Is there a way to find the 'mode' of a histogram?

I generated a histogram (hist3) using two variables, is there a way to find the 'mode' from this histogram?

Thanks !

[Shashank Prasanna]

Are you fitting it to a certain multivariate distribution? We can then define mode. How does it look? does it have multiple modes?

[Sandy]

What I ended up doing was generating a contour graph for the data (2 variables). It contours the data by pairwise frequency distribution. The mode would be the highest point on this graph. That's what I am trying to find.

[dpb]

If have Curve Fitting Toolbox,

doc fit

If not you can fit a polynomial model in x and y by generating the coefficient matrix and using \ to solve. You can then solve that for a maximum either analytically or numerically.

Alternatively, you could use interp2 and interpolate on a finer mesh to approximate the maximum.

Similar to that would be the approach outlined earlier of just using the counts from hist3. More bins would produce closer estimates as long as the number of points is large enough per bin to avoid the noise.

Answer 1

[dpb]

Seems to me would be approximated as

[n,c]=hist3(....;
[mx,ix]=max(n);  % magnitude of most frequent bin

location would the the ix-th entry in the c cell-array of bin center values

--