Sridevi Polavaram <spolavar@gmu.edu>
Dear All:
Due to the modeling contest in China many students asked this question
regarding the width, height and depth computation.Since many students
are in the mailing list, I want to have all of them on the same page
regarding this functionality, I apologize to others if this is causing
inconvenience.
The older version of L-Measure computes w,h,d after doing the PCA
analysis first. This use to be the default setting to calculate the
w,h,and d values until the last release. However, the latest version
v4.1 which is currently available on the web,doesn't do the PCA
implicitly. Instead, it uses the original orientation of the given input
neuron to compute the w,h and d. In other words, the difference of Xmin
and Xmax values is the computed as width (considering the 95% of the
x-values and elimiting the 2.5% on either ends for outliers). Similarly,
the Y and Z axes apply to height and depth respectively. So, to
reiterate, the current version computes the w,h and d values without PCA
by simply assuming X, Y and Z axes for w, h, and d respectively.
On the other hand, the option to measure them using PCA is also still
available, but needs to be selected explicitly. Select the w,h,and d
functions in the function panel and check the PCA option and click the
Go button the values are now computed using the PCA.
For both the cases since this is a tree level function, you should make
note that there is only one value of width for the entire tree and in
the L-Measure output the min, avg and max columns have the same final
result. The total_sum column doesn't make sense for these three functions.
If you still have questions related to this issue, please feel free to
ask me, maybe emailing me directly instead of using the mailing list.
Especially, when there are too many people trying to seek help, I would
like to make the discretion of posting the answers to the mailing list.
Best of luck,
Sridevi |
发表于 2010-9-18 09:36:58
楼主