Thread Subject: operate over arrays with nans

dear all:
suppose i have an array X = [nan nan nan; nan 2 4]

for each column of X, i wish to find the mean of its non-nan elements.
in this case, i want the output to be [nan 3]. is there an easy and
efficient way to do this?

mean(X(~isnan(X))) yields 3, but i want the column by column mean.

gautam.

Gautam Sethi <gautamsethi@gmail.com> wrote in message <5dd47ef1-
3946-4565-b223-6f1bdc1f1131@i3g2000hsf.googlegroups.com>...
> dear all:
> suppose i have an array X = [nan nan nan; nan 2 4]
>
> for each column of X, i wish to find the mean of its non-nan
elements.
> in this case, i want the output to be [nan 3]. is there an
easy and
> efficient way to do this?
>
> mean(X(~isnan(X))) yields 3, but i want the column by column
mean.
>
> gautam.


"column by column" or "row by row" ?

X(~isnan(X)): collect and generate a "NEW VECTOR" which
contains finite elements of X

So, mean(X(~isnan(X))) returns scalar result.

Anh Huy Phan
RIKEN - BSI

Gautam Sethi <gautamsethi@gmail.com> wrote in message
<5dd47ef1-3946-4565-b223-
6f1bdc1f1131@i3g2000hsf.googlegroups.com>...
> dear all:
> suppose i have an array X = [nan nan nan; nan 2 4]
>
> for each column of X, i wish to find the mean of its non-
nan elements.
> in this case, i want the output to be [nan 3]. is there
an easy and
> efficient way to do this?
>
> mean(X(~isnan(X))) yields 3, but i want the column by
column mean.
>
> gautam.

There's NANMEAN in the Stats Toolbox, or you can use this
approach:

q = ~isnan(X) ; % define non-NaNs
Y = X ; % operate on a copy
Y(~q) = 0 ;
M = sum(Y) ./ sum(q) ;

hth
Jos