Re: Potential Addition: hist3

[Daniel J Sebald]

On 06/21/2012 10:14 PM, Andrew Lampinen wrote:
> Hello,
>
> Attached is a hist3.m file for a bivariate histogram function, somewhat
> similar to matlab's hist3. It's one of the main functions I've missed in
> octave, so I wrote this and thought I'd submit it to you for
> consideration, since it seems like it would be useful for matlab
> compatibility, etc. My code is probably not the most efficient, since
> I've not had extensive experience writing for octave, but at least it's
> a starting point.

Hi Andrew,

Yes, there may be some ways to speed up the code.  In particular, this loop:

for i = 1:length (X) %For each data point
  x_loc = find(x_bins > X(i,1))(1) - 1;
  y_loc = find(y_bins > X(i,2))(1) - 1;
  A(x_loc, y_loc) += 1;
endfor

Could possibly be done without looping or without using find().  Here's 
a path to explore:

* Imagine some affine transform on X, i.e., Y = m*X + b so that if one 
were to toss the fractional part of Y what would be left is the bin number.
* Use sum(Y == i, 1) to figure the count.

Now, this all depends on the number of bins compared to the number of 
samples.  If the number of bins is large compared to the number of 
samples then your first method is more efficient.

Another thing to consider might be the use of standard deviation rather 
than min and max for determining bin width.  The problem with min/max is 
that for some distributions those outliers can vary wildly so the 
histogram could end up with a resolution not matching the bulk of the 
data so well.  Try using standard deviation for determining bin size, 
and either toss the outliers or group the outliers with the extreme 
bins.  (A few extra at the extrema is the way a Galton's board can look 
if the span of the bins isn't very wide: 
http://ptrow.com/articles/Galton_June_07.htm)

Dan