You could play it with humans but it would take months or years for a tournament

like Risk?

lol

yeah lol Risk is the perfect counterexample - it takes months to play and humans still play it (although most not more than once...)

I don't use any history yet. It's just there to keep contenders guessing :)

I tried to play it sometimes but never finished

Risk is amazing :)

My idea is much much simpler than Risk but will take much longer due to being played on a large grid

@MartinBüttner Does this opinion mean you have played Risk on a computer and not had to count up the number of pieces in different regions every single move...?

no, I've only played it on a board

You clearly have a lot of patience

:D

It also doesn't work if there are any living beings in your house...someone will step on it eventually

it's not like you can't remember how many units you got last round and just check what changed.

@feersum we have a table.

@MartinBüttner lol yeah I may not have been playing it as efficiently as you

still doesn't work with cats ;)

or someone will bump into the table

Someone needs to invent magnetic wall Risk

(although people will still brush against it)

@Geobits Does confusion still make you punch yourself if you provide an invalid command? (i.e. I want my Codemon to not do anything that turn)

@trichoplax I've actually got this 2210 (or something) A.D. edition... you can just play classic risk with it, but it also has its own crazy ruleset... the rule are super interesting, but I haven't yet figured out a way to play that unless you're with really hardcore gamers, because there's so much strategy that the game is basically reduce to 5 reaaaally long rounds.

one thing that's really interesting is that you can bid for turn order (each round) with some resource you have

I was an avid Risk player when I was a kid. I don't remember counting anything though. Just strive to get connected regions and bunch armies at the frontiers. Or have Risk rules evolved since the last century?

There goes my argument that some games are too drawn out for humans...

I think Risk always had new troops allocated based on existing numbers

(The version I played certainly wasn't from this century)

Did anyone ever play a card&board game called S.P.I.Vs?

That took a long time...

hi people

@MartinBüttner turning Risk into a wargame sounds like a strange idea to me. There are plenty of dedicated such wargames, while the main appeal of Risk is its simplicity IMHO.

@Pyraminx hi

what is up

no one ever chats on the sites I go to

this chat is usually reasonably busy (at least at this time of day)

When you guys aren't here it's half-half between dead quiet and occasional chatting though

the last message is always more than 2 hours ago on the sites i like

@Pyraminx Here it will depend on your time zone

I'm in UTC -8:00

Btw my rats are at 6000+ after 250 runs or so. Maybe it's just a fluke, but I keep my fingers crossed.

daytime and late evening in the UK and US tend to have people around. There are people around at other times too but not so many

@kuroineko I don't know where you get your motivation when you're already at the top :)

"Because it's an awesome question"

except Frenchmen with insomnia :). As for motivation, well it's just beating the problem. So much the better if I come on top, but that's secondary.

It's funny because no one actually talks about anything related to the site of the chat room. Everyone gets sidetracked to somewhere else.

Here we're about 50/50 on topic and off topic, which I think is a good balance

Freedom, freedom, freeedooom, yeah

bye guys i have to go

Bye for now :)

soon, not now

ok, now.

bye then

BYE!

About the GA challenge, I dream of someone posting a new answer that would mop the floor with my rats, so that I could learn or understand something new.

Nathan has a new GA challenge in the sandbox that I'm looking forward to seeing posted

@Geobits I'm running a mini Codemon tournament. How do I reduce the time limit?

@trichoplax I was in the discussion that lead to this (sandboxed) challenge, but I'm not convinced it could work.

nevermind I found it.

@kuroineko do you see a way of changing it to make it work?

If I did, I would have posted something. I have thought about it a bit, but came up blank.

Hi

hi there

Hi

I have another tiny javaScript question.

I have an html5 canvas that's detecting mouseUp events; if I drag the cursor off the canvas and then release the mouse button, it doesn't trigger

so I gave the HTML tag an id, "parent", and put in a line: parent.onmouseup = XZmouseUp;

is that bad form?

I have zero JS knowledge

have you tried adding an event listener to your document body?

that would catch the mouse unless you move it outside the browser's display area

no. I only need to detect mouseUp events outside of the canvas. What I did works, I just want to know if it's bad style somehow

lol @feersum

of course

prospective employers see my website, i'd rather not have it filled with horrible code any more than I need to

if you can't take delight in bad style, you do not belong in codegolf!

Do prospective employers see that you use PPCG?

i guess they could, i use my real name on stackexchange

so you are already doomed, might as well :)

i don't see how being on here is in any way negative :|

I was just joking because some people frown on golf

really?

Well I'm unemployed, and my last job was case worker in child protection, so I can't help you with employer-seducing tips. Your concern comforts me in my decision to leave the silly world of software industry, though.

Would you recommend against working in programming?

anyway, not a big deal, i just felt weird putting <html id="parent">

Wait, why do you need to give the html tag an ID?

scrolls up

@EricTressler I would certainly not name it "parent", regardless of what an employer might think of it

Oh, couldn't you just do document.documentElement.onmouseup = XZmouseUp?

does that work? let me try

yep, thanks

@Sp3000 Choosing a no-op does not preempt confusion. You still punch yourself :)

Does confusion have any affect on swapping?

Right now, I have a competition of about 20 completely random bots going on.

No, you can still swap as normal.

@trichoplax I have no authority to do any kind of recomendation, really. I was a real-time embedded software programmer until the NASDAQ collapse turned French industrial software business into a quagmire, and then I quit. Something tells me things have not changed for the better since then, so I'd rather live off state charity than go back there.

(maybe a confused Codemon should accidentally swap with the other team)

@kuroineko I see. I'm in the UK so not the same but I'm interested in opinions as I'm wondering about whether to learn to be a programmer of some sort

In my experience it beats working on an aircraft ramp. I don't know what you're comparing it to, though ;)

lol

@trichoplax well after having been a project manager for 3 or 4 years and a consultant for a couple more, the only thing I can say is that it's not a job for me. Too many asses to kick or lick, too many lies to tell for my taste. Trouble with French software industry is, you can't stay a mere programmer after you turned 35 or so. And being a manager and/or well-paid liar err... I mean consultant is an entirely different job.

I quit my job at a research lab when it turned into full-time programming of stuff I didn't care about

@kuroineko I've heard similar from others - "career progression" means moving away from what you enjoy. I'd be happy staying with programming for less money - I'm not looking for a lot

well trouble is, when you gather experience people won't hire you as a mere programmer anymore. When I asked to remain a simple programmer/designer, people looked at me with a kind of pity, as if I was mildly retarded or something.

People already look at me like that so perhaps I can work around it :)

@EricTressler Are you still interested in working in programming? Did it just put you off that one job or programming as a job altogether?

hahaha that would be nice. However, from my experience, you'll have to look hard for that kind of humor in software industry. The dominant ideology is rather a kind of machismo that leaves precious little room for dreamers or losers.

i'd like a job that has some design and analysis components as well

for 3 years, that job was pretty good, but the last year I was there, it was almost entirely coding, and it stressed me out too much

implementing poorly-described algorithms in literature

Yes I like the idea of a range of different tasks

my idea was to pick a domain where experience makes you valuable enough to force your hierarchy to give you some freedom.

The trouble in domains like web design is that originality is intrinsically bad. What makes money is to copy the last whim of fashion in the quickest and dirtiest way.

Thanks for the insight. I certainly want to be able to have originality, so I'll bear that in mind

I'm going to sleep now. Good night all

good night, young padawan

lol :)

Meh. I found a good strategy that gets about 50/50 with Kids and Campers, only to have its scored tanked by those damned metapods.

I'm still running a random simulation.

10 rounds, 50 random bots.

To see what happens.

Hmm. I don't know if 10 rounds is enough to overcome the inherent variance, but it'll be interesting to see what it comes up with.

The goal being to determine what move/type combinations are inherently better.

Each move/type combination could occur in multiple bots.

How are you deciding which move to use when? Tossing dice?

Each single bot's score will be pretty random, I guess, but a trend such as "teams with poison score higher" should be possible to find.

Throwing dice as of right now.

Stats evened out or random?

even stats

The tournament just finished.

5 seconds ago

Now take the winner and run 50 slight variances on him :D

Against the current field

16		4x12x1x5X2x12x13x14X2x0x7x1
16		4x9x13x3X0x12x8x1X1x8x6x7
15		0x12x7x6X2x4x12x7X2x4x7x5
14		3x4x7x6X1x10x9x6X2x4x7x6
14		1x5x9x6X1x9x5x3X1x14x7x4
14		1x8x13x7X1x5x3x10X4x3x0x10
14		3x0x1x8X3x7x6x11X2x13x4x14
13		0x1x6x13X0x13x6x2X0x10x2x13
13		4x10x3x1X1x7x10x4X1x3x10x7
13		3x5x9x3X3x8x5x3X2x13x14x12
13		4x1x2x14X4x1x9x12X4x5x13x4
13		2x7x12x5X2x7x13x14X0x11x1x10
12		3x3x11x1X1x7x4x12X3x2x1x0
12		2x13x7x14X4x3x0x9X0x1x2x0
12		2x0x14x2X1x9x5x12X0x12x2x7
12		4x2x13x5X1x8x9x13X2x2x6x8
12		3x3x8x6X4x12x0x10X2x8x7x5

That's typeXmoveXmoveXmove*3, right?

yes

uppercase Xs split the numbers into three groups.

Right

I don't see much of a pattern, unfortunately. The most common among the top 3 looks like 12(Vine), for instance, and that doesn't make much sense at all :)

I'm worried that type differences play a large part.

If one type is more common, than any bot with move strong against that type will win more often.

True. The answer to that is 5000 bots instead :P

(with 50 rounds)

I'll set something up to run overnight.

Is there a detailed log file created?

That shows who beat whom?

If you want just the match results, run java Tournament 2, and to pipe it to a file, try java Tournament 2 > results.txt

That'll get rid of the turn-by-turn but still show "A beat B" and the running scores for each round.

So, I'm going to start a second random tournament.

I've gotten the Tournament 2 to work.

Well, redirecting it to a file should work. I'm on Ubuntu, but I believe it works the same on Windows if you're running that.

What's the easiest way to check to see if a move has used all of its uses?

Those are fields 8/9/10 (0-indexed) in each member's string.

It'll be 0 for "none left", positive for "x left", and negative for "unlimited".

If you pin one move to whatever their type's unlimited pure-attack move is, it might help mitigate your typing issues. With pure random, there's a high chance of not having any good attack (for that type).

Then the other two as random.

okay, I'm giving each Codemon their type's primary attack and setting it to use that attack if the random move has no uses.

Sounds a bit better than pure dice roll. Hopefully something clearer will emerge overnight ;)

how many bots and rounds?

A billion and two thousand :P

No really, I don't know how long your 10/50 took.

1.5 hours.

I think.

way too long

Damn, really?

10*50 is 250 battles.

Oh, that actually makes sense.... if they run out of moves they'll just sit staring at each other for 1000 turns :D

Instead of the more normal 20 or so turns a battle takes.

With your backup attack, it shouldn't take nearly as long.

I've been doing 100*6 (300 battles) in about 5-6 minutes.

Though they're more targeted.

I wonder if my IDE is the problem.

As part of AP CompSci, I use an IDE called BlueJ.

I dunno. I'm more inclined to think you're hitting the turn limit with a few bots. No IDE should slow it down 20x like that.

Unless you're running it in some crazy debug mode, I guess.

I reduced the turn limit to 50, in that tournament.

It's not that.

It's just slow.

Ah. Weird then.

Well, if you want it to run overnight assuming the same speed, you need about five times as many battles. 25 rounds *100 bots would do it.

If you suspect your IDE, you could try a short one on the command line to see if that helps. I'm out for now though, good luck!

bye!

a little over 100 bots and 25 rounds has started

Hey, I just finished the controller, and I'm looking to post this question: meta.codegolf.stackexchange.com/a/2571/20198

Can I get any last minute feedback?

Just checking - if someone sticks to a deterministic strategy and moves a bot away, leaving an open spot below me

Then I get to name them and they have to move back to the same spot?

yes

Interesting...

@Sp3000 Do you know anything about machine learning?

Er... little bit? How come?

I was thinking something like Multinomial Logistic Regression, Naive Bayes Classifiers or Markov (possibly hidden) models

What do you mean by scoring the inputs?

(also when in doubt, pulling up a machine learning library and throwing everything at it to see what's best is an option)

^ Did that in mathematica and it used "Markov" which I assume is a Markov chain

and say the input was "What's the time in Phoenix, Arizona?"

There would be some list of probabilities or scores

referring to how certain the algorithm is that it is asking for the function

like maybe it returns [1.8483 -> time, .235 -> date, .004 -> maps]

or something

I've got it working currently but it's using a VERY naive approach

Markov doesn't sound too bad, it'd be a bit like this

Do you have any training data? Like example inputs and correct responses

Are you asking for a yes or no, or for me to send example data?

The answer is yes

Just a yes/no

Then yes

(Well, it would be yes either way XD )

Ahaha k :P

In mathematica, I used the classify function and everything works well but I want to limit it to one programming language to reduce execution time and computational steps

and I don't want to re-make it in mathematica

Well in my limited machine learning knowledge I know there's tf-idf which can help you find which words are more significant than others

So something like "What" wouldn't score very well with that

mm I see, but again that would just score the words

I thought you wanted to score the words? Or did you mean something else?

No like

Sorry, I didn't explain it well

Basically, it should return the correct function. Like, if it is inferred that the user is asking the time, we should execute the time function. If they are asking for directions, then the maps function, etc.

So the classes for the model would be those functions

but it has to take the string input, and from that, infer what function the user is trying to call from the natural language input

How did Naive Bayes go? Out of curiosity (I don't know much about the Multinomial one)

(if I remember right that one's not very resource intensive)

Like how does it work or how did testing go? If you're talking about testing, then I haven't specifically tested it yet, just been looking around trying to find something suited before I spend the time implementing it

Otherwise, en.wikipedia.org/wiki/Naive_Bayes_classifier

I like how you automatically cover both cases of my ambiguous questions :P I meant how did testing go

But let's see, hm...

I was just wondering if that was getting annoying or not x)

Nope, it's very useful :)

Alrighty

Well if it helps, the most machine learning-y thing I've done is an assignment which was a basic spam filter, in which we removed stop words (like "is", "a", etc.) from the documents, used tf-idf to score each word then applied Naive Bayes to that

Which is a little more than I've done

I also tried some Porter stemming, but that one seems to be okay sometimes and not as good at other times

I have a basic understanding of the concepts and an interest but almost no experience with it

I was thinking of creating my own function where each function has certain keyword:offset pairs associated with it, and then mapping the words in the input over that, modifying the score for each function by that offset each time

But I would think that machine learning would have a better success rate and be less of a pain to upkeep

Probably, if it's natural language like that

Why does NLP have to be so difficult XD

I wanted to do it this semester but they cancelled it this year :(

Aw damn :c It's a really fun topic

@MartinBüttner you will like this rednuht.org/genetic_walkers

That reminds me...

@MartinBüttner You will like this youtube.com/watch?v=8vzTCC-jbwM (and half of this guy's channel)

@MartinBüttner no, you will like my link better

I do like your link though :P Watching people fall

@Sp3000 wait for your "Martin you will like <link>" day

@Sp3000 sooo negative. Its all about people trying to rise up as the generations go.

I'm watching the cars one now and they're all getting stuck in the same ditch

@trichoplax My first job was creating a computer version of a play-by-mail strategy war game. When I joined the company, the rule book was about 200 pages long. By the time I left, the requests for clarifications from the developers had made it grow to 300 pages. I'm not sure how long the original mail version took, but I would guess that a game lasted several years.

@Sp3000 that's great!

although not a very good advert for genetic programming :)

but then nothing is...

Well that guy's vids were actually what got me into genetic programming, so...

whistles

@Sp3000 oh! I thought it was established 15 years ago that GAs were basically pointless

@Sp3000 at best they implement stochastic hill climbing

Oh? For something that's useless there sure are a lot of papers still coming out on it...

@Sp3000 right..like neural networks pre 1995 :)

@Sp3000 I mean stochastic hill climbing is useful but rarely the best thing to do

(I just like them cos they're fun)

(To watch)

@Sp3000 I love anything with graphics too :)

Not even that - I did a genetic Hello World once just for the sake of seeing the chars change

:)

meh, I wish GolfScript didn't print the trailing newline

then I could do CH's challenge in 4

Ah :/

Funny how xnor took advantage of that in Python

at least I have beaten Optimizer for once :P

@Optimizer neat

@MartinBüttner not like u have not before

but that is clever

if GolfScript didn't print the trailing newline (or CJam treated the empty stack as q), I'd have ..n+ (or __S+ in CJam)

I still do not understand the TECO logic

how it works. .

I wonder if you can use o somehow to take advantage of the fact that squares are the sum of two triangular numbers

@Sp3000 haha, that's pretty amazing too

@Optimizer it's basically the same thing as mine, except that L is implicit

or sum of n odd digits

@MartinBüttner isn't he doing n^2 + 1 + n^2 then ?

no

V -> current buffer -> n^2

in JS you'd have an implicity s = "" and then his code would be console.log(s); s+= "1" ; console.log(s); (ignore the fact that log prints newlines)

1 -> 1 -> n^2 + 1

V -> current buffer = n^2 + 1

no, V only has N characters

how does V gets updated ?

with 1\

actually V isn't updated at all... from what I understand it's just a command with an implicit source for printing

so if V\1V prints 1 , V\1VV prints 11 , V\1VV\1 prints 111 and V\1VV\1V prints 111111 no ?

no

wait

V1\VV1\ still prints 11 (because there's no additional V)

If only CJam had infinite zeroes at the bottom like Prelude...

and then V1\VV1\V prints 1111 (because now the buffer contains two 1s)

@Sp3000 well, there's an active feature request to implicitly perform q if you try to operate on an empty stack

that would have helped, too

@MartinBüttner because its VV instead of V ?

I really don't understand where you're stuck on this :D ... if JavaScript didn't help here is the CJam translation: V -> Lo ... 1\ -> L'1+:L;

I am stuck because V prints the line content as is. I am not sure how you are interpreting it that its soo clear.

"the line in the text buffer". (i.e. L in my CJam translation) not the line of the source code

of course. let me again show you the text buffer after each step for N = 2

dude, just copy the CJam snippets together and put some eds in between

(that is, LoL'1+:L;Lo)

ohhhhhhhhhhhh

thats what you converted

are you taking "the text buffer" to mean STDOUT?

by ur understanding

V prints to STDOUT, right ?

\1 does not add to STDOUT ?

I'm pretty sure "the text buffer" is some piece of memory and "printing" goes to STDOUT... so V doesn't modify "the text buffer" at all

@Optimizer I'm pretty sure it doesn't, because how would V read from its own STDOUT?

its not at all clear.

@MartinBüttner that is what I understood

Did you guys have a question about the TECO?

Optimizer does :D

@feersum it can need some explanation

at least to me its totally unclear

the text buffer is basically a string

V prints it

\1 appends "1" to it

ok

@feersum so no extra newline as such ?

There are no auto-newlines anywhere

only the = command would put a newline if it is not in the string

@MartinBüttner just like in CJam. This is the equivalent I was thinking : _'1]s_ if _ worked with empty stack.

well that's not really reading STDOUT though

if _ worked with the empty stack you could do 4 bytes though

I wonder who will be the first to try posting the 1-byte HQ9+ solution?

oh

well that would hopefully be downvoted into oblivion

also, it's arguable that Q reads the source code

Yes, I assumed it's invalid because of that

@Lembik What general optimisation techniques don't implement stochastic hill climbing?

@PeterTaylor I can only name simulated annealing and genetic programming

And Great Deluge, which has a pretty cool name

@Sp3000 true :)

@Lembik I can only name those and actual hill climbing. But simulated annealing is definitely stochastic hill climbing. Well, I suppose that if all of the known options are "rarely the best thing to do", that would explain why optimisation is so hard ;P

@PeterTaylor yes.. :) I think my argument would be that stochastic hill climbing in its simplest form is easier to implement and understand than GA which introduces spurious vocab and hope that some magic might happen

basically it is stochastic hill climbing + hocus pocus

in fact there is more to the Metropolis Hastings algorithm and MCMC

which are much more interesting

One @ at a time is enough.

argh.. I did that again .. sorry

I blame the muscle memory in my fingers :)

I've skimmed the Wikipedia pages for Metropolis Hastings and MCMC and I don't see the connection.

They both talk about sampling from a probability distribution. Does that implicitly lead to an optimisation technique?

(Other than the "pick a random value and see whether it's better than the best found so far", which seems pretty sub-optimal)

yes.. if you define the distribution to make high scoring instances have higher probability

then you get an optimisation algorithm

you can define the probability of an instance any way you like.. it's just a function from instances to probabilities

Ok. So it's another form of stochastic hill-climbing?

@PeterTaylor well sort of. It has the very strong provable property that at convergence you sample from exactly the probability distribution you have specified

of course convergence may be a long way off :)

so it is related to stochastic hill climbing but not really the same thing

And if there's a large plateau of near-maximum values, it's not going to do very well.

right.. so the whole magic is in the moves you specify. Which is of course like saying, you will have to solve the problem :)

which is always going to be the case

The description in terms of random walks with rejection criteria sounds very hill-climbing-esque.

it is certainly related... as I say the main difference is the provable outcome

in terms of the distribution you end up sampling from

the general rule of thumb is to have local moves which are common and global moves which are rare

look at the opening sentence of cs.princeton.edu/courses/archive/spr06/cos598C/papers/…

I'm impressed: although that citation is wrong, Google Scholar put the correct one before the cited one in its results for my search. computer.org/csdl/mags/cs/2000/01/c1022.pdf

interesting

in other news.. It seems I have found some suffix array construction code that takes < 30 seconds for my kmer input

@Geobits I would have wanted the random tournament results by now, but my computer auto-hibernated after only a few rounds.

I've timing it to be about 15 seconds per round.