**** IEVS ***** Warren D. Smiths's 2007 infinitely extendible voting system comparison engine ****
Copyright (C): Warren D. Smith 2007. Anybody can use this non-commercially
provided they acknowledge me, notify me (and make available for my unrestricted use)
re any & all code improvements created or bugs found. If you want commercial use,
then we'll negotiate.
On Unix, Mac OSX, and Linux: Compilation: gcc IEVS.c -Wall -lm -O6 -o IEVS
for extra speed (but less safety) add the #define NDEBUG line at top.
On MSWINDOWS: change #define MSWINDOWS to TRUE. Should work, see notes below by Cary -
thanks much to David Cary for the MSWINDOWS port!! (Except that he introduced a bunch
of bugs with careless typecasts, which hopefully I now have fixed...)
What election methods are available?: fgrep EMETH IEVS.c
What utility-generators now available?: fgrep UTGEN IEVS.c
This program should also work (with minor modifications) under microsoft windows OS.
David Cary successfully ported it and his notes on how he did it, are below.
As a side effect, Cary also noticed how to speed up Yee-picture-generation significantly
and fixed a bug in OutputCompressedBarray(). :)
If you do such porting, please report your experience, how you compiled it,
etc. warren.wds AT gmail.com.
Software Architecture:
I. voting methods.
II. voting strategies.
III. ignorance generators
IV. utility generators
Anybody can add new voting methods, new strategies, or new utility generators.
The idea is to build a "Chinese menu" system which can investigate A*B*C*D kinds of scenarios
BUT the effort to write it is only A+B+C+D.
The information-flow-direction is IV-->III-->II-->I-->winners-->regrets.
I am initially writing this with not very many of each. It is a "skeleton"
system which can be fleshed out later by adding more of each of I,II,III,IV.
Please send your contributed routines to warren.wds AT gmail.com .
You should be able to make an implementation of a new method pretty easily by
imitating some already-implemented methods similar to yours.
BUGGY: Rouse??, Copeland??
As-yet unimplemented voting methods include:
ICA, Maxtree, range+runoff(based on range ballots with ties discarded),
RRV+runoff, Sarvo-methods, Banks set, Boehm, Dodgson.
Asset & Candidate-withdrawal-option-IRV (both don't really fit into our model).
To add a new voting method you need to:
add your new EMETH subroutine, and you need to change 3 more lines:
#define NumMethods PutCorrect#Methodshere
and the case lines in GimmeWinner() and PrintMethName().
(If you add a new "core" method also must change NumCoreMethods and
perhaps renumber the methods...)
Currently only rating-based, strict-ranking-based, and approval-based methods (& combinations)
are supported - equalities are not allowed in rankings. ("3-slot" methods
also supported.) To extend the code to allow
equalities, I suggest adding a boolean vector
with v[x]=TRUE meaning x is really tied with, not below,
the candidate immediately preceding. That is a later goal.
HonestyStrat() now supports an arbitrary mixture of honest and (one kind of) strategic voters.
More strategies can/will be added later.
3 utility generators now implemented (some parameterized), plus
another "reality-based" utility generator based on
Tideman-et-al's real-world election collection.
Future plans/ideas: translate from C into D?
(Templates & automatic array bounds checking would really help.)
auto-searcher for property violations?
L1 utils should be skewed distn.
Allow a user-selected voting-method subset.
Targeted BR finder (new user interface to BR finder).
Other strat generators.
Use of 2-candidate elections is knd of silly.
Need to make better controls on summarizers so you can ask it to summarize only
SUBSETS of the data (such as, honest voters only, large #s of candidates only, etc)...
More realistic strategies for plur+runoff, MCA, IRNR, Benham2AppRunoff?
Build good hybrid voting methods?
Add feature to input votes from user and tally election.
Two-humped Gaussian distributions?
RandomWinner regret mean & variance should be computed more exactly.
DMC & the like - do the approvals and rankings correspond with strategy? Should they?
Summarizer: Also look at dominance relations, "approval".
***************************
Now notes by David Cary, 2007-02-14 re modifications to port IEVS 2.58 from LINUX to
Windows XP using Microsoft Visual Studio .Net 2003.
David Cary's Changes (not listing ones WDS did anyhow) include:
1) Fix compile errors due to passing an int[] to a function that expects a uint[],
2) Fix compile errors due to passing a uint[] to a function that expects an int[],
3) Fix runtime error caused by opening the .bmp output file in text mode, to avoid
LF bytes being written as CR LF combinations.
4) Add an include for "stdafx.h".
This program was successfully compiled and run as a C++ program after making the following
adjustments to the default Visual Studio Project:
1) Increase the stack size to 4 MB (under Linker/System)
(the default is 1 MB, but sizeof(edata) is about 2.2 MB and lives on the stack).
All reported warnings were eliminated by:
1) Updating the Visual Studio Project to not report 64-bit issues
(under "C/C++" / General)
2) Demoting warning messages #4018 & 4244 to warning level 4 by adding command line options
/w44018 /w44244 under "C/C++" / Command Line.
These messages warn about implicit int/uint conversions and conversions from __int64
to unsigned int.
****************************
Now notes by Frank D. Martinez, 2014-04-16.
I am attempting to make the code more readible in the way I define
"more readible": as if I am reading something more like pseudocode.
I have changed the code from C to C++ because the STL helps facilitate
the "more readible" goal.
I am also working to reduce the number of global variables involved
because experience has shewn reducing global variables reduces the
risk of errors. Your "mileage" may vary.
I have a good deal more work to do as far as metaphorical "house
cleaning" so I can more easily understand how all the parts work
together.
****************************