Bug: can't cluster simple data without ccore=True

[Jeff-Tian]

x = xmeans([[0],[0],[10],[10],[20],[20]],[[5],[20]], 20, 0.025)
x.process()

throws: ValueError: math domain error
But if turn on ccore = True, then the result is correct:

x = xmeans([[0],[0],[10],[10],[20],[20]],[[5],[20]], 20, 0.025, ccore=True)
x.process()

[annoviko]

@Jeff-Tian, thank you for the bug reporting and code example. It's a definitely bug that will be corrected soon.

[annoviko]

@Jeff-Tian, I have corrected it on '0.8.dev' branch. It works for 'CCORE' because 'std::log(x)' is able to perform calculation for values that are less or equal to zero:

double value = std::log(0); /* equals to '-inf' */

But it does not work for Python - exception is thrown instead.

[Jeff-Tian]

Wow, checked it and it works now! Thank you for so quickly turnaround.

[annoviko]

@Jeff-Tian, if you are going to use '0.8.dev' - it a development branch. So, you need to build CCORE by yourself. For linux:

cd pyclustering/ccore
make ccore_x64 # if you are working on x64
make ccore_x86 # if you are working on x86

And for windows using visual studio project 2015: pyclustering\ccore\ccore.sln

Here is an instruction how to do that: https://github.com/annoviko/pyclustering/wiki/Core-of-the-PyClustering

If you faced with problems related to pyclustering, do not hesitate to ask for help.

[Jeff-Tian]

@annoviko, thank you for the information! Yesterday I just checked out '0.8.dev' branch and run the python version very well. If later I need to build a CCORE I will reference the information you provided.