Add preview of 1984-2019
The 2020 winners are hinted it, but the source code is not there,
nor are the hints. Someone could look into the HTML and
see what was commented out, but that is not a problem because
WHO won is known and the name of their AWARD is known already.
Moreover a number of the winners have talked about what their
entirs do, so the commented out spoiler text is not a bit problem.
One will find HTML commented out with:
<!-- XXX remove when 2020 is ready
2020 links here
XXX remove when 2020 is ready -->
The "XXX remove when 2020 is ready" string is found at the
start and end comment blocks. Again, soneone could read
what was commented out, but that is OK since most of that
information is known or can be known. The following files
have such HTML comment blocks in them.
./winners.html
./years-spoiler.html
./years.html
In addition, ./all/summary.txt was modified to not
include the 2020 entries for the momemt.
The 2020 winner code AND write-ups are NOT part of this set.
p.s. The all/all.tar.bz2 does not contain the 2020 source code (yet?).
This commit is contained in:
Landon Curt Noll committed
1 parent
28efc67f5d
commit
c0663537cb
1683 files changed
+220064
No files matched your search
@@ -0,0 +1,127 @@
|
||||
#!/usr/bin/env make
|
||||
#
|
||||
# 2011 makefile
|
||||
#
|
||||
# This work by Landon Curt Noll, Simon Cooper, and Leonid A. Broukhis
|
||||
# is licensed under:
|
||||
#
|
||||
# Creative Commons Attribution-ShareAlike 3.0 Unported License.
|
||||
#
|
||||
# See: http://creativecommons.org/licenses/by-sa/3.0/
|
||||
|
||||
|
||||
################
|
||||
# tool locations
|
||||
################
|
||||
#
|
||||
SHELL= /bin/bash
|
||||
CP= cp
|
||||
CPP= cpp
|
||||
GUNZIP= gunzip
|
||||
LD= ld
|
||||
MAKE= make
|
||||
RM= rm
|
||||
SED= sed
|
||||
TAR= tar
|
||||
TRUE= true
|
||||
|
||||
# Compiler warnings
|
||||
#
|
||||
CWARN= -Wall -W
|
||||
|
||||
# compiler standard
|
||||
#
|
||||
CSTD= -std=c99
|
||||
|
||||
# compiler bit architecture
|
||||
#
|
||||
# Some entries require 32-bitness,
|
||||
# other entries require 64-bitess.
|
||||
# By default we assume nothing.
|
||||
#
|
||||
ARCH=
|
||||
|
||||
# optimization
|
||||
#
|
||||
# Most compiles will safely use -O2. Some can use only -O1 or -O.
|
||||
# A few compilers have broken optimizers or this entry make break
|
||||
# under those buggy optimizers and thus you may not want anything.
|
||||
#
|
||||
OPT= -O2
|
||||
|
||||
# default flags for ANSI C compilation
|
||||
#
|
||||
CFLAGS= ${CWARN} ${CSTD} ${ARCH} ${OPT} -lm -D_BSD_SOURCE
|
||||
|
||||
# ANSI compiler
|
||||
#
|
||||
# Set CC to the name of your ANSI compiler.
|
||||
#
|
||||
CC= cc
|
||||
|
||||
|
||||
##############################
|
||||
# Special flags for this entry
|
||||
##############################
|
||||
#
|
||||
ENTRY= dlowe
|
||||
DATA= earlystop.pl
|
||||
|
||||
|
||||
#################
|
||||
# build the entry
|
||||
#################
|
||||
#
|
||||
all: ${ENTRY} ${DATA}
|
||||
@${TRUE}
|
||||
|
||||
${ENTRY}: ${ENTRY}.c
|
||||
${CC} ${CFLAGS} ${ENTRY}.c -o ${ENTRY}
|
||||
|
||||
# alternative executable
|
||||
#
|
||||
alt:
|
||||
@${TRUE}
|
||||
|
||||
|
||||
###############
|
||||
# utility rules
|
||||
###############
|
||||
#
|
||||
everything: all alt
|
||||
|
||||
clean:
|
||||
${RM} -f ${ENTRY}.o
|
||||
|
||||
clobber: clean
|
||||
${RM} -f ${ENTRY}
|
||||
|
||||
nuke: clobber
|
||||
@echo gnab gib!
|
||||
|
||||
install:
|
||||
@echo "Surely you're joking Mr. Feynman!"
|
||||
|
||||
# backwards compatibility
|
||||
#
|
||||
build: all
|
||||
|
||||
|
||||
##################
|
||||
# 133t hacker rulz
|
||||
##################
|
||||
#
|
||||
love:
|
||||
@echo 'not war?'
|
||||
|
||||
haste:
|
||||
$(MAKE) waste
|
||||
|
||||
waste:
|
||||
@echo 'waste'
|
||||
|
||||
easter_egg:
|
||||
@echo you expected to mis-understand this $${RANDOM} magic
|
||||
@echo chongo '<was here>' "/\\oo/\\"
|
||||
@echo Readers shall not be disallowed from being unable to partly misunderstand this final echo.
|
||||
@${TRUE}
|
||||
Binary file not shown.
@@ -0,0 +1,43 @@
|
||||
#include <stdlib.h>
|
||||
#include <string.h>
|
||||
#include <dirent.h>
|
||||
#include <stdio.h>
|
||||
#include <math.h>
|
||||
#include <time.h>
|
||||
#define _ float
|
||||
|
||||
_ CRUSH(_*LEG,int ARM,_*FINGER) {
|
||||
_ GORE; for(GORE=0,--ARM; ARM>=0; --ARM)GORE+=FINGER[ARM]*ARM
|
||||
[LEG]; return GORE; } _*BITE(){ _*BRAIN=calloc(sizeof(_),1<<17);int TOE
|
||||
=getc(stdin),EYE,SKULL=0; while((EYE=getc(stdin))!=EOF){ ++BRAIN[(TOE<<8) +
|
||||
EYE]; ++SKULL; TOE=EYE; } if(SKULL)for(TOE=0; TOE<8<<13; ++ TOE)BRAIN [TOE]/=
|
||||
SKULL; return BRAIN; } _ CHEW(_*GUT,_ BONE[][1<<16]){ int GRR; for(GRR=0; GRR<
|
||||
6; ++GRR){ GUT[GRR+256*256]=powf(1+expf(-CRUSH(BONE[GRR],1<<16,GUT)),-1); BONE
|
||||
[6][50+GRR]=GUT[256*256+GRR]*(1.-GUT[(8<<13)+GRR]); } BONE[6][81]=1/(1 +expf(-
|
||||
CRUSH(GUT+(1<<16),6,BONE[6]))); return BONE[6][82]=BONE[6][81]*(1.-BONE[6][81]
|
||||
),BONE[6][81]; } _ GNAW(_ FLESH,_ LEG[][2<<15],_*EYE){ int UG,MMM; LEG[6][13]=
|
||||
FLESH-CHEW(EYE,LEG); LEG[6][14]=LEG[6][ 82]*LEG[6][13]; for(UG=0; UG<6; ++UG){
|
||||
LEG[6][34]=LEG[6][UG+(1<<7)-14]*LEG[6][7<<1]*LEG[6][UG]; for(MMM=0; MMM<4<<14;
|
||||
++MMM)LEG [UG][MMM]+=LEG[6][34]*.3*EYE[MMM]; LEG[6][UG]+=.3*LEG[6][14
|
||||
]*EYE[ 256*256+UG]; } return powf(LEG[6] [13],2); } _
|
||||
**EAT( char*TOMB){ DIR*BONE = opendir( TOMB); int
|
||||
BRAIN= 0; struct dirent*TOOTH; _**BODY =0; while
|
||||
(BONE?(TOOTH= readdir(BONE)):0){ if( TOOTH-> d_name[0]
|
||||
!=46){ char*MOAN=malloc(strlen(TOMB)+strlen(TOOTH->d_name )+1); sprintf(
|
||||
MOAN, "%s%s",TOMB,TOOTH ->d_name); if(freopen(MOAN, "r",stdin)){ BODY=realloc(
|
||||
BODY,sizeof(_*)*(BRAIN+1)); BRAIN ++ [BODY]=BITE(); } } } return BODY=realloc(
|
||||
BODY,(1+BRAIN)*sizeof(_*)),BODY[BRAIN]=0,BODY; } int main(int GRR,char **UGH){
|
||||
_ BRAINS[7][1<<16],***CORPSES; int PUS,OOZE,UG; for(srand(time(0)),PUS=0; PUS<
|
||||
7; ++PUS)for(OOZE=0; OOZE<4<<14; ++OOZE)BRAINS[PUS][OOZE]=rand()/(_)RAND_MAX
|
||||
-.5;fread(BRAINS,sizeof(BRAINS),1,stdin); if(*UGH[1]==45){ GRR-=2; CORPSES
|
||||
=malloc( sizeof(_
|
||||
**)*GRR ); for(
|
||||
PUS=0; PUS<
|
||||
GRR; ++ PUS)CORPSES[PUS]=EAT(UGH[2+PUS]); for(UG=
|
||||
0; UG<atoi (&(UGH[1][1])); ++UG){ BRAINS[6][97]= 0; for(PUS
|
||||
=0; PUS<GRR; ++PUS)for(OOZE=0; CORPSES[PUS][OOZE]; ++OOZE)BRAINS [6][
|
||||
97]+=GNAW(1.-(_) PUS/(GRR-1),BRAINS,CORPSES[PUS][OOZE]); fprintf(
|
||||
stderr,"%d: %f\n",UG,BRAINS[6][97]); } fwrite( BRAINS, sizeof(
|
||||
BRAINS),1,stdout); } else for(UG=1; UG<GRR; ++UG)if(freopen(
|
||||
UGH[UG],"r",stdin))fprintf(stderr,"%s %f\n",UGH[UG],
|
||||
CHEW(BITE(),BRAINS)); return 0; }
|
||||
@@ -0,0 +1,37 @@
|
||||
#!/usr/bin/env perl
|
||||
|
||||
use strict;
|
||||
use warnings;
|
||||
|
||||
if ($#ARGV != 4) {
|
||||
die "usage: $0 <net> <train corpus 1> <train corpus 0> <test corpus 1> <test corpus 0>\n"
|
||||
}
|
||||
|
||||
my ($net, $train1, $train0, $test1, $test0) = @ARGV[0..4];
|
||||
my $error = undef;
|
||||
my $n = 1;
|
||||
|
||||
open TOUCH, '>', $net;
|
||||
close TOUCH;
|
||||
|
||||
while (1) {
|
||||
print "training $n to " . ($n+1000) . "...\n";
|
||||
system "./dlowe -1000 $train1 $train0 < $net > $net.tmp 2>/dev/null";
|
||||
rename "$net.tmp", $net;
|
||||
|
||||
my $current_error = 0;
|
||||
for my $r (split /\s+/, `./dlowe $test1* < $net 2>&1 | awk '{print \$NF}'`) {
|
||||
$current_error += ((1.0 - $r) ** 2);
|
||||
}
|
||||
for my $r (split /\s+/, `./dlowe $test0* < $net 2>&1 | awk '{print \$NF}'`) {
|
||||
$current_error += ((0.0 - $r) ** 2);
|
||||
}
|
||||
if (defined $error) {
|
||||
print "prior error = $error; current error = $current_error\n";
|
||||
if ($error - $current_error < 0.01) {
|
||||
last;
|
||||
}
|
||||
}
|
||||
$error = $current_error;
|
||||
$n += 1000;
|
||||
}
|
||||
@@ -0,0 +1,336 @@
|
||||
<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN" "http://www.w3.org/TR/html4/loose.dtd">
|
||||
<html>
|
||||
<head>
|
||||
<meta name="GENERATOR" content="mkd2html 2.1.3 DL=DISCOUNT">
|
||||
<META HTTP-EQUIV="Content-Type" CONTENT="text/html; charset=utf-8">
|
||||
<link rel="stylesheet"
|
||||
type="text/css"
|
||||
href="../hint.css">
|
||||
<title>Most self deprecating</title>
|
||||
</head>
|
||||
<body>
|
||||
<h1>Most self deprecating</h1>
|
||||
|
||||
<p>David Lowe<br>
|
||||
434 Tenney Drive<br>
|
||||
Rogue River, OR<br>
|
||||
97537<br>
|
||||
<a href="mailto:j.david.lowe@gmail.com">j.david.lowe@gmail.com</a><br>
|
||||
<a href="http://www.pootpoot.net/">http://www.pootpoot.net/</a></p>
|
||||
|
||||
<h2>Judges' comments:</h2>
|
||||
|
||||
<h3>To build:</h3>
|
||||
|
||||
<pre><code>make dlowe
|
||||
</code></pre>
|
||||
|
||||
<h3>To run:</h3>
|
||||
|
||||
<pre><code>./dlowe -<n_iterations> corpus1/ [...] corpus0/ < start.net > end.net
|
||||
|
||||
NOTE: The directory entries MUST end in a /
|
||||
|
||||
./dlowe file [file ...] < trained.net
|
||||
</code></pre>
|
||||
|
||||
<h3>Try:</h3>
|
||||
|
||||
<pre><code>Download the dlowe-aux-data.tar.bz2 file from:
|
||||
|
||||
curl -O http://www.ioccc.org/2011/dlowe/dlowe-aux-data.tar.bz2
|
||||
|
||||
tar -jxf dlowe-aux-data.tar.bz2
|
||||
|
||||
./dlowe dlowe.c < ioccc-winlose-trained.net
|
||||
./dlowe dlowe < ioccc-winlose-trained.net
|
||||
|
||||
./dlowe xor-0/00 < xor-trained.net
|
||||
./dlowe xor-0/11 < xor-trained.net
|
||||
./dlowe xor-1/01 < xor-trained.net
|
||||
./dlowe xor-1/10 < xor-trained.net
|
||||
|
||||
./dlowe hint.html < english-trained.net
|
||||
|
||||
./dlowe dlowe-aux-data.tar.bz2 < png-trained.net
|
||||
</code></pre>
|
||||
|
||||
<h3>Selected Judges Remarks:</h3>
|
||||
|
||||
<p>This entry is a ghoulish example of a brain (dead?) neural network classifier.
|
||||
It comes with a set of corpora for you to try.</p>
|
||||
|
||||
<h4>ioccc-winlose-trained.net</h4>
|
||||
|
||||
<p>You can create your own trained networks. For example, we trained a network
|
||||
on the C code of entries that were supplied to the 20Th IOCCC.</p>
|
||||
|
||||
<p>The ioccc-winlose-trained.net was trained using the earlystop.pl tool:</p>
|
||||
|
||||
<pre><code>rm -f ioccc-winlose-trained.net
|
||||
./earlystop.pl ioccc-winlose-trained.net ioccc_won_training/ ioccc_lost_training/ ioccc_won_test/ ioccc_lost_test/
|
||||
</code></pre>
|
||||
|
||||
<p>where:</p>
|
||||
|
||||
<pre><code>ioccc_won_training/
|
||||
1/2 of C code of the 20th IOCCC winners (except for dlowe.c from 2011)
|
||||
plus 1/2 of the C code winners from prior IOCCC contests
|
||||
|
||||
ioccc_won_test/
|
||||
1/2 of C code of the 20th IOCCC winners (except for dlowe.c from 2011)
|
||||
plus 1/2 of the C code winners from prior IOCCC contests
|
||||
|
||||
ioccc_lost_training/
|
||||
1/2 of C code from entries that that did not win
|
||||
|
||||
ioccc_lost_test/
|
||||
1/2 of C code from entries that that did not win
|
||||
|
||||
NOTE: The above directies were not shipped and the C code from
|
||||
entries that that did not win was flushed to /dev/null.
|
||||
</code></pre>
|
||||
|
||||
<p>The earlystop.pl tool wrote:</p>
|
||||
|
||||
<pre><code>training 1 to 1001...
|
||||
training 1001 to 2001...
|
||||
prior error = 56.746974766043; current error = 58.095691135305
|
||||
</code></pre>
|
||||
|
||||
<p>The resulting ioccc-winlose-trained.net file was included in the
|
||||
dlowe-aux-data.tar.bz2 file in the above mentioned URL.</p>
|
||||
|
||||
<p>The dlowe.c was explicitly excluded from this trailing set. So this test is interesting:</p>
|
||||
|
||||
<pre><code>./dlowe dlowe.c < ioccc-winlose-trained.net
|
||||
</code></pre>
|
||||
|
||||
<p>The match of dlowe.c to the lose/win trailed network was:</p>
|
||||
|
||||
<pre><code>dlowe.c 0.125108
|
||||
</code></pre>
|
||||
|
||||
<p>NOTE: The non-artificially intelligent decisions of the IOCCC judges include random
|
||||
quirks and their non-linear judgment calls not found in the ioccc trained network.
|
||||
Just because C code is or isn’t scored highly by this network does not
|
||||
mean it will win or lose a future IOCCC. On the other hand … it just might!</p>
|
||||
|
||||
<h4>other data sets to try</h4>
|
||||
|
||||
<p>While we cannot supply you with the losing IOCCC source code, the author did supply a number
|
||||
data sets on which to test:</p>
|
||||
|
||||
<p>We created the english-trained.net as follows:</p>
|
||||
|
||||
<pre><code>./dlowe -8000 english-1/ english-0/ < /dev/null > english-trained.net
|
||||
|
||||
english-0/
|
||||
Non-English (French) text
|
||||
english-1/
|
||||
English (non-French) text
|
||||
</code></pre>
|
||||
|
||||
<p>We created the png-trained.net as follows:</p>
|
||||
|
||||
<pre><code>./dlowe -8000 png-1/ png-0/ < /dev/null > png-trained.net
|
||||
|
||||
png-0/
|
||||
Non-png (gif) images
|
||||
png-1/
|
||||
png (non-gif) images
|
||||
</code></pre>
|
||||
|
||||
<p>We created the xor-trained.net as follows:</p>
|
||||
|
||||
<pre><code>./dlowe -8000 xor-1/ xor-0/ < /dev/null > xor-trained.net
|
||||
|
||||
xor-0/
|
||||
Data that XORs to 0
|
||||
xor-1/
|
||||
Data that XORs to 1
|
||||
</code></pre>
|
||||
|
||||
<h2>Author’s comments:</h2>
|
||||
|
||||
<h1>Synopsis:</h1>
|
||||
|
||||
<p>This is an artificially intelligent judging tool to help the IOCCC judges.
|
||||
Here’s to shorter, more frequent contests!</p>
|
||||
|
||||
<h1>Description:</h1>
|
||||
|
||||
<p>This is a multilayer perceptron (a feedforward artificial neural network)
|
||||
which can be trained, using on-line backpropagation, to classify input files.</p>
|
||||
|
||||
<p>It has a fixed topology of 2<sup>16</sup> input neurons, 6 hidden neurons and 1 output
|
||||
neuron.</p>
|
||||
|
||||
<p>The neurons' activation function is the logistic function 1 / (1 + e ^ -x).</p>
|
||||
|
||||
<h2>Classifying</h2>
|
||||
|
||||
<pre><code>./dlowe file [file ...] < trained.net
|
||||
</code></pre>
|
||||
|
||||
<p>To classify files, one specifies a trained network (on stdin) and one or more
|
||||
files to classify. The program will output one line per successfully-classified
|
||||
file to stderr, with the filename and the classification: a number between 0
|
||||
and 1.</p>
|
||||
|
||||
<p>The interpretation of the classification number depends on how the network was
|
||||
trained, but it’s geared toward interpretation as a <em>probability</em> or a
|
||||
<em>confidence</em>.</p>
|
||||
|
||||
<h2>Training</h2>
|
||||
|
||||
<pre><code>./dlowe -<n_iterations> corpus1/ [...] corpus0/ < start.net > end.net
|
||||
|
||||
NOTE: The directory args must end in a / for them to work.
|
||||
</code></pre>
|
||||
|
||||
<p>To train a network, one specifies a starting network (on stdin), two or more
|
||||
corpora (directories containing training data), and the number of training
|
||||
iterations to run. The program will write some progress data to stderr and,
|
||||
when it’s done, will serialize the updated network to stdout.</p>
|
||||
|
||||
<p>If no input network is given, a random new network will be generated as a
|
||||
starting point (i.e. provide an empty stdin).</p>
|
||||
|
||||
<p>The first corpus will be assigned a target value of 1. The last will be
|
||||
assigned a target value of 0. Intervening directories (if any) will be assigned
|
||||
intermediate target values.</p>
|
||||
|
||||
<p>The learning rate is hard-coded as 0.3. No momentum factor is used.</p>
|
||||
|
||||
<h3>png corpora</h3>
|
||||
|
||||
<p>The png-1 corpus was obtained by manually scraping the first results from an
|
||||
https://images.google.com search for “obfuscate filetype:png”.</p>
|
||||
|
||||
<p>The png-0 corpus was obtained by manually scraping the first results from an
|
||||
https://images.google.com search for “obfuscate filetype:gif”.</p>
|
||||
|
||||
<p>Results:</p>
|
||||
|
||||
<ul>
|
||||
<li>using these corpora as a training set</li>
|
||||
<li>using several hundred random <em>.gif and </em>.png files from my home computer as
|
||||
a test set</li>
|
||||
<li>using the included “earlystop.pl” script, which stopped after 67000 iterations.</li>
|
||||
<li>interpreting output of “> 0.5” as “probably a .png” and “< 0.5” as “probably
|
||||
a .gif”</li>
|
||||
<li>produced a network with about 87% accuracy</li>
|
||||
</ul>
|
||||
|
||||
|
||||
<h3>english corpora</h3>
|
||||
|
||||
<p>The english-1 corpus was obtained by manually scraping the first results from
|
||||
a https://www.google.com/webhp?lr=lang_en search for “paris filetype:txt”.</p>
|
||||
|
||||
<p>The english-0 corpus was obtained by manually scraping the first results from
|
||||
a https://www.google.com/webhp?lr=lang_fr search for “paris filetype:txt”.</p>
|
||||
|
||||
<p>Results:</p>
|
||||
|
||||
<ul>
|
||||
<li>using these corpora as a training set</li>
|
||||
<li>using several dozen additional text files (obtained by the same methods)
|
||||
as a test set</li>
|
||||
<li>using the included “earlystop.pl” script, which stopped after 10000
|
||||
iterations.</li>
|
||||
<li>interpreting output of “> 0.5” as “probably english” and “< 0.5” as
|
||||
“probably french”</li>
|
||||
<li>produced a network with 100% accuracy</li>
|
||||
</ul>
|
||||
|
||||
|
||||
<h3>xor corpora</h3>
|
||||
|
||||
<p>The xor-1 corpus consists of two files containing ‘01’ and ‘10’ respectively.</p>
|
||||
|
||||
<p>The xor-0 corpus consists of two files containing ‘00’ and ‘11’ respectively.</p>
|
||||
|
||||
<p>Results:</p>
|
||||
|
||||
<ul>
|
||||
<li>using these corpora as a training set</li>
|
||||
<li>takes about 8000 iterations to learn xor to within a tolerance of <0.01</li>
|
||||
</ul>
|
||||
|
||||
|
||||
<h1>Limitations</h1>
|
||||
|
||||
<p>The program can’t tell you anything meaningful about files with less than two
|
||||
bytes in them (I’m looking at you, smr.c!)</p>
|
||||
|
||||
<p>You must include the trailing directory separator on training directories
|
||||
(this allows the program to be portable without wasting precious bytes on
|
||||
figuring out how to concatenate directory and file names…)</p>
|
||||
|
||||
<p>Serialized network files are only portable between systems with the same
|
||||
floating-point representation and endianness.</p>
|
||||
|
||||
<p>Making sure not to overfit the network to the training data is a bit of a
|
||||
black art. I have enclosed ‘earlystop.pl’, a wrapper script that implements
|
||||
a simple ‘early stopping’ algorithm; other techniques are possible.</p>
|
||||
|
||||
<p>Bad input (e.g. nonexistent files, non-numeric number of iterations, etc.)
|
||||
tends to result in empty output.</p>
|
||||
|
||||
<p>Given exactly one corpus, the program will crash or produce garbage.</p>
|
||||
|
||||
<p>Leaks memory and file descriptors while processing files.</p>
|
||||
|
||||
<p>Will crash and die horribly if it runs out of memory.</p>
|
||||
|
||||
<p>The Microsoft C compiler doesn’t provide a dirent API, so to get this working
|
||||
on a Windows system you’ll need cygwin+gcc (tested) or a dirent compatibility
|
||||
library (untested, but they do exist).</p>
|
||||
|
||||
<p>Backpropagation doesn’t always converge: if you play with this long enough,
|
||||
you’ll eventually have a training session that completely fails to converge.</p>
|
||||
|
||||
<h1>Obfuscation:</h1>
|
||||
|
||||
<p>Zombies! (Since neural networks are modeled after BRAINS, ya know? And
|
||||
corpus sounds a lot like corpse. And I have 4- and 7-year-old kids ;)</p>
|
||||
|
||||
<p>Neural networks are interesting, and while the math isn’t terribly difficult,
|
||||
their behavior is difficult to fully understand.</p>
|
||||
|
||||
<p>Had some extra space in one of the major data structures, and it seemed a
|
||||
shame to waste it.</p>
|
||||
|
||||
<p>Similarly, I think three file pointers ought to be enough for anyone.</p>
|
||||
|
||||
<p>Lots of magic numbers expressed in various ways.</p>
|
||||
|
||||
<p>Random abuse of random C trivia.</p>
|
||||
|
||||
<p>Some textual changes were made to maximize the classification of this program
|
||||
by a network trained to recognize ioccc winners ;)</p>
|
||||
|
||||
<p>… but mostly zombies!</p>
|
||||
|
||||
<hr />
|
||||
|
||||
<!--
|
||||
(c) Copyright 1984-2012, [Leo Broukhis, Simon Cooper, Landon Curt Noll][judges] - All rights reserved
|
||||
This work is licensed under a [Creative Commons Attribution-ShareAlike 3.0 Unported License][cc].
|
||||
|
||||
[judges]: http://www.ioccc.org/judges.html
|
||||
[cc]: http://creativecommons.org/licenses/by-sa/3.0/
|
||||
-->
|
||||
|
||||
<TABLE><TR>
|
||||
<TD><a rel="license" href="http://creativecommons.org/licenses/by-sa/3.0/"><img alt="Creative Commons License" style="border-width:0" src="http://i.creativecommons.org/l/by-sa/3.0/88x31.png" /></a></TD>
|
||||
<TD><P>© Copyright 1984-2012,
|
||||
<A HREF="/judges.html">Leo Broukhis, Simon Cooper, Landon Curt Noll</A>
|
||||
- All rights reserved<BR>
|
||||
This work is licensed under a <a rel="license" href="http://creativecommons.org/licenses/by-sa/3.0/">Creative Commons Attribution-ShareAlike 3.0 Unported License</a>.</P></TD>
|
||||
<TD> <!--<a href="http://validator.w3.org/check?uri=referer"><img src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01 Transitional" height="31" width="88"></a>--></TD>
|
||||
</TR></TABLE>
|
||||
</body>
|
||||
</html>
|
||||
@@ -0,0 +1,287 @@
|
||||
# Most self deprecating
|
||||
|
||||
David Lowe
|
||||
434 Tenney Drive
|
||||
Rogue River, OR
|
||||
97537
|
||||
<j.david.lowe@gmail.com>
|
||||
<http://www.pootpoot.net/>
|
||||
|
||||
|
||||
## Judges' comments:
|
||||
### To build:
|
||||
|
||||
make dlowe
|
||||
|
||||
### To run:
|
||||
|
||||
./dlowe -<n_iterations> corpus1/ [...] corpus0/ < start.net > end.net
|
||||
|
||||
NOTE: The directory entries MUST end in a /
|
||||
|
||||
./dlowe file [file ...] < trained.net
|
||||
|
||||
### Try:
|
||||
|
||||
Download the dlowe-aux-data.tar.bz2 file from:
|
||||
|
||||
curl -O http://www.ioccc.org/2011/dlowe/dlowe-aux-data.tar.bz2
|
||||
|
||||
tar -jxf dlowe-aux-data.tar.bz2
|
||||
|
||||
./dlowe dlowe.c < ioccc-winlose-trained.net
|
||||
./dlowe dlowe < ioccc-winlose-trained.net
|
||||
|
||||
./dlowe xor-0/00 < xor-trained.net
|
||||
./dlowe xor-0/11 < xor-trained.net
|
||||
./dlowe xor-1/01 < xor-trained.net
|
||||
./dlowe xor-1/10 < xor-trained.net
|
||||
|
||||
./dlowe hint.html < english-trained.net
|
||||
|
||||
./dlowe dlowe-aux-data.tar.bz2 < png-trained.net
|
||||
|
||||
### Selected Judges Remarks:
|
||||
|
||||
This entry is a ghoulish example of a brain (dead?) neural network classifier.
|
||||
It comes with a set of corpora for you to try.
|
||||
|
||||
#### ioccc-winlose-trained.net
|
||||
|
||||
You can create your own trained networks. For example, we trained a network
|
||||
on the C code of entries that were supplied to the 20Th IOCCC.
|
||||
|
||||
The ioccc-winlose-trained.net was trained using the earlystop.pl tool:
|
||||
|
||||
rm -f ioccc-winlose-trained.net
|
||||
./earlystop.pl ioccc-winlose-trained.net ioccc_won_training/ ioccc_lost_training/ ioccc_won_test/ ioccc_lost_test/
|
||||
|
||||
where:
|
||||
|
||||
ioccc_won_training/
|
||||
1/2 of C code of the 20th IOCCC winners (except for dlowe.c from 2011)
|
||||
plus 1/2 of the C code winners from prior IOCCC contests
|
||||
|
||||
ioccc_won_test/
|
||||
1/2 of C code of the 20th IOCCC winners (except for dlowe.c from 2011)
|
||||
plus 1/2 of the C code winners from prior IOCCC contests
|
||||
|
||||
ioccc_lost_training/
|
||||
1/2 of C code from entries that that did not win
|
||||
|
||||
ioccc_lost_test/
|
||||
1/2 of C code from entries that that did not win
|
||||
|
||||
NOTE: The above directies were not shipped and the C code from
|
||||
entries that that did not win was flushed to /dev/null.
|
||||
|
||||
The earlystop.pl tool wrote:
|
||||
|
||||
training 1 to 1001...
|
||||
training 1001 to 2001...
|
||||
prior error = 56.746974766043; current error = 58.095691135305
|
||||
|
||||
The resulting ioccc-winlose-trained.net file was included in the
|
||||
dlowe-aux-data.tar.bz2 file in the above mentioned URL.
|
||||
|
||||
The dlowe.c was explicitly excluded from this trailing set. So this test is interesting:
|
||||
|
||||
./dlowe dlowe.c < ioccc-winlose-trained.net
|
||||
|
||||
The match of dlowe.c to the lose/win trailed network was:
|
||||
|
||||
dlowe.c 0.125108
|
||||
|
||||
NOTE: The non-artificially intelligent decisions of the IOCCC judges include random
|
||||
quirks and their non-linear judgment calls not found in the ioccc trained network.
|
||||
Just because C code is or isn't scored highly by this network does not
|
||||
mean it will win or lose a future IOCCC. On the other hand ... it just might!
|
||||
|
||||
#### other data sets to try
|
||||
|
||||
While we cannot supply you with the losing IOCCC source code, the author did supply a number
|
||||
data sets on which to test:
|
||||
|
||||
We created the english-trained.net as follows:
|
||||
|
||||
./dlowe -8000 english-1/ english-0/ < /dev/null > english-trained.net
|
||||
|
||||
english-0/
|
||||
Non-English (French) text
|
||||
english-1/
|
||||
English (non-French) text
|
||||
|
||||
We created the png-trained.net as follows:
|
||||
|
||||
./dlowe -8000 png-1/ png-0/ < /dev/null > png-trained.net
|
||||
|
||||
png-0/
|
||||
Non-png (gif) images
|
||||
png-1/
|
||||
png (non-gif) images
|
||||
|
||||
We created the xor-trained.net as follows:
|
||||
|
||||
./dlowe -8000 xor-1/ xor-0/ < /dev/null > xor-trained.net
|
||||
|
||||
xor-0/
|
||||
Data that XORs to 0
|
||||
xor-1/
|
||||
Data that XORs to 1
|
||||
|
||||
## Author's comments:
|
||||
# Synopsis:
|
||||
This is an artificially intelligent judging tool to help the IOCCC judges.
|
||||
Here's to shorter, more frequent contests!
|
||||
|
||||
# Description:
|
||||
This is a multilayer perceptron (a feedforward artificial neural network)
|
||||
which can be trained, using on-line backpropagation, to classify input files.
|
||||
|
||||
It has a fixed topology of 2^16 input neurons, 6 hidden neurons and 1 output
|
||||
neuron.
|
||||
|
||||
The neurons' activation function is the logistic function 1 / (1 + e ^ -x).
|
||||
|
||||
## Classifying
|
||||
|
||||
./dlowe file [file ...] < trained.net
|
||||
|
||||
To classify files, one specifies a trained network (on stdin) and one or more
|
||||
files to classify. The program will output one line per successfully-classified
|
||||
file to stderr, with the filename and the classification: a number between 0
|
||||
and 1.
|
||||
|
||||
The interpretation of the classification number depends on how the network was
|
||||
trained, but it's geared toward interpretation as a *probability* or a
|
||||
*confidence*.
|
||||
|
||||
## Training
|
||||
|
||||
./dlowe -<n_iterations> corpus1/ [...] corpus0/ < start.net > end.net
|
||||
|
||||
NOTE: The directory args must end in a / for them to work.
|
||||
|
||||
To train a network, one specifies a starting network (on stdin), two or more
|
||||
corpora (directories containing training data), and the number of training
|
||||
iterations to run. The program will write some progress data to stderr and,
|
||||
when it's done, will serialize the updated network to stdout.
|
||||
|
||||
If no input network is given, a random new network will be generated as a
|
||||
starting point (i.e. provide an empty stdin).
|
||||
|
||||
The first corpus will be assigned a target value of 1. The last will be
|
||||
assigned a target value of 0. Intervening directories (if any) will be assigned
|
||||
intermediate target values.
|
||||
|
||||
The learning rate is hard-coded as 0.3. No momentum factor is used.
|
||||
|
||||
### png corpora
|
||||
|
||||
The png-1 corpus was obtained by manually scraping the first results from an
|
||||
https://images.google.com search for "obfuscate filetype:png".
|
||||
|
||||
The png-0 corpus was obtained by manually scraping the first results from an
|
||||
https://images.google.com search for "obfuscate filetype:gif".
|
||||
|
||||
Results:
|
||||
|
||||
* using these corpora as a training set
|
||||
* using several hundred random *.gif and *.png files from my home computer as
|
||||
a test set
|
||||
* using the included "earlystop.pl" script, which stopped after 67000 iterations.
|
||||
* interpreting output of "> 0.5" as "probably a .png" and "< 0.5" as "probably
|
||||
a .gif"
|
||||
* produced a network with about 87% accuracy
|
||||
|
||||
### english corpora
|
||||
|
||||
The english-1 corpus was obtained by manually scraping the first results from
|
||||
a https://www.google.com/webhp?lr=lang_en search for "paris filetype:txt".
|
||||
|
||||
The english-0 corpus was obtained by manually scraping the first results from
|
||||
a https://www.google.com/webhp?lr=lang_fr search for "paris filetype:txt".
|
||||
|
||||
Results:
|
||||
|
||||
* using these corpora as a training set
|
||||
* using several dozen additional text files (obtained by the same methods)
|
||||
as a test set
|
||||
* using the included "earlystop.pl" script, which stopped after 10000
|
||||
iterations.
|
||||
* interpreting output of "> 0.5" as "probably english" and "< 0.5" as
|
||||
"probably french"
|
||||
* produced a network with 100% accuracy
|
||||
|
||||
### xor corpora
|
||||
|
||||
The xor-1 corpus consists of two files containing '01' and '10' respectively.
|
||||
|
||||
The xor-0 corpus consists of two files containing '00' and '11' respectively.
|
||||
|
||||
Results:
|
||||
|
||||
* using these corpora as a training set
|
||||
* takes about 8000 iterations to learn xor to within a tolerance of <0.01
|
||||
|
||||
# Limitations
|
||||
The program can't tell you anything meaningful about files with less than two
|
||||
bytes in them (I'm looking at you, smr.c!)
|
||||
|
||||
You must include the trailing directory separator on training directories
|
||||
(this allows the program to be portable without wasting precious bytes on
|
||||
figuring out how to concatenate directory and file names...)
|
||||
|
||||
Serialized network files are only portable between systems with the same
|
||||
floating-point representation and endianness.
|
||||
|
||||
Making sure not to overfit the network to the training data is a bit of a
|
||||
black art. I have enclosed 'earlystop.pl', a wrapper script that implements
|
||||
a simple 'early stopping' algorithm; other techniques are possible.
|
||||
|
||||
Bad input (e.g. nonexistent files, non-numeric number of iterations, etc.)
|
||||
tends to result in empty output.
|
||||
|
||||
Given exactly one corpus, the program will crash or produce garbage.
|
||||
|
||||
Leaks memory and file descriptors while processing files.
|
||||
|
||||
Will crash and die horribly if it runs out of memory.
|
||||
|
||||
The Microsoft C compiler doesn't provide a dirent API, so to get this working
|
||||
on a Windows system you'll need cygwin+gcc (tested) or a dirent compatibility
|
||||
library (untested, but they do exist).
|
||||
|
||||
Backpropagation doesn't always converge: if you play with this long enough,
|
||||
you'll eventually have a training session that completely fails to converge.
|
||||
|
||||
# Obfuscation:
|
||||
|
||||
Zombies! (Since neural networks are modeled after BRAINS, ya know? And
|
||||
corpus sounds a lot like corpse. And I have 4- and 7-year-old kids ;)
|
||||
|
||||
Neural networks are interesting, and while the math isn't terribly difficult,
|
||||
their behavior is difficult to fully understand.
|
||||
|
||||
Had some extra space in one of the major data structures, and it seemed a
|
||||
shame to waste it.
|
||||
|
||||
Similarly, I think three file pointers ought to be enough for anyone.
|
||||
|
||||
Lots of magic numbers expressed in various ways.
|
||||
|
||||
Random abuse of random C trivia.
|
||||
|
||||
Some textual changes were made to maximize the classification of this program
|
||||
by a network trained to recognize ioccc winners ;)
|
||||
|
||||
... but mostly zombies!
|
||||
|
||||
--------------------------------------------------------------------------------
|
||||
<!--
|
||||
(c) Copyright 1984-2012, [Leo Broukhis, Simon Cooper, Landon Curt Noll][judges] - All rights reserved
|
||||
This work is licensed under a [Creative Commons Attribution-ShareAlike 3.0 Unported License][cc].
|
||||
|
||||
[judges]: http://www.ioccc.org/judges.html
|
||||
[cc]: http://creativecommons.org/licenses/by-sa/3.0/
|
||||
-->
|
||||
Reference in new issue
Block a user