A math/algorithm puzzle!

Hi all...

Using VB in a appplication that needs to solve a bit of a math/algorithm
problem, and thought the good people here would have interesting ideas on
how to solve it...

Without going into too much detail, basically the code must find/suggest the
best pairs of numbers from a single list, and be able to exclude "outliers".

Pairing would be the best pairings possible by closest value.

Example:

data:
25
22
31
24
25
22
28
18
23

I would want to be able to adjust the sensitivity of the code to start to
trim outliers from the data, like "18" would seem to be the biggest
troublemaker in creating similar pairs... resulting in a list like this:


31 28
25 25
24 23
22 22
excluded: 18


The largest delta is three in this case... but you can see how the
logic/sorting might become complex, especially when the list of data becomes
much larger.

Thoughts?

Thx,

SMS

Why not just have VBA sort your data and then compare the difference between teh first and second numbers with the difference between the last and second to last numbers in your list? This would identify your greatest outlier and it could be handled accordingly.

rci

At first glance...

Why not work out the mean and standard deviation of your numbers and say...

a) exclude any number greater or less than 3 s.d.'s (as outliers)
b) try pairing numbers by their distance above and below the mean e.g check
for numbers within +/- 0.5 sd of mean and pair them, then look at numbers +/-
0.51 to 1sd from mean, then 1.01 to 1.5sd etc.

Just a preliminary thought...

Alex

"rci" wrote:

> Hi all...
>
> Using VB in a appplication that needs to solve a bit of a math/algorithm
> problem, and thought the good people here would have interesting ideas on
> how to solve it...
>
> Without going into too much detail, basically the code must find/suggest the
> best pairs of numbers from a single list, and be able to exclude "outliers".
>
> Pairing would be the best pairings possible by closest value.
>
> Example:
>
> data:
> 25
> 22
> 31
> 24
> 25
> 22
> 28
> 18
> 23
>
> I would want to be able to adjust the sensitivity of the code to start to
> trim outliers from the data, like "18" would seem to be the biggest
> troublemaker in creating similar pairs... resulting in a list like this:
>
>
> 31 28
> 25 25
> 24 23
> 22 22
> excluded: 18
>
>
> The largest delta is three in this case... but you can see how the
> logic/sorting might become complex, especially when the list of data becomes
> much larger.
>
> Thoughts?
>
> Thx,
>
> SMS
>

Another suggestion:

FIrst sort the numbers. This will put close numbers next to each other.
Then calculate the differences, for example A1 to A2, A2 to A3, etc. and
select the smallest differences for the pairs. Outliers would have large
differences from both their neighbors.
--
Gary's Student


"Alex" wrote:

> rci
>
> At first glance...
>
> Why not work out the mean and standard deviation of your numbers and say...
>
> a) exclude any number greater or less than 3 s.d.'s (as outliers)
> b) try pairing numbers by their distance above and below the mean e.g check
> for numbers within +/- 0.5 sd of mean and pair them, then look at numbers +/-
> 0.51 to 1sd from mean, then 1.01 to 1.5sd etc.
>
> Just a preliminary thought...
>
> Alex
>
> "rci" wrote:
>
> > Hi all...
> >
> > Using VB in a appplication that needs to solve a bit of a math/algorithm
> > problem, and thought the good people here would have interesting ideas on
> > how to solve it...
> >
> > Without going into too much detail, basically the code must find/suggest the
> > best pairs of numbers from a single list, and be able to exclude "outliers".
> >
> > Pairing would be the best pairings possible by closest value.
> >
> > Example:
> >
> > data:
> > 25
> > 22
> > 31
> > 24
> > 25
> > 22
> > 28
> > 18
> > 23
> >
> > I would want to be able to adjust the sensitivity of the code to start to
> > trim outliers from the data, like "18" would seem to be the biggest
> > troublemaker in creating similar pairs... resulting in a list like this:
> >
> >
> > 31 28
> > 25 25
> > 24 23
> > 22 22
> > excluded: 18
> >
> >
> > The largest delta is three in this case... but you can see how the
> > logic/sorting might become complex, especially when the list of data becomes
> > much larger.
> >
> > Thoughts?
> >
> > Thx,
> >
> > SMS
> >