Baker, Christopher
CEB1%%CEC
	P3:  	left a core file, blog ok, check spelling, one example?
		wow, you gave up too soon, and no boxes.  skin is not so bad.
		this is like a 9 for skin and a 0 for blobs.
		directory looks ok.  cs102 java subdirectory is always bad.
		i'd think better if the blob detector notes were in the blog.

	P2:	scally-fuller-baker, no blog, defeats 4, ties 1
		shallow strategy

	P1:	baker-schoen, 5% every 7 seconds and an analysis of bd.
		my notes say your behavior was good.

Chen, Kevin
kpc3%%cec

	P3:	k9 dir seems a bit empty -- no examples.  like the awk --
		looks like it was written by machine -- you don't need
		"declarations" you know...  ouch, wolf dir has a core file.
		hey, i'd rather you linked me than see pretty parchment
		html background.  this is a 10 presentation (for the
		tech description and pics, not for the html, which is nice).
		peformance is only 7ish though.  you have a face-size
		constraint, but still find smallish "faces".

	P2:	chen-eder-harris, group blog, defeats 1?, ties 1
		plays only as red, strategy shows some thought

	P1:	chen-eder, accepts anything over 50%, what are "good pairs"?
		my notes have good comments.

Cheng, Nicholas
NAC1%%CEC

	P3:	hey, i liked the face on the page so i can learn names!
		i like the NR step 1 idea.  source looks fine.  face
		detection is only ok because skin detection is only ok.
		this is a good effort, though.

	P2:	lin-cheng-halley, source code is gz'ed java, lots of
		functions here, bmp screen shot ouch, defeats 9,
		was in final.  gotta like this effort.

	P1:	lin-cheng-halley, models mood... bmp
		screen shots again.  looking at code:  recalculates
		mood on each waking...  mood model is simple,
		code is substantial.
		my notes say your behavior is ok.

Chernoff, Rebecca
RMC2%%CEC

	P3:  	wow, another person leaving a core file for me.  is it
		my unraw that creates these things?  huge log files... hmm...
		wolf blog is unimpressive, though the first transformations
		sound good.  nothing in k9?  should have dumped the output			of face to a file so we could figure out why cjpeg doesn't
		like it.  a bit of a punt on this one, but i can feel
		your frustration.  also, the filemod times suggest you
		were in a hurry, or ftp'ing a bit.

	P2:	chernoff-dewulf, defeats 13, ties 2, one of the four
		best players, strategy itself is not too intricate, 
		writeup on mtd's account.

	P1:     chernoff-clark:  happiness rises with time?  assuming
		things go as expected, i assume.  i have to read the
		code... (that's never a good thing to say about the
		blog)... seems ok.  my notes say your perf was good.

Clark, Robert
RPCLARK%%ARTSCI

	P3:  	yes, you set the bar for performance for now.  
		i especially like the ability to find big foreground
		faces and small background faces in the same image.
		profiles are impresive too.  i think the big hand on the left
		could be filtered, and it's too bad you didn't get
		the segments joined in 2639.  i think the long drapes
		could be filtered, too.  you should say what you mean
		by "recursively" anything.  is the power of this approach
		your human picked rgb region?

	P2:	freestone-clark, defeats 10, ties 3.  approach is novel.

	P1:     chernoff-clark:  happiness rises with time?  assuming
		things go as expected, i assume.  i have to read the
		code... (that's never a good thing to say about the
		blog)... seems ok.  my notes say your perf was good.

DeWulf, Michael
MTD1%%CEC

	P3:	wow, another student who "declares" variables at the top.
		i kind of like the way you did it.  you and others should
		not indent with tabs.  this is good work, mike.  i'm a bit
		curious about the double dewulf stamping in the upper left.
		it makes me think you used photoshop, cause i don't see
		anything in the code to produce it.  also, why java source
		and gawk source?  the algorithm is sound and the description
		is good.  the work on k9 looks pretty bad.  i guess i need
		to see more of the development path using the java code.  
		thank you for the footnote about clark's code.  i see there
		is almost no similarity.  this is definitely an extra-base
		hit.

	P2:	chernoff-dewulf, defeats 13, ties 2, one of the four
		best players, strategy itself is not too intricate, 
		writeup on mtd's account.

	P1:	i am looking at some reasonable claims for behavior,
		and code in mike-jason/dewulf... this looks like a 
		reasonable attempt.

Eder, Derek
DWE1%%CEC

	P3:	you have a proj3_blog.txt in a blogs dir without
		your name in the filename...  uh, another YUV guy.  
		in past years, just using YUV was enough to get an A...
		oh how the expectations rise even though people think the
		standards are falling.  this blog is very good reading.
		200 candidates sounds like a lot.  you should have included
		your output.  why not choose the first box arbitrarily
		when there are overlapping ones?  i am not so impressed
		by the face finding!  writeup 9, performance 6.

	P2:	chen-eder-harris, group blog, defeats 1?, ties 1
		plays only as red, strategy shows some thought

	P1:	chen-eder, accepts anything over 50%, what are "good pairs"?
		my notes have good comments.

Fan, Amy
AF2%%CEC

	P3:	yeesh, another core file.  k9 blog so far nothing 
		to impress.  i don't see anything on wolf...

	P2:	team fan, late, no teammates, doesn't look too bad,
		how does it play?  no score.log seems to have team-fan 
		in it.

	P1:	fan-hussain-krettek:  5% every 2 min, analysis of force,
		what is a "better evaluation of average"?
		my notes have good comments.

Franco, David
DF1%%CEC

	P3:	k9 seems like a decent start.  another copy of finder.awk...
		well, we have an honest blog with a good picture of what
		was accomplished.  you get a 10 for clarity, but a solid
		"yuck" for performance.  actually, the skin was not so bad,
		but you need to try to filter it and draw some boxes.

	P2:	ratkowski-franco.  loses to all.  26k of perl, why is it
		so bad?

	P1:	oh yes, ratkowski-franco.  scores bundles, tracks mood.
		my notes include some weird behaviors -- all due to bugs?

Freestone, Amy
AMF4%%CEC

	P3:	nice blog, reasonable approach.  fascinating source code...
		should have tried more images, though i guess the machine
		was loaded dec 14, 3pm.  another core file sigh.  the
		performance is not great, but this is also an extra base
		hit.  something like a solid 8/10.  A-.  however you want
		to say it.  met expectations.

	P2:	freestone-clark, defeats 10, ties 3.  approach is novel.

	P1:	freestone-khatri, nice progress chart, 28% above the
		left corner?  separate rates for accept and offer concession.
		predefined schedule.  my notes say ok.

Froderman, Jason
jkf1%%cec

	P3:	yo yo, you know i can't read your php if you don't link
		to it.  the montage of skin still disturbs me.  but the
		skin detector is really not so hot, and the face boxes
		are nowhere to be found.  this is an A if it were a
		web class ;) but pretty much a bunt single (or even
		a sacrifice bunt) as ai goes.  i'll assume that if i could
		read the learning method in the php, it'd be a single.

	P2:	froderman-krettek, defeats 14, ties 1, only US Womens'
		Soccer Team is better.  a reasonable strategy too.
		code is even readable.

	P1:	a froderman only after team bust, i am looking thru code 
		for strategy...  seems to be an i-learned-gawk project.

Fuller, Benjamin
BMF1%%CEC

	P3:	another core file and a very nice 4-line awk file.  
		another fleck-forsyth follower.  what's with the function
		name?  no actual blog?  ok, good contact line drive out.
		i see you got started, but didn't get too far.

	P2:	scally-fuller-baker, no blog, defeats 4, ties 1
		shallow strategy

	P1:	ouch -- offers above equilibrium increase goodness?
		seems to be a weak criterion... looking at harris-fuller,
		sort of a discourse model of happiness.  this is ok.
		my notes not so good...

Halley, Philip
pjhalley%%artsci

	P3:	i see this 'my%20stuff%20is%20in%20the%20blogs%20dir'
		but nothing there.  let's assume it's a cruel mistake
		and we'll fix the grade later.  oh, the old old blogs dir.
		there it is.  i see some hefty c program and some big
		resolution images.  the performance is actually pretty
		darned good.  9+/10 maybe.  it looks like all you did 
		was xform the pixels, do a simple test for skin, 
		and do some kind of density check.  i'd like to have
		this in text instead of having to read it in c.  like
		you doubled and got thrown out trying to go for third.
		also known as a very exciting sac fly.

	P2:	lin-cheng-halley, source code is gz'ed java, lots of
		functions here, bmp screen shot ouch, defeats 9,
		was in final.  gotta like this effort.

	P1:	lin-cheng-halley, models mood... bmp
		screen shots again.  looking at code:  recalculates
		mood on each waking...  mood model is simple,
		code is substantial.
		my notes say your behavior is ok.

Hamer, Timothy
TDH1%%CEC

	P3:	ok, we have a nice writeup of a skin-but-not-face
		result.  i'm looking at some depressing awk code here.
		it's probably not a good idea to carry $ references
		far away from the getline.  it's always better to
		read into an array before processing, though i suppose
		you have a nose for inline computations... 
		what?  you don't give me the original pics?  hmm.  
		the skin-detecting results were actually quite good
		i thought.  we have an infield single here, but the 
		batter didn't hustle to first and maybe got thrown out.
		no explanation for not using our pics?  i don't mind,
		i'm just curious.  well, not that curious.

	P2:	hamer-schoen, no blog, proj2, tourney winner, actually
		a TERRIBLE program that defeats only two others, code
		seems to reveal a well-thought-out strategy.  like
		rat, we have lots of code and poor performance, hmm...

	P1:	scally-hamer, 3x your security!  that's not going to work
		all the time... sec level strange.  please apostrophize your
		contractions.  how do you schedule your acc level drops?
		your neg is also chmod'ed wrong.
		my notes say the code didn't work.

Harris, Brandon
BBH2%%CEC

	P3:	hey, apostrophize your possessives please.  
		i like the alogrithm.  esp. the part about marking
		non-skin.  genuis!  the method has real promise and
		i think you should've shown it on all of the images.
		some of these function names should be tarantino movies.
		home run.  not the kind you put a tape measure to, but
		definitely cleared the fences.

	P2:	chen-eder-harris, group blog, defeats 1?, ties 1
		plays only as red, strategy shows some thought

	P1:	ouch -- offers above equilibrium increase goodness?
		seems to be a weak criterion... looking at harris-fuller,
		sort of a discourse model of happiness.  this is ok.
		my notes not so good...

Horvath, Alexa
AKH2%%CEC

	P3:	for someone who couldn't djpeg on the last 
		demo day, this is great stuff.  i like the low-false-positive
		tight-threshold output that is very picky about what it
		colors.  your indenting and commenting needs a bit more
		discipline (sorry -- you know it doesn't matter what
		your code looks like in this class, but i notice these 
		things when i am looking around).  if i can find a blog,
		i'm thinking home run... ahh ah ah, no it's off the wall
		for a double... wait -- there's the blog!  ok, we must
		have had fan interference.  it's out of the park after
		all.  i would just have liked to have seen the effect
		of skin-id and relaxing, so i could guess where the power
		of the method comes from.  it actually looks like #3 is
		your main advantage here.

	P2:	horvath-simon, scorebaby, defeats 4, ties 3, ouch a core
		file, claims to be rule based, well described strategy,
		and look at that code.  once again, lots of brains, not
		a great result.  this is why i don't like soccer as an
		ai project... this would never happen with stratego...

	P1:	horvath-simon, linear drop with time. 

Hussain, Rumana
RSH2%%CEC

	P3:	yes, i see the similarity to eder-boxes.  yeah yeah,
		more toddler-can-do-it commentary.  the source code
		seems reasonable enough and definitely looks like it
		took effort.  i've done this before where i've left
		all sorts of overlapping boxes in the image.  you need
		to bring yourself to that final decision -- is it or
		isn't it a face?  which box is it?  i actually think you
		took this farther than eder.  or perhaps you just
		ran on better images.  I'm thinking bunt single here.

	P2:	hussain-khatri, spry, ouch another core file, there is
		quite a bit of logic in the code and well described, 
		and yes, it defeats 1 and ties 2.

	P1:	fan-hussain-krettek:  5% every 2 min, analysis of force,
		what is a "better evaluation of average"?
		my notes have good comments.

Khatri, Ashish
ahk1%%cec

	P3:	hey that's not a radius.  that's a square.  you
		could approximate the inscribed circle by using a conditional
		to skip the corners...  apostrophize your possessives please...
		this is A work and A level results.  i'd need to see more
		examples... you should follow horvath's lead and show
		me all the bad ones, too.  i'm thinking this is the
		off-the wall double.  maybe a triple...

	P2:	hussain-khatri, spry, ouch another core file, there is
		quite a bit of logic in the code and well described, 
		and yes, it defeats 1 and ties 2.

	P1:	freestone-khatri, nice progress chart, 28% above the
		left corner?  separate rates for accept and offer concession.
		predefined schedule.  my notes say ok.

Krettek, John
jk3%%cec

	P3:	wow, i can put a face to this name.  wow, you can't
		even test me on it!  this code is heavy.  what is a line
		like this doing?  all_lines_new = (all_lines_new line_new"\n");
		are you building recursively parenthesized structure?
                jimminy.  you're taking 455 concurrently, right?  i
                wouldn't call this a lifecycle, which has other connotations.
		ouch these results are not so good.  skin is all over,
		like a bad csi episode.  and even with the nuisance skin,
		boxes could be better (apply those constraints, man!).
		and you did what i do when i have a bad algorithm -- i
		keep running it on more images to see if it's gonna get
		better... your fellow students must hate the cycles you
		were eating, unless, as i suspect, you kept yourself on
		a cec machine.  why so bad?  well, the boxes that are
		adjacent should be merged, or the larger one taken.  i see
		you actually put some thought and work into the skin
		detector.  i've been there before.  i think the normalization
		of the images might be important in training a skin detector.
		let's call this a 2-rbi single where you got thrown out
		at second.

	P2:	froderman-krettek, defeats 14, ties 1, only US Womens'
		Soccer Team is better.  a reasonable strategy too.
		code is even readable.

	P1:	fan-hussain-krettek:  5% every 2 min, analysis of force,
		what is a "better evaluation of average"?
		my notes have good comments.

Lin, Thomas
tkl2%%cec

	P3:	sorry about the wolf downtime.  it was almost certainly
		something your fellow students were doing to the machine!
		now why would you delete your project 1 blog?  jimminy.
		aha, a directory with, um, a core file in it.  there's
		a reasonable-looking face program in here.  looks like you
		got stuck on the skin-finder, which is just awful.  forget
		the stanford name -- i think this was stanford, wyoming.
		the skin code you got from them stunk.  i'm calling this
		one a hit-by-pitch.

	P2:	lin-cheng-halley, source code is gz'ed java, lots of
		functions here, bmp screen shot ouch, defeats 9,
		was in final.  gotta like this effort.

	P1:	lin-cheng-halley, models mood... bmp
		screen shots again.  looking at code:  recalculates
		mood on each waking...  mood model is simple,
		code is substantial.
		my notes say your behavior is ok.

McCarthy, Stephen
SJM1%%CEC

	P3:	jank.us?  you're kidding, right?  and it's offline.
		probably your laptop and you're gone for the holidays.
		ok, i see an impressive c program here, but no output.
		this is an I.

	P2:	mccarthy-satkin-tipton, US Women's Soccer Team, enough said.

	P1:	mccarthy-tipton-satkin, will time out instead of responding
		with force... looks like dumb behavior when it arises, 
		but actually is smart, i guess.  will slow down rate of
		concession.  will threaten.  let's see this actual rate in
		the code... this is the crucial test?
		if(sqrt(secondsPassed/totalTime) > (om/totalOffers))
		hmm...  my notes on your behaviors ok.

Ratkowski, Richard
RRR1%%CEC

	P3:	aha, perl, i have missed your ugliness.  this is not
		bad performance.  i'm more worried about the false
		positives than the false negatives.  ouch, a core file.
		we seem to have lots of jpg files that are actually
		ppm's ...  i like knowing what i can delete first to
		recover space at the end of the semester ... method?
		please apostrophize your possessives.  please don't
		let a process be ran, and don't let a picture be within
		10% of the border of a picture.  i didn't quite understand
		how you got http://cec.wustl.edu/~rrr1/newMultiSquare.jpg,
		which seems to be interesting.  otherwise, an A for your
		method(s).  and the writeup is actually good despite the
		too-busy-to-edit english.  another example that clarity
		is not about good grammar or good spelling (though the
		latter do help).  i'll call this a solid rbi double.

	P2:	ratkowski-franco.  loses to all.  26k of perl, why is it
		so bad?

	P1:	oh yes, ratkowski-franco.  scores bundles, tracks mood.
		my notes include some weird behaviors -- all due to bugs?

Satkin, Scott
SBS1%%CEC

	P3:	wow, this is an uninformative blog!  ok, it's not
		so bad, as the comment on the right is good and the
		examples are good.  but the problem decomposition
		statement -- i mean, how else would you do it?
		you should try animating the a,b,c jpg's in your k9
		04-03 directory.  the color changes are very entertaining
		(though i take it that only the successive boxing
		is important).  for some reason your msu/hmc
		skin detector is better than others who used the latter.
		are you limiting the faces to three by hand somehow?
		the code looks good.  very inviting.  wish i had time
		(or interest!) to read it carefully.  i mean, i keep
		re-reading it and i don't see where the a,b,c come
		from, nor why the colors change each time.  this is
		a solid double also.

	P2:	mccarthy-satkin-tipton, US Women's Soccer Team, enough said.

	P1:	mccarthy-tipton-satkin, will time out instead of responding
		with force... looks like dumb behavior when it arises, 
		but actually is smart, i guess.  will slow down rate of
		concession.  will threaten.  let's see this actual rate in
		the code... this is the crucial test?
		if(sqrt(secondsPassed/totalTime) > (om/totalOffers))
		hmm...  my notes on your behaviors ok.

Scally, Andrew
ags1%%cec

	P3:	scally!  i can't find the blog.  aha, got it.
		i see some good looking code here, but not a lot of
		images that is has been tried on.  it's like finding
		a bat, and no baseballs.  or a kfc bucket, and no chicken.
		faceFinal.jpg is even quite promising.  clearly an
		out-of-time attempt.  i'm calling this strike three
		on the pitcher with the bases loaded and one out.
		you could've moved the runner over and didn't.  don't worry.
		cardinals just signed mulder and he can't hit either.
		you also seem to have the "do the computation as i read in
		the file" affliction.  it's a good way to preserve
		bandwidth if you are programming network devices, but in ai,
		we usually want to touch the data a few times.  read it in
		first, then start working on it.

	P2:	scally-fuller-baker, eh blog, defeats 4, ties 1
		shallow strategy

	P1:	scally-hamer, 3x your security!  that's not going to work
		all the time... sec level strange.  please apostrophize your
		contractions.  how do you schedule your acc level drops?
		your neg is also chmod'ed wrong.
		my notes say the code didn't work.

Schoen, Luke
LRS1%%CEC

	P3:	face blobulation?  sounds like botoxulation.
		you should say how you tweaked the YUV/shiny criteria.
		you know, for all the great talk, the results are so-so.
		i'm going to call this a derek jeter single.  it's just
		a single, no matter how amazing the athlete/algorithm
		behind the hit.  and the yankee fans and ny press go WILD
		in awe of jeter's amazing single!!!!

	P2:	hamer-schoen, no blog, proj2, tourney winner, actually
		a TERRIBLE program that defeats only two others, code
		seems to reveal a well-thought-out strategy.  like
		rat, we have lots of code and poor performance, hmm...

	P1:	baker-schoen, 5% every 7 seconds and an analysis of bd.
		my notes say your behavior was good.

Simon, Joshua
JAS6%%CEC

	P3:	josh -- the link is locked up.  we'll call this
		an I for now.  actually, there may be enough info on k9
		to assign a good grade... the images show only a so-so
		performance (as above, so-so, not so-sa), and i don't see
		too much in the cgi's... 

	P2:	horvath-simon, scorebaby, defeats 4, ties 3, ouch a core
		file, claims to be rule based, well described strategy,
		and look at that code.  once again, lots of brains, not
		a great result.  this is why i don't like soccer as an
		ai project... this would never happen with stratego...

	P1:	horvath-simon, linear drop with time. 

Tipton, Eric
ECT1%%CEC

	P3:	rgb seemed to work better than yuv for this project
		anyway!  ha ha ha.  pixel bashing is all about wasting your
		life tweaking parameters and re-running the data set.  i
		wish i had a year back of my life that i've spent pixel-bashing.
		this is infield-single performance.  like 5/10.  i think
		the issue is that your rgb range is too permissive of white
		areas that are skin-under-flash-like.  actually most of the 
		skin is not under flash.  it's still a hit and not an out.

	P2:	mccarthy-satkin-tipton, US Women's Soccer Team, enough said.

	P1:	mccarthy-tipton-satkin, will time out instead of responding
		with force... looks like dumb behavior when it arises, 
		but actually is smart, i guess.  will slow down rate of
		concession.  will threaten.  let's see this actual rate in
		the code... this is the crucial test?
		if(sqrt(secondsPassed/totalTime) > (om/totalOffers))
		hmm...  my notes on your behaviors ok.