The Nineteenth Byte

Sparr

3:12 AM

@NathanMerrill CodeBots2Java needs to handle division and modulus by zero in some way

I'm just returning zero, for now

Sparr

3:31 AM

All( BotAt(D) , None( BotAt(Add(D,1)) , BotAt(Add(D,2)) , BotAt(Add(D,3)) ) )

that should trigger if there's a bot in front of me and no bots on other sides of me, right?

@NathanMerrill this doesn't seem to work ^^

Sparr

3:59 AM

@NathanMerrill also, I think the search for valid conditions after Start doesn't loop back to the beginning

Nathan Merrill

4:24 AM

@Sparr I don't know of a better way to handle divide by 0. I'll check tomorrow as to why that isn't working and I'm fairly positive that looping back around works, but I'll double check tomorrow

Sparr

returning zero seems the easiest course

making the whole line a noop is also an option

Nathan Merrill

oooh...I think I know why the condition doesn't trigger

hmm...maybe I don't

Yes, the search for valid conditions does loop back to the beginning

I'll test/figure out the other one tomorrow

Sparr

Something is wrong with bots that don't Start at the beginning, I think

simply adding a "Flag" line at the top of my bot ruins its performance

and it has no absolute line numbers

Sparr

4:52 AM

also, I think this is top of the leaderboard next time you run it: codegolf.stackexchange.com/questions/37308/…

user2179021

6:15 AM

hi @MartinBüttner

Peter Taylor

7:18 AM

@MartinBüttner Not sure. I think that if someone wants to do that as a challenge they should first write a reference implementation to get a feel for how trivial it is.

Martin Büttner

10:24 AM

@PeterTaylor I might try a naive implementation in Julia an see if it's possible to find any that way at all

Martin Büttner

11:37 AM

@user2179021 I'm trying to work on a proper submission for your challenge. But I seem to be misunderstanding something. Say, for every point in Y, I try to rotate the first point in X onto that point. Shouldn't there be one rotation where the second point in X falls exactly (to machine precision) on another point in Y?

user2179021

1:12 PM

@MartinBüttner hi

Martin Büttner

@PeterTaylor okay, it's trivial to find them for 16-bit floats (they start around magnitude 40-50). and not too hard for 32-bit floats either (the first pair is (i,j) = (2897,2896); (x,y) = (3220,2532)). however, this challenge has the same problem as the euler brick one: every integer multiple of a solution is also a solution. so I need a way to encourage searching for more than just multiples of a few hardcoded ones

@user2179021 hi

user2179021

@MartinBüttner let me check something about the data creation. I am slightly worried I didn't update the code in the question

@MartinBüttner in relation to your question

Martin Büttner

@PeterTaylor there was a bug in my code. float16 collisions start at (i,j) = (24,22); (x,y) = (31,10) and float32 at (i,j) = (1483,1471); (x,y) = (1675,1248)

user2179021

@MartinBüttner ok back.. are you not finding the second point which matches exactly?

Martin Büttner

@user2179021 no, they are usually by about 10^-2 or 10^-3 off, and if I take that match I still get an overall distance about as bad as your generic optimiser

there might be a bug in my code of course

but I thought I'd check with you that this should actually be the case in the first place

user2179021

1:29 PM

@MartinBüttner It is possible I mucked something up. Would you mind if I just recreated the data?

Martin Büttner

@user2179021 no, go ahead. I'll give it another go then.

user2179021

ok.. let me do that

ah.. found the problem

@MartinBüttner The data creation code was missing "+np.random.random()"

it should all make sense now

sorry about that

so in answer to your question.. no :)

@MartinBüttner let me know if it makes sense now please

Martin Büttner

ah I see.

that does make sense

otherwise the challenge would probably have been quite trivial

user2179021

or at least easier :)

also.. what a "python-bashing" tag :)

Martin Büttner

does random.random() return in [0,1)?

user2179021

1:36 PM

what about a...

@MartinBüttner "Return random floats in the half-open interval [0.0, 1.0)"

so it doesn't include 1.0

yes :)

Martin Büttner

so "yes" :D

user2179021

what is your approach going to be? I am intrigued

Martin Büttner

not sure I'll still be bothered to work it out with that change :D ... but I think I should be able to figure out a handful of candidate rotations, about which I can minimise.

user2179021

sounds cool (not that part about potentially giving up !)

it occurred to me that there is probably an exactly O(n^2) solutuion

an exact O(n^2) solution I mean

which is fairly useless for my largest test of course

I hate to think this problem has defeated the Internet :)

Martin Büttner

well how exact can it be? in the end you still need to figure out the best exact rotation which is going to be different from your rotation due to the random offsets

Nathan Merrill

1:50 PM

@Sparr I believe that the conditions are working properly, but that the highlighting I give them at the beginning is not

I'm not sure about that though

Sparr

Next time I'm working on bots I'll make some test cases

I ended up just replacing my All(x,None(x,x,x)) with All(x,Not(x),Not(x),Not(x)) and it worked fine

Nathan Merrill

well, its interesting because I had my BotAt condition return False every time

and that line you gave me to test always showed blue on the first turn

but Insidious does fantastic

188.1

Sparr

yeah :)

I'm sorta sad at how stagnant the field is after a few hundred turns in CB2

New Sandboxed Posts

0

A: Sandbox for Proposed Challenges

Still Trying To Think of a Good Name Prior to the discovery of fingerprints and DNA testing, the British police used an anthropometric system to identify repeat offenders. It involved a certain parts of the human body. It was believed that the size of these body parts did not change after adulth...

Sparr

most bots stop moving

it's a lot easier to get frozen in CB2

Nathan Merrill

1:53 PM

for sure

also, I made a pretty major fix with the latest release. only a few of the O-arguments were working. I just pushed the fix

Martin Büttner

@PeterTaylor Okay, finally found some pairs for 64-bit floats. I estimate that they all appear beyond magnitudes of 10^7. so by starting at 50M, I found (e.g.) (i,j) = (50000000,45000018); (x,y) = (50682794,44229583)

Martin Büttner

2:42 PM

@COTO you might be interested in this ^

(this is about false positives on a circle lattice)

COTO

3:17 PM

@MartinBüttner: Indeed. Confirmed that the former vector has magnitude `0.6726813227673505132725932e8` and the latter has magnitude `0.6726813227673505876019945e8`, differing in the 17th digit (would be the 16th in a normalized float).

I'm not a great student of algebraic geometry, but perhaps there's an efficient way to determine all integer solutions to the Diophantine equation `x^2+y^2-w^2-z^2=n`, in which case you could simply step `n` over a reasonable range of values to determine different potential sets of false positives.

@Sparr: I've been thinking of different ways to put together a bot for CB2, but I'm having a hard time with the freezing issue too. There are multiple ways to freeze now, and no efficient ways to block instructions.

user2179021

@MartinBüttner I meant exact to within machine precision

hi @COTO

has anyone asked question that requires limited memory? I would like to specify that the space usage has to be "constant" which makes sense from a theory point of view but I am not sure how to specify that for code

any idea?

COTO

Hi, Lembik.

user2179021

@COTO Do you have any interest in my question by any chance? codegolf.stackexchange.com/questions/37269/…

it is feeling slightly lonely and I love it so much :)

COTO

3:34 PM

You state that the measure of distance between the points isn't symmetric, but how so?

Given any point in X, it will have a closest partner Y, with a corresponding distance D.

If we use Y as a reference, its closest partner will be X, also with difference D.

The various distances may be summed up in different ways, but they should still sum to the same total.

user2179021

@COTO oh I can explain that :)

let X have [1] and Y have [2,3]

the distance from X to Y is 1

from Y to X it is 3

1+2

COTO

Oh, you're saying that the number of points might not be equal.

No, actually you're right. I just worked up a counterexample.

OK, false alarm. My apologies.

user2179021

no problem :)

I hope this has inspired you to post an answer :)

COTO

I'll submit something this afternoon.

user2179021

the number of points may well not be equal. In the data I give they are not

cool!

if you look at the data creation code you will see timesY has 10 times as many points as timesX

user2179021

4:02 PM

meta.codegolf.stackexchange.com/a/2145/9206 please comment

New Sandboxed Posts

0

A: Sandbox for Proposed Challenges

Produce a factorization picture The task is simple. Your code should take in one positive integer n less than 1000 and output a picture in any standard graphical format (jpeg, png etc) of the factorization of n. What is this picture? It should look exactly like the corresponding picture from htt...

Nathan Merrill

6:12 PM

...who controls the twitter for code golf?

CodeGolf SE

@StackCodeGolf, The Stack Exchange network

A Q&A site for programming puzzle enthusiasts and code golfers

2k tweets, 178 followers, following 0 users

aka...who wrote the program to do it?

Vi.

8:22 PM

Should there be a Capture The Flag challenge on PPCG?

0

A: Sandbox for Proposed Challenges

Capture the flag king-of-the-hill Basic rules There is a rectangular map of cells, each bot has current cell. Bots compete 1 vs 1 at a time, let's call them red and blue. There are N instances of red bot and N instances of the blue bot on the map. Additionally, there is unmovable red base and ...

Martin Büttner

8:44 PM

I haven't read your spec yet, but I think @Geobits was working on one, too

New Sandboxed Posts

9:31 PM

0

A: Sandbox for Proposed Challenges

False Positives on an Integer Lattice fastest-code arithmetic Background When working on a grid of integer coordinates, you sometimes want to know whether two vectors (with integer components) have the same magnitude. Of course, the Euclidian norm of a vector is given by √(Δx² + Δy²) So a n...

Transcript for