[SciPy-User] Value that compare two

[josef.pktd at gmail.com]

On Tue, Apr 2, 2013 at 10:15 AM, Florian Lindner <mailinglists at xgm.de> wrote:
> Hello,
>
> this is not exactly a scipy question... but I want to implement it with scipy.
> ;-)
>
> I have two datasets of shape: (n, 2), each row consists of a coordinate and a
> pressure value from experiments or simulations.
>
> I want to compare these two sets and get some kind of integral distance value.
>
> delta = abs(data2 - data1)
> delta[:,0] = data1[:,0] # I don't want to delta the coordinates
> sum_delta = np.trapz(delta[:,1], x = delta[:,0])
>
> This works fine, but I also want to have a normalized delta value (aka
> percentage).
>
> Before I try to invent another wheel which at the end will look rather
> rectangular:
>
> Is there some best practice way to compute such a value?
>
> If one could also give a quotable source of the algorithm it would be even
> more perfect!

I'm more familiar with quadratic than absolute error for comparing
functions or distributions in applications

http://books.google.ca/books?id=my-i9VNzCfAC&pg=PA260&lpg=PA260&dq=mise+mean+integrated+absolute+error&source=bl&ots=WQls8u4iYi&sig=eipR2JTk4dgAfxqWEFdB9bjzL_0&hl=en&sa=X&ei=tQNbUaTQC7PM0gHsxoGoAg&ved=0CC8Q6AEwAA#v=onepage&q=mise%20mean%20integrated%20absolute%20error&f=false

ISE integratead squared error
MISE mean integrated squared error
MIAE mean integrated absolute error

these are common measures for non-parametric estimation.

google search for mean integrated absolute error gave me the above link.

There are various other functional distance measures, where I also
know mainly the goodness-of-fit measures for probability
distributions.

Josef

>
> Thanks,
>
> Florian
> _______________________________________________
> SciPy-User mailing list
> SciPy-User at scipy.org
> http://mail.scipy.org/mailman/listinfo/scipy-user