Computation time performance profiling of feature calculation

[fedorov]

Some information that gives user the idea about the relative computation time could be helpful

[JoostJM]

I did a profiling using PyCharm on the Lung2 testcase, using the current master and all features enabled, with only original image, here is a snapshot of the most time intensive functions:

[JoostJM]

Call diagram (with time needed):

[fedorov]

The last one looks very helpful! Nice work!

[JoostJM]

As you can see, glcm appears to be the main bottleneck. Today, I tried a different approach to calculating the matrix and this resulted in about 5x speed increase for glcm alone (~3-4x faster for the whole calculation).

[fedorov]

Sounds promising! Also, thank you for going over the open issues and cleaning up the list. This is a good thing to do prior to the release. I have several urgent deadlines and a lot of work that needs to be done in the next week or so, so I apologize I won't have time to look into this soon.

[JoostJM]

Addressed by #135

[JoostJM]

Implementation of C which yields a large improvement: #158, total computation time was 13101 ms.
Here, shape and glrlm are not yet implemented, but code has two extra feature classes compared to profiling shown in this issue

[JoostJM]

With all matrix calculation in C, as well as surface area calculation, total computing time for 5 test cases is now just over 6 seconds

[JoostJM]

Call graph using Cmatrices #158 and reduction of SimpleITK calls proposed in #167:

[JoostJM]

Resolved by #202