| Use of ++ operator to count frequency of an entry (Was: Majority Voting) [message #65473] |
Sun, 08 March 2009 06:26  |
Allan Whiteford
Messages: 117
Registered: June 2006
|
Senior Member |
|
|
Hi,
A couple of weeks ago it was pointed out that the ++ operator used in a
vectorised form would work on repeated indices meaning that it's
possible to do this:
IDL> a=[1,2]
IDL> ++a[[0,0,0,1,0,0]]
IDL> print,a
6 3
here a[0] has been incremented five times because the index 0 appears
five times.
It was pointed out that this meant that David's mode calculation example of:
array = [1, 1, 2 , 4, 1, 3, 3, 2, 4, 5, 3, 2, 2, 1, 2, 6, -3]
distfreq = Histogram(array, MIN=Min(array))
maxfreq= Max(distfreq)
mode = Where(distfreq EQ maxfreq) + Min(array)
Print, mode
could be re-written as:
array = [1, 1, 2 , 4, 1, 3, 3, 2, 4, 5, 3, 2, 2, 1, 2, 6,-3]
f=intarr(max(array)-min(array)+1)
f[array-min(array)]++
junk=max(f,idx)
mode=idx + min(array)
print,mode
(in fact I see David's article on this has already been updated to
include the ++ solution). It was also shown that this solution wasn't
any slower than using histogram. It's not necessarily better but in
certain situations could be easier to read.
Anyway, it was pointed out that this meant that a++ behaves differently
from a+=1 which some people didn't like and others didn't mind. Some of
us (ok, mainly me) worried that using this behaviour of the ++ operator
leading to problems later if it turned out that the behaviour wasn't
intended by ITTVIS and they would try to rationalise ++ with +=.
I volunteered to contact them to query if the behaviour was intended and
if it could be considered future-proof. The response included:
"the output of the following commands seem to be the expected one (and
should not change in the future)
a = [1, 2]
a [[0, 0, 1, 0]]++
print,a"
so, as promised, this is me sharing the result of the query with the
rest of the group. Everyone can feel safe about relying on the ++
operator to work on repeated indices.
Thanks,
Allan
|
|
|
|
| Re: Use of ++ operator to count frequency of an entry (Was: Majority Voting) [message #65551 is a reply to message #65473] |
Mon, 09 March 2009 08:40  |
Allan Whiteford
Messages: 117
Registered: June 2006
|
Senior Member |
|
|
Carsten,
Carsten Lechte wrote:
> Allan Whiteford wrote:
>
>>
>> a = [1, 2]
>> a [[0, 0, 1, 0]]++
>> print,a"
>>
>
> Is a line missing here?
Kind of but not really: I was trying to cut and paste verbatim from the
e-mail from ITTVIS without modifying what they'd said. The answer (see
below) was implied in my original question so they didn't quote it in
their reply.
> What is the expected value of a
> according to ITT? [4,3]?
>
Yep, [4,3] which is what the current implementation gives.
Thanks,
Allan
|
|
|
|
|
|
| Re: Use of ++ operator to count frequency of an entry (Was: Majority Voting) [message #65553 is a reply to message #65473] |
Mon, 09 March 2009 05:33 |
Jeremy Bailin
Messages: 618
Registered: April 2008
|
Senior Member |
|
|
On Mar 8, 9:26 am, Allan Whiteford <allan-remove-th...@-and-this.phys-
dot-strath.ac.uk> wrote:
> Hi,
>
> A couple of weeks ago it was pointed out that the ++ operator used in a
> vectorised form would work on repeated indices meaning that it's
> possible to do this:
>
> IDL> a=[1,2]
> IDL> ++a[[0,0,0,1,0,0]]
> IDL> print,a
> 6 3
>
> here a[0] has been incremented five times because the index 0 appears
> five times.
>
> It was pointed out that this meant that David's mode calculation example of:
>
> array = [1, 1, 2 , 4, 1, 3, 3, 2, 4, 5, 3, 2, 2, 1, 2, 6, -3]
> distfreq = Histogram(array, MIN=Min(array))
> maxfreq= Max(distfreq)
> mode = Where(distfreq EQ maxfreq) + Min(array)
> Print, mode
>
> could be re-written as:
>
> array = [1, 1, 2 , 4, 1, 3, 3, 2, 4, 5, 3, 2, 2, 1, 2, 6,-3]
> f=intarr(max(array)-min(array)+1)
> f[array-min(array)]++
> junk=max(f,idx)
> mode=idx + min(array)
> print,mode
>
> (in fact I see David's article on this has already been updated to
> include the ++ solution). It was also shown that this solution wasn't
> any slower than using histogram. It's not necessarily better but in
> certain situations could be easier to read.
>
> Anyway, it was pointed out that this meant that a++ behaves differently
> from a+=1 which some people didn't like and others didn't mind. Some of
> us (ok, mainly me) worried that using this behaviour of the ++ operator
> leading to problems later if it turned out that the behaviour wasn't
> intended by ITTVIS and they would try to rationalise ++ with +=.
>
> I volunteered to contact them to query if the behaviour was intended and
> if it could be considered future-proof. The response included:
>
> "the output of the following commands seem to be the expected one (and
> should not change in the future)
>
> a = [1, 2]
> a [[0, 0, 1, 0]]++
> print,a"
>
> so, as promised, this is me sharing the result of the query with the
> rest of the group. Everyone can feel safe about relying on the ++
> operator to work on repeated indices.
>
> Thanks,
>
> Allan
Excellent news, thanks for confirming that with ITT!
-Jeremy.
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
