#2 MAFIA: Incorrect supports output in -mfi mode

[Blhue]

Issue:
------

In -mfi mode, the supports for some maximal itemsets may be
output incorrectly. Specifically, they can be shuffled with other
output itemsets.

Example:
--------

I reproduced this with various supports on the chess dataset. I
compared output with multiple other data mining tools and also
manually counted supports (using grep) to verify the problem.
Please let me know if you need specific datasets and steps to
reproduce.

Solution:
---------

In SortLMFI(), subranges in the MFI vector are sorted. However,
the SupportCountList vector (which contains the corrosponding
supports) is not updated to match.

I added three statements to fix the problem; see 'diff -c' output
below:

*** ../../mafia-cvs/Mafia/src/Mafia.cpp Thu Dec 4 13:42:20 2003
--- ./Mafia.cpp Sun Feb 6 15:10:50 2005
***************
*** 531,536 ****
--- 531,542 ----
MFI[left] = MFI[right];
MFI[right] = tempBitmap;
tempBitmap = NULL;
+
+ // SUGGESTED FIX
+ int tempSupport = SupportCountList[left];
+ SupportCountList[left] = SupportCountList[right];
+ SupportCountList[right] = tempSupport;
+
left++;
right--;
}

Aside:
------

I do not know why I didn't observe faulty behavior in -fci mode.