[MINC-users] Compute overlap?

[Alex Zijdenbos]

On Wed, Aug 31, 2011 at 9:44 AM, Andrew Janke <a.janke at gmail.com> wrote:
> Hi Alex,
>
> On 31 August 2011 02:31, Alex Zijdenbos <zijdenbos at gmail.com> wrote:
>>    return (2.0 * $stats[1]) / (2.0 * $stats[1] + $stats[0]);
>
> You sure of that 2.0 * in the denominator?

Yup :)

> Wikipedia says something different:
>
>   http://en.wikipedia.org/wiki/Dice%27s_coefficient

Actually, it does not; the difference is that I use -add in the
mincmath call, so $stats[1] becomes n_intersection, and $stats[0] is
the sum of the relative complements. In other words, if the regions
are A and B, you calculate the denominator as

 n(A) + n(B)

which is equal to

 2 * n(A ^ B) + n(A\B) + n(B\A)

which is what I calculate (couldn't find the correct math symbols but
you get the picture). Same difference though.

> Although I am sure there are more things to compare with than just
> Dice's coefficient.

That too - I'm always fascinated how many ways there are to turn the 4
values of a 2x2 confusion matrix into a number :) voldiff calculates
some but there are many more.

-- A