr...@gmail.com> wrote:
> On Nov 12, 4:02=A0pm, Jugoslav Dujic <jdu...@yahoo.com> wrote:
>
> > Take a look at Michel Olagnon's (who used to frequent clf):
>
> >http://www.fortran-2000.com/rank/
>
> > In particular, INDMED and VALMED are based on Knuth. I used some
> > of other Michel's routines, but not these ones.
>
> Thanks. I'll take a look at VALMED. In fact, I have a question
> already. VALMED starts with:
>
> Recursive Function D_valmed (XDONT) Result (res_med)
>
> I thought that function definitions always needed a return type like
> "real function foo()". When I compile my own code with gfortran it
> always complains if I don't include a return type, but Michel's module
> compiles just fine the way it is. What's Michel doing that allows him
> to get away with a function without a return type?

Removing the large comment from Michel's code, I finds
the first 3 lines:

Recursive Function D_valmed (XDONT) Result (res_med)
      Real (kind=3Dkdp), Dimension (:), Intent (In) :: XDONT
      Real (kind=3Dkdp) :: res_med

The return type is declared in line 3.

--
steve
0
steve
11/12/2010 7:59:39 PM
Daniel Carrera <dcarrera@gmail.com> wrote:

> Recursive Function D_valmed (XDONT) Result (res_med)
> 
> I thought that function definitions always needed a return type like
> "real function foo()". When I compile my own code with gfortran it
> always complains if I don't include a return type, but Michel's module
> compiles just fine the way it is. What's Michel doing that allows him
> to get away with a function without a return type?

He presumably declares the type later in the code in a separate
statement. Not only is that allowed, it is what I prefer and recommend.
There can be problems with declaring the type directly on the function
statement; it works in most cases, but not all of them. Declaring the
type in a separate type declaration statement works in *ALL* cases. I
prefer to do things that work all the time so that I don't need to worry
about checking when I have crossed the boundary between the cases that
work and the ones that don't.

Also note that when you declare the result variable to have a separate
name, it is the result variable whose type gets declared - not the
function.

As an aside, no, you don't have to explicitly declare the type. If you
don't declare the type, implicit typing applies, just as with any other
variable. I disrecommend the use of implicit typing, and if you use
impkicit none, the compiler will bitch about it, but it is allowed. If
your compiler complains when you omit the type, then probably you are
using implicit none (which I do recommend). Either that or you are doing
something that is incompatible with the implicit type. Be aware that
implicit typing does *NOT* mean that an entity has no type; it just
means that the type is determined by the implicit typing rules instead
of by an explicit type declaration. I wouldn't mention this except that
I wonder why you would ever be trying to write code that doesn't declare
the return type and also isn't deliberately using implicit typing; the
only reason I can think of (other than just an accident) is that one
might think that failing to explicitly declare the type somehow left it
generic.

-- 
Richard Maine                    | Good judgment comes from experience;
email: last name at domain . net | experience comes from bad judgment.
domain: summertriangle           |  -- Mark Twain
0
nospam
11/12/2010 8:05:11 PM
On Nov 12, 8:59=A0pm, steve <kar...@comcast.net> wrote:
> On Nov 12, 11:47=A0am, Daniel Carrera <dcarr...@gmail.com> wrote:
>
>
>
>
>
> > On Nov 12, 4:02=A0pm, Jugoslav Dujic <jdu...@yahoo.com> wrote:
>
> > > Take a look at Michel Olagnon's (who used to frequent clf):
>
> > >http://www.fortran-2000.com/rank/
>
> > > In particular, INDMED and VALMED are based on Knuth. I used some
> > > of other Michel's routines, but not these ones.
>
> > Thanks. I'll take a look at VALMED. In fact, I have a question
> > already. VALMED starts with:
>
> > Recursive Function D_valmed (XDONT) Result (res_med)
>
> > I thought that function definitions always needed a return type like
> > "real function foo()". When I compile my own code with gfortran it
> > always complains if I don't include a return type, but Michel's module
> > compiles just fine the way it is. What's Michel doing that allows him
> > to get away with a function without a return type?
>
> Removing the large comment from Michel's code, I finds
> the first 3 lines:
>
> Recursive Function D_valmed (XDONT) Result (res_med)
> =A0 =A0 =A0 Real (kind=3Dkdp), Dimension (:), Intent (In) :: XDONT
> =A0 =A0 =A0 Real (kind=3Dkdp) :: res_med
>
> The return type is declared in line 3.

Interesting. I didn't know you could do that. I was beginning to think
that he was relying on implicit types in some way. I notice that he
doesn't use "IMPLICIT NONE".

Thanks. I learned something new today.

Daniel.
0
Daniel
11/12/2010 8:16:08 PM
On Nov 12, 9:05=A0pm, nos...@see.signature (Richard Maine) wrote:
> He presumably declares the type later in the code in a separate
> statement. Not only is that allowed, it is what I prefer and recommend.

Yes, he does. I didn't realize that was possible until Steve pointed
it out.


> There can be problems with declaring the type directly on the function
> statement; it works in most cases, but not all of them. Declaring the
> type in a separate type declaration statement works in *ALL* cases. I
> prefer to do things that work all the time so that I don't need to worry
> about checking when I have crossed the boundary between the cases that
> work and the ones that don't.

Sounds reasonable. I'm surprised that declaring the type on the
function statement can fail. Can you give me an example? Anyway, I
certainly want to use methods that work all the time.


> If your compiler complains when you omit the type, then probably you are
> using implicit none (which I do recommend). Either that or you are doing
> something that is incompatible with the implicit type.

Yes, I always use implicit none. I only have a few weeks knowledge of
Fortran 95, but I learned to use implicit none from the very
beginning. Implicit typing scares me. I'm surprised that Michael
doesn't use implicit none.


> I wouldn't mention this except that
> I wonder why you would ever be trying to write code that doesn't declare
> the return type and also isn't deliberately using implicit typing; the
> only reason I can think of (other than just an accident) is that one
> might think that failing to explicitly declare the type somehow left it
> generic.


It was just an accident. I just forgot to enter the type once and saw
the compiler error. I just "understood" that every function needs a
return type, and I was surprised when I didn't see one in Michael's
code. I never thought of implicit typing (a concept that still seems
very odd to me).

Cheers,
Daniel.
0
Daniel
11/12/2010 8:30:35 PM
Daniel Carrera <dcarrera@gmail.com> wrote:
> On Nov 12, 9:05�pm, nos...@see.signature (Richard Maine) wrote:
>> He presumably declares the type later in the code in a separate
>> statement. Not only is that allowed, it is what I prefer and recommend.
 
> Yes, he does. I didn't realize that was possible until Steve pointed
> it out.
 
>> There can be problems with declaring the type directly on the function
>> statement; it works in most cases, but not all of them. Declaring the
>> type in a separate type declaration statement works in *ALL* cases. I
>> prefer to do things that work all the time so that I don't need to worry
>> about checking when I have crossed the boundary between the cases that
>> work and the ones that don't.
 
> Sounds reasonable. I'm surprised that declaring the type on the
> function statement can fail. Can you give me an example? Anyway, I
> certainly want to use methods that work all the time.

The separate return variable is required for direct recursion.
(When a function calls itself directly.)  Inside a function without
a separate return variable, either declared on the FUNCTION statement
or in a separate statement, the function name is a variable, and
not a function reference.

-- glen
0
glen
11/12/2010 8:43:13 PM
Daniel Carrera <dcarrera@gmail.com> wrote:

> On Nov 12, 9:05 pm, nos...@see.signature (Richard Maine) wrote:

> > There can be problems with declaring the type directly on the function
> > statement; it works in most cases, but not all of them. Declaring the
> > type in a separate type declaration statement works in *ALL* cases. I
> > prefer to do things that work all the time so that I don't need to worry
> > about checking when I have crossed the boundary between the cases that
> > work and the ones that don't.
> 
> Sounds reasonable. I'm surprised that declaring the type on the
> function statement can fail. Can you give me an example? Anyway, I
> certainly want to use methods that work all the time.

Looking at the code in question, I see that it isn't actualy too far
from the main case that I think fails. Basically, the problems come from
ordering requirements. The function statement is always the first
statement in the function; there are no exceptions to that.

There are some things that you can't use until after the statement that
defines them. If you want to use any of those things on the function
statement, there can be a problem in that the function statement is not
likely to be after the statement where they are defined.

A derived type is one of those things. Normally you can't use a derived
type until after its definition. I think (not worth checking, but I
think it is right) that a special-case exception was made for using a
derived type in a function statement just in order to allow it. I don't
tend to like such special cases and I'd have probably voted against
making one for this (don't recall whether I ever got a chance to make
such a vote - might have). Special cases tend to grow as one discovers
that you didn't catch quite everything. Indeed, this special case didn't
catch everything.

I think the missing case is type parameters. One often (and preferably)
specifies type parameter values using named constants. I do no think
there is a simillar exception to allow using a named constant in a
function statement before its definition. So if you try to do something
like

  real(my_kind) function f(x)
    use module_with_a_definition_of_my_kind
    ...

I think you are out of luck. It is possible that I've gotten confused on
the details and that this particular case works. But if it isn't exactly
this one, it is something simillar. One of the advatanges of doing the
type declararation in a separate statement is that I don't actually have
to keep track of the fine points about exactly which cases work.

Posted excerpts of the posted code do show that it uses a named constant
for a kind type parameter, so it is at least close to this case. I don't
see a USE in the posted extract though, so perhaps the kind type
parameter might be accessed via host association. That would be a case
that was ok.... well as long as it wasn't an interface body.

-- 
Richard Maine                    | Good judgment comes from experience;
email: last name at domain . net | experience comes from bad judgment.
domain: summertriangle           |  -- Mark Twain
0
nospam
11/12/2010 9:21:31 PM
On Nov 12, 10:21=A0pm, nos...@see.signature (Richard Maine) wrote:
> A derived type is one of those things. Normally you can't use a derived
> type until after its definition...
>
> I think the missing case is type parameters. One often (and preferably)
> specifies type parameter values using named constants. I do no think
> there is a simillar exception to allow using a named constant in a
> function statement before its definition. So if you try to do something
> like
>
> =A0 real(my_kind) function f(x)
> =A0 =A0 use module_with_a_definition_of_my_kind
> =A0 =A0 ...
>
> I think you are out of luck.


I just found another example. Suppose you want a function that returns
an array:

    function myfunc() result(res)
        real, dimension(3) :: res
        res =3D (/ 2.3, 2.4, 2.5 /)
    end function

This compiles an runs correctly. But when I tried move the type
declaration to the function definition, the compilation failed.

Cheers,
Daniel.
0
Daniel
11/12/2010 10:38:25 PM
Daniel Carrera <dcarrera@gmail.com> wrote:

> I just found another example. Suppose you want a function that returns
> an array:
> 
>     function myfunc() result(res)
>         real, dimension(3) :: res
>         res = (/ 2.3, 2.4, 2.5 /)
>     end function
> 
> This compiles an runs correctly. But when I tried move the type
> declaration to the function definition, the compilation failed.

Good point, thought I'll make a minor terminology correction. You
actually can still put the type into the function statement. You just
can't put the dimension attribute there. The dimension is not part of
the type even though you normally put it in the type declaration
statement; it is an attribute.

You can do (and I just checked to make sure that I hadn't missed
something - worked fine).

    real function myfunc() result(res)
        dimension :: res(3)
        res = (/ 2.3, 2.4, 2.5 /)
    end function

I do not recommend that at all. I personally find it abominable. But it
is standard conforming.

Now that you mention attributes, that reminds me that things like the
pointer attribute have simillar issues in not being specifiable on the
function statement. But I don't even want to mention that it is a
possibility for a function to return a pointer because that might prompt
someone to write such functions, which I advise against. Pretend I
didn't say anything. :-)

Functions that return allocatables are a much better example in that I
think they can be good ideas.

-- 
Richard Maine                    | Good judgment comes from experience;
email: last name at domain . net | experience comes from bad judgment.
domain: summertriangle           |  -- Mark Twain
0
nospam
11/13/2010 12:54:35 AM
Hi, I didn't read everything above.
But a search did not find the term "partial".
IMHO, quantiles, median belongs to it, are fastest found with "partial
sorting".
"Partial sorting" means that the data are not fully sorted, it is only
sorted so far that all smaller values are before the median and all
larger
values are behind, but they are not exactly sorted.
There are good algorithms available if you just look for
"partial sorting".
Wolfgang

On Nov 12, 7:54=A0pm, nos...@see.signature (Richard Maine) wrote:
> Daniel Carrera <dcarr...@gmail.com> wrote:
> > I just found another example. Suppose you want a function that returns
> > an array:
>
> > =A0 =A0 function myfunc() result(res)
> > =A0 =A0 =A0 =A0 real, dimension(3) :: res
> > =A0 =A0 =A0 =A0 res =3D (/ 2.3, 2.4, 2.5 /)
> > =A0 =A0 end function
>
> > This compiles an runs correctly. But when I tried move the type
> > declaration to the function definition, the compilation failed.
>
> Good point, thought I'll make a minor terminology correction. You
> actually can still put the type into the function statement. You just
> can't put the dimension attribute there. The dimension is not part of
> the type even though you normally put it in the type declaration
> statement; it is an attribute.
>
> You can do (and I just checked to make sure that I hadn't missed
> something - worked fine).
>
> =A0 =A0 real function myfunc() result(res)
> =A0 =A0 =A0 =A0 dimension :: res(3)
> =A0 =A0 =A0 =A0 res =3D (/ 2.3, 2.4, 2.5 /)
> =A0 =A0 end function
>
> I do not recommend that at all. I personally find it abominable. But it
> is standard conforming.
>
> Now that you mention attributes, that reminds me that things like the
> pointer attribute have simillar issues in not being specifiable on the
> function statement. But I don't even want to mention that it is a
> possibility for a function to return a pointer because that might prompt
> someone to write such functions, which I advise against. Pretend I
> didn't say anything. :-)
>
> Functions that return allocatables are a much better example in that I
> think they can be good ideas.
>
> --
> Richard Maine =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0| Good judgment come=
s from experience;
> email: last name at domain . net | experience comes from bad judgment.
> domain: summertriangle =A0 =A0 =A0 =A0 =A0 | =A0-- Mark Twain

0
wolfgang
11/13/2010 9:15:56 AM
On 11/12/2010 09:30 PM, Daniel Carrera wrote:
> On Nov 12, 9:05 pm, nos...@see.signature (Richard Maine) wrote:
>> He presumably declares the type later in the code in a separate
>> statement. Not only is that allowed, it is what I prefer and recommend.
>
> Yes, he does. I didn't realize that was possible until Steve pointed
> it out.
>
>
>> There can be problems with declaring the type directly on the function
>> statement; it works in most cases, but not all of them. Declaring the
>> type in a separate type declaration statement works in *ALL* cases. I
>> prefer to do things that work all the time so that I don't need to worry
>> about checking when I have crossed the boundary between the cases that
>> work and the ones that don't.
>
> Sounds reasonable. I'm surprised that declaring the type on the
> function statement can fail. Can you give me an example? Anyway, I
> certainly want to use methods that work all the time.
>
>
>> If your compiler complains when you omit the type, then probably you are
>> using implicit none (which I do recommend). Either that or you are doing
>> something that is incompatible with the implicit type.
>
> Yes, I always use implicit none. I only have a few weeks knowledge of
> Fortran 95, but I learned to use implicit none from the very
> beginning. Implicit typing scares me. I'm surprised that Michael
> doesn't use implicit none.
>

Well, I write a lot of quick and dirty programs, but when it is code 
that I want to keep or publish, I do check for undeclared variables.
In the case of the sorting and ranking routines, I compiled them with an 
F compiler, that enforces implicit none, so everything is declared.

Regarding the way declarations are arranged, some features come from the 
fact that I used a pre-processor to build the various type versions of 
the routines, by lack of a suitable mechanism in the language (at least 
at that time, I don't know if something has appeared now).

The code looked at first like:
Module m_mrgrnk
Integer, Parameter :: kdp = selected_real_kind(15)
public :: mrgrnk
private :: kdp
private :: R_mrgrnk, I_mrgrnk, D_mrgrnk
interface mrgrnk
   module procedure D_mrgrnk, R_mrgrnk, I_mrgrnk
end interface mrgrnk
contains

Subroutine D_mrgrnk (XDONT, IRNGT)
$define MYTYPE "Real (kind=kdp)"
$include "mrgrnk.inc"
End Subroutine D_mrgrnk

Subroutine R_mrgrnk (XDONT, IRNGT)
$define MYTYPE "Real"
$include "mrgrnk.inc"
End Subroutine R_mrgrnk

Subroutine I_mrgrnk (XDONT, IRNGT)
$define MYTYPE "Integer"
$include "mrgrnk.inc"
End Subroutine I_mrgrnk
End Module m_mrgrnk



0
Michel
11/15/2010 8:20:52 AM
On 11/12/2010 04:02 PM, Jugoslav Dujic wrote:
>
> On 12.11.2010. 10:52, Daniel Carrera wrote:
>> Thanks for the help.
>>
>> Now I know that I need to write my own function. Thanks for the
>> suggestions for algorithms. I'll try to pick something that I
>> understand, but is efficient. I have an idea for an algorithm based on
>> the quicksort / mergesort idea. Probably not as efficient as James'
>> code, but it should be good enough, and it'll be something I
>> understand.
>>
>> Thanks for the help.
>
> Take a look at Michel Olagnon's (who used to frequent clf):
>
> http://www.fortran-2000.com/rank/
>
> In particular, INDMED and VALMED are based on Knuth. I used some
> of other Michel's routines, but not these ones.
>

I programmed Knuth's algorithm for the sake of sport, but in practice it 
is so complex that you would have to go to huge values of N to see some 
advantage over the simple pivoting strategy.
0
Michel
11/15/2010 8:27:08 AM
In article <8kc96tFfv4U1@mid.individual.net>,
Michel Olagnon  <molagnon@ifremer-a-oter.fr> wrote:
>
>I programmed Knuth's algorithm for the sake of sport, but in practice it 
>is so complex that you would have to go to huge values of N to see some 
>advantage over the simple pivoting strategy.

That is true, but the statistical one is enough simpler and faster
that you start to see gains at quite low values - if I recall, it's
about 20-25.


Regards,
Nick Maclaren.
0
nmm1
11/15/2010 8:49:44 AM
Reply:

Similar Artilces:

Rank the arrary, then Find the duplicated values in the array and replace rank value with the mean of the ranks for the duplicated elements in Array.
for example: Array Rank replaced_rank ------ ------- -------- 1 1 (1+2)/2 1 2 (1+2)/2 2 3 (3+4+5)/3 2 4 (3+4+5)/3 2 5 (3+4+5)/3 3 6 6 the 'replaced_rank' array is what I needed. how to get it? "yu " <mothee.yue@gmail.com> wrote in message <jluclt$r47$1@newscl01ah.mathworks.com>... > for example: > > Array Rank replaced_rank > ------ ------- -------- > 1 1 (1+2)/2 > 1 ...

finding values in one array and using those locations to change values in one channel of a separate 3 channel array
So I have a labeled array (array1) with connected regions of interest (background is all zeros, connected regions are all 1's for the first region, all 2's for the second, 3's for the 3rd etc.) I also have a vector (vector1) of the region labels I find important (ex. 1,6,9). I want to find the locations of these values in the labeled array and then change values in one channel of a separate 3 channel array at the same locations (want to color certain parts of an image green based regions of interest found in another image). I can use the below code to change all channels, but do...

Find a closest value of array and test which of the values gives the highest function value as below...
Hello all, i have one problem, if someone helps me, i will really appreciate. So please, HELP ME ! I have one array, let's say it is is like this: A=[3, 5, 4, 6.2, 7.1, 7.8, 9.3, 10, 11, 13] I have to find the closest value of 7 (<=7). And then, i have to search in this vector for all values that are smaller than the closest value that we found (from the value that we found in descending order one by one search till the first value of array), and to see that which of these values gives as maximum value of function f(i), the function is like below: f(i) = 5/i+i/6+7/i So, i have...

finding a set of values in another set array of values
hi i'm trying to find how to find a set of values in another array set of values... lets say i have: pf is an array of values which is structured 1x46 pf(1)=[61,62,63,64,65] pf(2)=[62,63,64,65,61] pf(3)=[63,64,65,61,62] .... pf(46)=[...] and so on... what i need to find is: among these sets of pf from index 1 to 46, i just want to extract indexes of those with values [61,62,63,64,65] thnx. nora "Nora Ramli" <noraznita86@hotmail.com> wrote in message <h6b1cp$774$1@fred.mathworks.com>... > hi > > i'm trying to find how to find a set of values in ...

Finding the two closest array values to some arbitrary value
I would like to identify the row indices of two minimum values in a cell array, containing data Zi. Searching for a single value is simple (see below) but how to I search for the minimum and the closest value to that minimum? minimum value search: c=find(min(Z)==Z); %find indices "Daniel " <elefant2_2000@yahoo.de> wrote in message <hdud6u$dpn$1@fred.mathworks.com>... > I would like to identify the row indices of two minimum values in a cell array, containing data Zi. Searching for a single value is simple (see below) but how to I search for the minim...

Finding lowest numeric value in an array with undef values
Suppose I have an array: my @array = (1, 2, undef, 3); I want to find the smallest numeric value (ignoring undef values). My first inclination is to use the min() method of List::Util, but the undef value causes problems (the method throws a warning and then returns the undef). I could do this: use List::Util qw{ min }; print min ( grep /\d+/, @array ); which returns the expected value (1), but strikes me as kludgy (and still throws a warning). Any better suggestions? Thanks! -- The best way to get a good answer is to ask a good question. David Filmer (http://DavidFilmer.com...

Given an array a[N] with value 1...N, find a repeted value
Anyone has the solution to the question "Given an array a[N] with value 1...N, find a repeted value" in C code? Thanks a lot <cai_rongxi@hotmail.com> wrote in message news:1103992195.221718.104740@c13g2000cwb.googlegroups.com... > Anyone has the solution to the question "Given an array a[N] with value > 1...N, find a repeted value" in C code? > > Thanks a lot My solution is to ask your class teacher. cai_rongxi@hotmail.com wrote: > Anyone has the solution to the question "Given an array a[N] with value > 1...N, find a repeted value" ...

Replacing some # of values of an array with a different array of same # of values
Hi all, I hope my title makes sense. I have a feeling that what I am trying to do is not necessarily difficult, but I'm just not sure how to approach this problem without doing something extremely inefficient. I have an array of values, e.g. : temp = [1 2 3 50 1 2 3 60 1 2 3 50 1 2 3]; I want to find all instances of '50' and '60' and replace them with alternative values. altern = [80 70 80]; The number of values that I identify in temp will always match the number of alternative values that I want to insert. I was thinking that I would use find() to lo...

Finding values (text,numbers, regexps) in objects (arrays or key/value pairs)
I have an array of objects and I want to retrieve specific elements that match certain criteria, i.e. find value A in an object B The problem is that I cannot know what is A (text, object, reg exp, array) and what is B (text, object, array). Also, I must assume that B may be an array or a key/value object and that A may be anywhere (including 'children' objects). This is usually the case when I have a JSON response from an external source and want to find something in there but without hard-coding any specific methods or search cases. I think the best way to achieve this is to have...

Find values in an array
Hi all, Have been using this code for years to add unique values to an array. Is there a better way - faster? For z = 1 To cntTbl If tmpTblSpc = TblSpc(z) Then TblFnd = True Exit For End If DoEvents Next z If TblFnd = False Then cntTbl = cntTbl + 1 TblSpc(cntTbl) = tmpTblSpc TblFnd = False txtLog = txtLog & tmpTblSpc & vbCrLf End If Thanks Dan... On Tue, 6 Apr 2004 14:47:39 -0400, "Dan E." <formulazss@twcny.rr.com> ...

Find Values in an array
Hi there, I want to find the indices of a specific value say 78.0577 in an array called ABC. I thought the code was: >ind=find(ABC==78.0577) however the result I get given back is: ind = Empty matrix: 0-by-1 What am I doing wrong? Many thanks On Oct 21, 10:00=A0am, "Pete Davis" <peted.da...@gmail.com> wrote: > Hi there, > > I want to find the indices of a specific value say 78.0577 in an array ca= lled ABC. I thought the code was: > > >ind=3Dfind(ABC=3D=3D78.0577) > > however the result I get given back is: > ind =3D > > =A0 =A0Empty...

Finding values in an array
I have an array/matrix which is 10x365, and I need to find a way to search for the values under a certain value, eg less than 0, then tell me how many values in the are less than zero. Thanks for any help. On 6/14/2013 12:46 PM, Brian wrote: > I have an array/matrix which is 10x365, and I need to find a way >to search for the values under a certain value, eg less than 0, then >tell me how many values in the are less than zero. > > Thanks for any help. > may be EDU>> A=rand(5); EDU>> length(find(A<0.5==1)) %how many less than 0.5? 1...

Find an array in an array
Hi In excel sheet, I have the column like this: 0 0 0.001(error data) 0 0 10 20 30 50 20 5 0 0 0.002(error data) 0 0 0 I would like to pick up only this array (set of numbers between zeros) 0 10 20 30 50 20 5 0 and would like to know where they are. Thank you in advance!! atk On 6/25/2012 10:43 PM, ATSUKI OTO wrote: > Hi > > In excel sheet, I have the column like this: > > 0 > 0 > 0.001(error data) > 0 > 0 > 10 > 20 > 30 > 50 > 20 > 5 > 0 > 0 > 0.002(error data) > 0 > 0 &...

Finding a value in a array efficiently
I was asked an apparently simple question -- what is the most efficient way in IDL to determine if a particular scalar value is in an array. The loop method is straightforward: found = 0 for i=0,n-1 do begin if array[i] EQ value then begin found = 1 goto, DONE endif endfor DONE: This ugly code will end the loop as soon as a match is found. A first attempt at vectorized code might be index = where(array EQ value, ng) found = ng gt 0 But this code both creates an unnecessary index array, and requires searching the entire array even if the match is found in ...

finding dominant value in an array
I'm looking to implement a subroutine or function that accepts a vector of integers and returns the dominant (most prominent) value. For example, passed 7,3,9,7,7,2,4,1 it will return 7. Can't seem to envision anything simple and elegant. Anyone have code that does this or a pointer in the right direction? TIA. On Nov 11, 10:47=A0am, Mike <rawlin...@yahoo.com> wrote: > I'm looking to implement a subroutine or function that accepts a > vector of integers and returns the dominant (most prominent) value. > For example, passed 7,3,9,7,7,2,4,1 it will return 7. =A0Can't seem to > envision anything simple and elegant. Anyone have code that does this > or a pointer in the right direction? =A0TIA. program testing integer :: i(10) i =3D [7, 3, 9, 7, 7, 2, 4, 1, 5, 6] print *, dom(i) contains function dom(i) integer dom integer, intent(in) :: i(:) integer j, k, m m =3D -1 do j =3D minval(i), maxval(i) k =3D count(i =3D=3D j) if (k > m) then dom =3D j m =3D k endif end do end function dom end program testing -- steve On 2010-11-11 14:47:39 -0400, Mike <rawlins02@yahoo.com> said: > > I'm looking to implement a subroutine or function that accepts a > vector of integers and returns the dominant (most prominent) value. > For example, passed 7,3,9,7,7,2,4,1 it will return 7. Can't seem to...

how do i find the max value out of an array?
Hi, i have a text file which is comma seperated and i have extracted few values from the text in an array for example, 30, 1, 4,5, 31, 4, 2, 3, 32, 2,3,0, 38 my goal is to find the max of 30, 31, 32, 38 (i, i+4, i + 8, i+ 12....so on) i have put a split statement and in a loop and captured these 4 elements in an array. how do i find the max of these values, which is in an array using perl? thanks IJALAB <balaji.draj@gmail.com> wrote: >i have a text file which is comma seperated and i have extracted few >values from the text in an array for example, >30, 1, 4,...

try to find median value
i have trouble about finding the median value for my vector.. i have 1x30000 vector and have been sorted.. how can i get the value of median? thanks for help :) "Ermitha" wrote in message <jm0no8$2vr$1@newscl01ah.mathworks.com>... > i have trouble about finding the median value for my vector.. i have 1x30000 vector and have been sorted.. how can i get the value of median? thanks for help :) help median Bruno "Bruno Luong" <b.luong@fogale.findmycountry> wrote in message <jm0png$9jt$1@newscl01ah.mathworks.com>... > "Ermitha" wrote in mes...

Finding a value within an array
Given the array A={1 2 3; 'aa' 'bb' 'cc'; 4 5 6}, I would like to be able to find the index of an input variable, where let's say input=1. I tried to use [row,col]=find(A==input). However this gives the error message "Undefined function or method 'eq' for input arguments of type 'cell'." I also tried to use find(strcmp(A, input)) but that also gives an error since the input 1 is not a string. Do you have any suggestions? On 6/24/2012 2:18 PM, conan.so@gmail.com wrote: > Given the array A={1 2 3; 'aa' 'bb' '...

Finding closest value from array
Hello all, I have a matrix of 4 columns (x,y,a,b) and N rows: A = [ 1 2 74 87 1.05 2 2 3 4.05 3.05 3 4 ..... ..... ]; Now I have some points (x, y) B = [ 0.95 1.05 1.52 2.36 4.54 5 .... ....]; So I need to find the points closest to B(x,y) in matrix A. and then assign (a,b) values from A to corresponding values in B! Any help would be greatly appreciated! Thanks "anand makwana" wrote in message <mbgm1h$lf2$1@newscl01ah.mathworks.com>... > I...

find out the max value in array
Hi, I have the following problem,if I get the values of x-axes and y-axes like thak: x=get(src,'XData');y=get(src,'YData'); then I want to find out the max value of the points 0 to 200 in y-axes and its correspond x's value,I wrote the code as: y=y(0:200); [maxvalue,maxindex]=max(y); xvalue=x(maxindex); yvalue=maxvalue; when I run this code it has a wrong as: ??? Subscript indices must either be real positive integers or logicals. ??? Error while evaluating uimenu Callback. Why and what should I do for it? Thanks for you help! "lee rose" <blusunrui@gmai...

array of hashes (finding value)
I'm just looking for other possible solutions. Sorry, this is kinda a noob question. Currently I loop through the array and check the value of each hash key to find values. Is that the only and/or best way to find matches? example: @invoices = [{:id => 1, :first_name => 'nate'}, {:id => 2, :fist_name => 'greg'}, {:id => 3, :first_name =>}] @invoices.each do |invoice| if invoice[:id] == params[:contact_id] bla... bla... end end -- Posted via http://www.ruby-forum.com/. [Note: parts of this message were removed to make it a legal post.] On Mon, Jul 14, 2008 at 3:22 PM, Nate Leavitt <nateleavitt@gmail.com> wrote: > I'm just looking for other possible solutions. Sorry, this is kinda a > noob question. > > Currently I loop through the array and check the value of each hash key > to find values. Is that the only and/or best way to find matches? > > example: > > @invoices = [{:id => 1, :first_name => 'nate'}, {:id => 2, :fist_name => > 'greg'}, {:id => 3, :first_name =>}] > > @invoices.each do |invoice| > if invoice[:id] == params[:contact_id] > bla... bla... > end > end > -- > Posted via http://www.ruby-forum.com/. > > With the construct you have there it would be faster to just store the names in an array at the index corresponding to :id. Although if this is coming out of Rails that probably isn't a viable op...

find nearest value in array
I have an x,y,z matrix of zeros and ones and I want to find the nearest point containing a zero to a specified point containing a one, eg. [0 0 0 0 1 1 1 1 0 0; 0 0 0 1 1 1 1 1 0 0; 0 0 0 1 1 1 1 1 0 0; 0 0 0 1 1 1 1 1 0 0] and iwant to find the point containing a zero which is closest to say point (2,5) which would be in this case (1,4) . find seems to work on lines and columns but not in a 2D way. Any help? Rory. Hi, try this : M=[0 0 0 0 1 1 1 1 0 0 0 0 0 1 1 1 1 1 0 0 0 0 0 1 1 1 1 1 0 0 0 0 0 1 1 1 1 1 0 0] x=2 y=5 [r,c]=ind2sub(size(M),find(M==0)); d=(r-x).^2+(c-y).^2; idx=f...

finding key values in array
In my 2d array, I have ten columns with keys. To determine which set of keys to print out, I have the link set up to transer the year and the rank. $find=$array[$year][$rank] does nothing. How do I properly go about retrrieving a particular key value? With only these paramters to work with? On Wed, 04 Dec 2013 00:26:34 -0500, richard wrote: > In my 2d array, I have ten columns with keys. > To determine which set of keys to print out, I have the link set up to > transer the year and the rank. > > $find=$array[$year][$rank] > > does nothing. It as...

Finding coordinates of a value in an array
I have a 90 by 100 array, and I would like to extract the row and column number for the maximum element in each row. For examle if I have an array which is: 1 2 1 1 1 3 4 1 1 I would like to extract the fact that in the first row the highest value is in row 1 column 2, in 2nd row the highest value is in row 2 column 3 in 3rd row the highest value is in row 3 column 1 so that I get an array like this: 2 3 1 Showing which column the highest value is in for each row. Is this possible? and can anyone tell me how to do it? Thanks Jon In article <1140894708.409834.193720@e56g2000cwe.goog...

Web resources about - Find the median value of an array. - comp.lang.fortran

Median plane - Wikipedia, the free encyclopedia
Text is available under the Creative Commons Attribution-ShareAlike License ;additional terms may apply. By using this site, you agree to the ...

Median Line Analysis (@JF991) on Twitter
Twitter Registrieren Anmelden Folgen Median Line Analysis @JF991 Markets, Trading, Commodities. FX, Median Line Analysis. No advice. http://signalinea.com ...

What Sydney's median house price buys you in Brisbane
We&#8217;ve all seen the stories about absolute dumps selling in Sydney for astronomical prices. It's hard to believe that the Sydney median ...

Nearly a third of Sydney suburbs now have median house price of $1 million
Sydney's housing boom has&#160;continued for so long and been so widespread that almost one third of the city's suburbs are now members of&#160;the ...

The humble cornerstore key to lifting median house values - The Courier-Mail Search Search
THE humble cornerstore could be the missing link to jump-starting the value of your home, new research has found - with urban areas especially ...

The humble cornerstore key to lifting median house values
... has found - with urban areas especially the big winners. An AMEX study found the number of small businesses in a suburb could lift median house ...

Perth’s top performing suburbs for median real estate price increase - PerthNow Search Search
A LITTLE known suburb with a population of less than 1500 is Perth’s top performing suburb for median house price growth.

Melbourne - Sydney - House Prices - Apartments - Median
Melbourne's median house price growth exceeded Sydney's during the March quarter, according to new data released on Wednesday.

Median house price hits $600,000
Melbourne's median house price has topped $600,000 for the first time despite signs of a slowing market.

NZ median house price hits record
The NZ housing market was more buoyant in November for properties over $NZ1 million, contrasting with fewer sales of lower priced houses below ...

Resources last updated: 3/1/2016 8:36:44 AM