2011-09-20T00:16:43 *** dvladim has quit IRC (Quit: Konversation terminated!) 2011-09-20T00:17:22 *** antimatroid has joined #aichallenge 2011-09-20T00:37:39 *** onensora has quit IRC (Ping timeout: 252 seconds) 2011-09-20T00:40:08 *** Cyndre has joined #aichallenge 2011-09-20T01:15:25 *** chris___0076 is now known as chris_0076 2011-09-20T01:17:29 *** bhasker has joined #aichallenge 2011-09-20T01:41:42 *** bhasker has quit IRC (Ping timeout: 252 seconds) 2011-09-20T01:59:37 *** Palmik has joined #aichallenge 2011-09-20T02:20:10 *** amstan has quit IRC (Ping timeout: 260 seconds) 2011-09-20T03:20:42 *** Kingpin13 has joined #aichallenge 2011-09-20T03:21:42 *** aerique has joined #aichallenge 2011-09-20T03:26:08 *** olexs has joined #aichallenge 2011-09-20T03:26:23 *** ltriant has quit IRC (Quit: Computer has gone to sleep) 2011-09-20T04:05:06 *** Palmik has quit IRC (Ping timeout: 258 seconds) 2011-09-20T04:13:41 *** Palmik has joined #aichallenge 2011-09-20T04:30:35 *** Palmik has quit IRC (Ping timeout: 276 seconds) 2011-09-20T04:42:55 *** eashoka has joined #aichallenge 2011-09-20T05:06:42 *** kaemo has quit IRC (Quit: *puff*) 2011-09-20T05:10:19 *** Palmik has joined #aichallenge 2011-09-20T05:11:17 *** ibdknox has quit IRC (Remote host closed the connection) 2011-09-20T05:37:58 *** mcstar has joined #aichallenge 2011-09-20T05:52:52 *** sigh has joined #aichallenge 2011-09-20T06:28:53 janzert : what if we set the starting skill of a new version (of a bot) to what it is already instead of forcing it to come up from 0. It is far more likely that the bot skill is around what it was rather than be near 0. This way, we wouldn't need ghost ranking as well. Impact on other bots also would be greatly reduced (other bots loosing on skill level by playing against a bot which is considerably lower ranked than it's actu 2011-09-20T06:34:00 janzert : There may be a very good reason why it is done in the way it's done at the moment. So, thought of running it against you b4 posting a possible annoyingly late feature request :-) anybody else who know why a new version of a bot has to come from ground up instead of starting from where it was ? 2011-09-20T06:47:13 *** olexs1 has joined #aichallenge 2011-09-20T06:48:48 *** olexs has quit IRC (Ping timeout: 260 seconds) 2011-09-20T07:02:09 i suppose if a bot did get worse, than it was b4 than maybe it would reach its lower ranking in a greater time than if it was started from the bottom 2011-09-20T07:02:45 i stress suppose, i dont know 2011-09-20T07:03:27 although, there are 2 parameters in computing trueskill, maybe it would be enough to only reset sigma, and still get a decent behavior 2011-09-20T07:07:18 *** Palmik has quit IRC (Remote host closed the connection) 2011-09-20T07:07:22 mcstar: that a good idea... 2011-09-20T07:07:41 i don't think any system works properly for resubmissions 2011-09-20T07:07:41 *** berak has joined #aichallenge 2011-09-20T07:07:47 but that may work the best 2011-09-20T07:09:24 *** jbroman has quit IRC (Quit: Scotty, beam me up!) 2011-09-20T07:11:24 *** pgpaskar has quit IRC (Ping timeout: 258 seconds) 2011-09-20T07:11:24 *** Zepp has quit IRC (Ping timeout: 258 seconds) 2011-09-20T07:11:24 *** jbroman has joined #aichallenge 2011-09-20T07:11:53 *** Zepp has joined #aichallenge 2011-09-20T07:16:51 *** pgpaskar_ has joined #aichallenge 2011-09-20T07:21:40 deltor : is it too late to suggest improvements to the scoring system ? 2011-09-20T07:21:54 *** jbroman has quit IRC (Quit: Scotty, beam me up!) 2011-09-20T07:22:24 *** jbroman has joined #aichallenge 2011-09-20T07:23:39 eashoka: ask mcleo when he is around 2011-09-20T07:23:42 *** Zepp has quit IRC (Ping timeout: 252 seconds) 2011-09-20T07:23:48 deltor : I feel something like this may work better than the current : ((Ants_killed - Ants_Lost) * weight) + Remaining_Ants_ at_the_End_of_the_game 2011-09-20T07:23:54 and its good to talk to jan.zert too 2011-09-20T07:23:58 seems he isnt around 2011-09-20T07:24:05 mcstar : ah ok, sure. tx 2011-09-20T07:25:58 *** delt0r_ has quit IRC (Ping timeout: 258 seconds) 2011-09-20T07:34:36 ive found an error in sage 2011-09-20T07:36:24 *** jbroman has quit IRC (Quit: Scotty, beam me up!) 2011-09-20T07:37:24 *** jbroman has joined #aichallenge 2011-09-20T07:39:01 *** delt0r_ has joined #aichallenge 2011-09-20T07:41:05 *** Zepp has joined #aichallenge 2011-09-20T07:49:36 *** onensora has joined #aichallenge 2011-09-20T08:01:30 *** eashoka has quit IRC (Quit: Page closed) 2011-09-20T08:02:00 delt0r: are you familiar with fft? 2011-09-20T08:02:12 (dumb question, sure you are :)) 2011-09-20T08:10:47 yes 2011-09-20T08:10:52 fft ftw 2011-09-20T08:11:00 oh yea fftw 2011-09-20T08:11:01 fftw ftw 2011-09-20T08:11:08 lol 2011-09-20T08:11:12 so 2011-09-20T08:11:40 ... 2011-09-20T08:11:48 im wondering if you consider fft-ing a signal and extracting its 0-component vs. integrating with a more sophisticated integrator 2011-09-20T08:12:07 considering if its good enough* 2011-09-20T08:12:21 sry, seems my english is broken today .) 2011-09-20T08:12:33 the zero component is just the DC level 2011-09-20T08:12:40 sure 2011-09-20T08:12:41 or the avarege 2011-09-20T08:12:46 avarage 2011-09-20T08:12:48 the integral of the series 2011-09-20T08:13:00 yea--up to a factor 2011-09-20T08:13:02 but im wondering about accuracy here 2011-09-20T08:13:10 it is identical 2011-09-20T08:13:20 for the zero component 2011-09-20T08:13:36 unless you go all complex numbers... 2011-09-20T08:13:38 why? 2011-09-20T08:13:46 in which case it still is 2011-09-20T08:14:09 if the signal is almost harmonic than the error is low i imagine 2011-09-20T08:14:11 because cos(x*0) is always 1 2011-09-20T08:14:36 but i also think that a trpezoid rule, or simpson's or romberg's integrator gives better results 2011-09-20T08:14:54 hence the zero component is literally just the integral of the signal 2011-09-20T08:15:18 delt0r: that i know, i know the "interpretation" 2011-09-20T08:15:27 well with fft there is the "window" function 2011-09-20T08:15:31 im wondering if fft is accurate enough 2011-09-20T08:15:43 at the dc level this is == to the "interpolation" method 2011-09-20T08:16:14 basically its a linear interp. 2011-09-20T08:16:16 ok 2011-09-20T08:16:20 fft is perfectly reversible with the proper window function and bandwidth 2011-09-20T08:16:41 i know that, again thats not my question 2011-09-20T08:16:53 mcstar: not quite, the default "square" window is a sinc interpolator 2011-09-20T08:17:19 yeah, square goes to sinc in freq. doamin 2011-09-20T08:17:23 yep 2011-09-20T08:18:22 honestly it is not often a plain Euler integrator wouldn't be good enough for the DC 2011-09-20T08:18:23 so i guess its notbad, since it already involves minimal number of arithmetic(fft) 2011-09-20T08:18:43 well not if you only want one component of the result 2011-09-20T08:18:56 i already have the fft of the signal 2011-09-20T08:18:59 since fft is n log n 2011-09-20T08:19:11 and a single component is just n 2011-09-20T08:19:16 oh ic 2011-09-20T08:19:17 and for 1 component its not more than an integrator 2011-09-20T08:19:24 yes 2011-09-20T08:19:30 the number of floating point operations that is 2011-09-20T08:19:40 why are you asking... you seem to know as much as i do? 2011-09-20T08:19:54 ? 2011-09-20T08:19:56 *** cichociemno has joined #aichallenge 2011-09-20T08:20:00 why not ask? 2011-09-20T08:20:06 one can always learn 2011-09-20T08:20:13 well for non dc components there is a extra * 2011-09-20T08:20:17 sure 2011-09-20T08:20:28 but its nothing really 2011-09-20T08:20:37 for example i misread numpy.fft manual 2011-09-20T08:21:01 this is how it maps k: [0 1 2 3 -3 -2 -1] 2011-09-20T08:21:11 but i interpreted it as [0 1 2 3 -1 -2 -3] 2011-09-20T08:21:11 odd 2011-09-20T08:21:44 i wrote a fft for my school project once 2011-09-20T08:21:46 its just a toy project, the real one uses cufft 2011-09-20T08:23:01 fft or dft? 2011-09-20T08:23:42 * mcstar is wonderign wheather it was cooley-tuckey or just summing exponentials 2011-09-20T08:23:57 my one? or what should you use? 2011-09-20T08:24:01 it was fft 2011-09-20T08:24:15 any power of 2 2011-09-20T08:24:27 full complex fft 2011-09-20T08:24:51 in fact what is dft? 2011-09-20T08:25:03 yours yes 2011-09-20T08:25:12 discrete fourier transform 2011-09-20T08:25:25 all fft must be discrete 2011-09-20T08:25:31 lol 2011-09-20T08:25:33 of course 2011-09-20T08:25:43 but fft is a dft 2011-09-20T08:25:52 and it has the nlogn characterstic 2011-09-20T08:25:58 the dft has n2 is think 2011-09-20T08:26:03 well i would use the d for things like dct 2011-09-20T08:26:07 or dst 2011-09-20T08:26:30 delt0r: dft is a common acronym 2011-09-20T08:26:34 well the fft only works for 2^n samples for any n 2011-09-20T08:26:44 never came across it sorry 2011-09-20T08:26:55 did real signal processing too 2011-09-20T08:27:15 and physics and 2 years EE 2011-09-20T08:27:21 cufft has an fft impl. that works for 2^n*3^m*5^k*7^p 2011-09-20T08:27:56 i would pretty much only use fftw now, it optimizes for whatever your n is 2011-09-20T08:28:08 but casting the problem in powers of 2 is not hard either 2011-09-20T08:28:29 in fact nowdays its pretty much R or matlab 2011-09-20T08:28:38 R uses fftw iirc 2011-09-20T08:29:00 i hate matlab, i rather use python, even if im not satisfied 2011-09-20T08:29:14 these days its 99% R 2011-09-20T08:29:26 but for 512^3 domains you cant really use any of it 2011-09-20T08:29:35 .8% mathmatica and .2% matlab --its what ppl send me 2011-09-20T08:29:55 i liked mathematica when i was young 2011-09-20T08:30:06 i even corrected a prof. that came to lecture about it 2011-09-20T08:30:18 well it is free for me... site license 2011-09-20T08:30:25 so is matlab. 2011-09-20T08:30:35 noob, didnt know he had to use ()'s for a plainfunction to use it correctly 2011-09-20T08:30:37 and MS word for that matter... but i don't touch that 2011-09-20T08:31:23 well i get that all the time... you are told 1 day before the class, that you need to give a lecture on XXX 2011-09-20T08:31:32 only you have never done/used XXX 2011-09-20T08:32:01 So i ask the students to correct me :) 2011-09-20T08:32:02 he was giving these 1 week lecture course regularly at the time, and was quite confident about it 2011-09-20T08:32:16 but it doesnt matter 2011-09-20T08:32:26 basically what he did, was some visualizations 2011-09-20T08:32:40 that was the most interesting part, but it gave my some credits so.. 2011-09-20T08:32:56 don't like mathmatica for that. sure its nice, but getting them out of mathmatica is a PITA 2011-09-20T08:33:35 really? 2011-09-20T08:33:46 i even used it to convert a bunch of images to pdf at one time :) 2011-09-20T08:33:47 yea--at least for publication quality 2011-09-20T08:33:53 perhaps the have fixed it now 2011-09-20T08:34:00 but it really was a pain 2011-09-20T08:34:08 well, those graphs look real good 2011-09-20T08:34:20 but it was hard to annotate in the past 2011-09-20T08:34:31 R works for me. and easier to import etc 2011-09-20T08:34:33 but nowadays, its can be edited in the gui 2011-09-20T08:34:34 and free 2011-09-20T08:35:09 but i didnt use mma for that, i liked the very fast prototyping in it 2011-09-20T08:35:28 I use to check my math 2011-09-20T08:35:34 I use it 2011-09-20T08:35:36 that too 2011-09-20T08:35:55 it gives the best results, no arguing about that 2011-09-20T08:36:04 esp the tedious stuff... 2011-09-20T08:36:20 i computed several integrals with it for which maxima or maple didnt give as nice results, if gave at all 2011-09-20T08:36:48 *** mceier has joined #aichallenge 2011-09-20T08:36:54 haven't used maxima or maple since my grad days... maple is still going? 2011-09-20T08:36:54 but 2011-09-20T08:37:02 ive got to do my stuff 2011-09-20T08:37:09 sure 2011-09-20T08:37:10 yes maple is in business 2011-09-20T08:37:21 ~15th version? i dont know 2011-09-20T08:37:24 never liked it 2011-09-20T08:39:51 delt0r: btw there are some practical differences between the discussed two methods it seems, but its too much to discuss it over irc i guess 2011-09-20T08:40:00 end-point related 2011-09-20T08:41:12 you mean with fft vers integration for one component 2011-09-20T08:41:48 yes 2011-09-20T08:41:58 ok i tell it 2011-09-20T08:42:14 lets say i have a sine wave, discretized over 2pi 2011-09-20T08:42:35 i would not think this would affect the dc component... ie the 2 different ways of doing boundary conditions in ffts 2011-09-20T08:42:42 and its important for the fft, for the signal to be truly periodic 2011-09-20T08:42:50 but the rest, yes i think it would 2011-09-20T08:42:54 meaning ar[N]==ar[0] 2011-09-20T08:43:16 but if i use an integrater, that gives a very large error if dx is not very small 2011-09-20T08:43:35 yea--and there is the ar[-N]==ar[N] iirc... it has been a while since i cared about that level of detail 2011-09-20T08:43:42 because it would expect ar[0]==ar[N-1] 2011-09-20T08:44:20 ?? why would you expect that? 2011-09-20T08:44:28 why whould it 2011-09-20T08:44:33 damit 2011-09-20T08:44:43 :) 2011-09-20T08:45:55 well, in numpy it seems trapz and simps both work this way 2011-09-20T08:46:30 anyway, they give 2 magnitude worse results 2011-09-20T08:46:33 than the fft 2011-09-20T08:46:48 1e-14 vs. 1e-16 2011-09-20T08:47:06 oh crap... that is not 2 real orders of mag 2011-09-20T08:47:14 that is double level noise 2011-09-20T08:47:18 thats obviously because the signal is very good for the fft, since its a sine 2011-09-20T08:47:37 yea-- but in general i would not count on it 2011-09-20T08:47:50 what did you mean by that? 2011-09-20T08:48:08 i know some would claim doubles are 1e-16... but reality is that for many things 1e-10 is dreaming 2011-09-20T08:48:39 well, i guess i could shift the signal to be all positive at least 2011-09-20T08:49:15 however for the record Gaussian quadrature can get very highly accurate results --but 1e-16 is still dreaming for the general case IMO 2011-09-20T08:49:45 yes, i wont argue with that 2011-09-20T08:49:51 also consider SNR 2011-09-20T08:50:07 if you sine wave has amplitude 1 2011-09-20T08:50:26 that is mmm forgot my dB 2011-09-20T08:50:46 14 orders of magnatude is that like 70dB? 2011-09-20T08:50:49 or 140dB 2011-09-20T08:51:32 dunno 2011-09-20T08:51:39 huge either way 2011-09-20T08:51:57 1 db is a 21x factor? 2011-09-20T08:52:01 i remember something like this 2011-09-20T08:52:08 3dB is a factor of 2 2011-09-20T08:52:13 shat 2011-09-20T08:52:21 sry i didnt use db in many years 2011-09-20T08:52:27 only in electronics lab 2011-09-20T08:52:28 neither 2011-09-20T08:52:35 at that time i knew it 2011-09-20T08:52:37 * delt0r googles 2011-09-20T08:53:17 yea its 140 dB 2011-09-20T08:53:56 nope 2011-09-20T08:53:59 70 2011-09-20T08:54:05 oh who cares 2011-09-20T08:54:10 dB is lame anyway 2011-09-20T08:54:16 * delt0r does some work 2011-09-20T08:54:24 ah yes, -2db is for halfing the amplitude 2011-09-20T08:54:29 some like that i remember 2011-09-20T08:54:47 140 2011-09-20T08:54:55 hell -3db* 2011-09-20T09:00:07 well, log10(2)~3.3 2011-09-20T09:00:22 anyway its -140db 2011-09-20T09:01:15 ah yes, gain=20log(a1/a2) 2011-09-20T09:01:23 gain=20log10(a1/a2) 2011-09-20T09:09:01 *** olexs1 has quit IRC (Read error: Connection reset by peer) 2011-09-20T09:15:43 *** Eruonen has joined #aichallenge 2011-09-20T09:17:31 *** onensora has quit IRC (Ping timeout: 258 seconds) 2011-09-20T09:22:09 *** olexs has joined #aichallenge 2011-09-20T09:35:56 *** foRei has joined #aichallenge 2011-09-20T09:49:03 *** cichociemno has quit IRC (Quit: Page closed) 2011-09-20T09:49:17 *** cichociemno has joined #aichallenge 2011-09-20T10:12:27 *** FireFly has joined #aichallenge 2011-09-20T10:18:43 *** Palmik has joined #aichallenge 2011-09-20T10:27:04 *** eashoka has joined #aichallenge 2011-09-20T11:17:44 *** aerique has quit IRC (Quit: ...) 2011-09-20T11:18:35 *** sigh has quit IRC (Remote host closed the connection) 2011-09-20T11:54:18 *** Akranis has joined #aichallenge 2011-09-20T12:00:59 *** AI|loop has joined #aichallenge 2011-09-20T12:10:32 *** McLeopold has joined #aichallenge 2011-09-20T13:01:39 *** AI|loop has left #aichallenge ("Leaving") 2011-09-20T13:08:24 *** cichociemno has quit IRC (Quit: Page closed) 2011-09-20T13:19:45 McLeopold: got another empty task on a worker 2011-09-20T13:19:45 *** UncleVasya has joined #aichallenge 2011-09-20T13:20:02 http://pastebin.com/Ggd0ZDEx 2011-09-20T13:25:04 janzert: I see a traceback for the utf8 issue 2011-09-20T13:25:28 great 2011-09-20T13:25:53 I think it may be a bad username? 2011-09-20T13:26:07 it's when the json encoding happens 2011-09-20T13:27:28 no, usernames are not known by the worker 2011-09-20T13:27:40 maybe something in stdout or stderr from a bot is writing unicode? 2011-09-20T13:27:46 but only sometimes? 2011-09-20T13:28:29 well the problem is that something is trying to interpret something as unicode when it isn't ;) 2011-09-20T13:28:31 so, info from the bots needs to go through sanitization before being sent back 2011-09-20T13:31:36 janzert: how in sync are the times on the workers and server? 2011-09-20T13:32:26 not sure, I would suspect that the global time is pretty close but the timezones are different 2011-09-20T13:32:51 can you tell me the matchup id the worker processed just before the crash? 2011-09-20T13:33:02 yeah looks like worker is UTC 2011-09-20T13:33:14 so add 4 hours to get beta server time 2011-09-20T13:33:46 305954 2011-09-20T13:34:19 oh, "trying to reset mysql..." 2011-09-20T13:34:32 sometimes the php mysqli fails 2011-09-20T13:35:12 did both workers go down? 2011-09-20T13:35:29 just one 2011-09-20T13:35:34 205 2011-09-20T13:36:00 okay, I'm not sure what to do about php, but the worker should just wait 20 seconds and try again 2011-09-20T13:38:03 hmm, quick read of worker.py:task() I would think that's what it would do already 2011-09-20T13:38:23 *** ibdknox has joined #aichallenge 2011-09-20T13:38:53 i.e. the else clause starting at line 549 should cover this I would have thought 2011-09-20T13:39:06 aichallenge: McLeopold epsilon * r94145ed / worker/worker.py : prevent worker from stopping - http://git.io/YH7Rww 2011-09-20T13:39:29 yep :) 2011-09-20T13:39:32 ahh 2011-09-20T13:39:36 :P 2011-09-20T13:39:55 okay, 2 more bugs on the worker, then we are really solid! 2011-09-20T13:42:06 touch wood 2011-09-20T13:42:19 knock on wood 2011-09-20T13:44:05 janzert: how about everything gathered from the bots is converted to utf8 right away, if there are errors, then the line is throw out? 2011-09-20T13:44:28 *** ibdknox has quit IRC (Remote host closed the connection) 2011-09-20T13:44:49 are some bots putting out illegal uft8 chars 2011-09-20T13:44:54 just one 2011-09-20T13:45:10 let me look at it some more 2011-09-20T13:45:27 yea--inputs should be sanitized 2011-09-20T13:45:37 * delt0r_ looks at my engine 2011-09-20T13:45:44 * delt0r_ looks at his engine 2011-09-20T13:46:07 janzert: line 140 in worker, but the replay object has the bad data 2011-09-20T13:46:18 random binary data from the bot should get a kick right? 2011-09-20T13:46:21 unfortunately the one traceback we have is when the worker tried to send a result that it had from a saved result 2011-09-20T13:46:26 not the initial try 2011-09-20T13:46:26 we could at least try to dump it to see what is happening 2011-09-20T13:46:35 sign of insanity when one refers to himself in the third place, /me doesnt help people stay sane 2011-09-20T13:46:48 lol 2011-09-20T13:47:05 * McLeopold types /me 2011-09-20T13:47:19 well,its funny, just dont get used to it 2011-09-20T13:47:55 Here was no activity for a long time earlier but now it's "wow!". Almost all issues are closed now, Thanks organizers/contributors. 2011-09-20T13:48:59 UncleVasya: if you are bored, just read the logs, they date back to 2010.09 2011-09-20T13:50:18 *** dvladim has joined #aichallenge 2011-09-20T13:53:37 janzert: how do I grep for unicode? 2011-09-20T13:55:01 pcregrep? 2011-09-20T14:01:25 ? 2011-09-20T14:01:50 I was looking at the submission that causes the issue to see if there was anything obvious 2011-09-20T14:02:38 the problem we're seeing is that the json encoder is running into something that isn't unicode, probably binary data 2011-09-20T14:03:02 McLeopold: pcregrep has the -u option to run in unicode mode 2011-09-20T14:03:05 ok 2011-09-20T14:03:18 was really looking for latin-1 or something 2011-09-20T14:03:40 it has to be something the bot is outputting, since it is only the one bot some of the time 2011-09-20T14:03:56 what language is it written in? 2011-09-20T14:04:10 cpp 2011-09-20T14:04:24 McLeopold: you can't really 2011-09-20T14:04:41 there are a few illegal byte sequences... but that it 2011-09-20T14:04:46 http://aichallengebeta.hypertriangle.com/profile.php?user=22 2011-09-20T14:04:58 a short random set of bytes still has a good chance of passing 2011-09-20T14:05:07 ie being legal utf8 2011-09-20T14:05:10 I now know a way of preventing any losses from being recorded by the system. I can now cheat my way to a win! 2011-09-20T14:05:22 the result has an 0xff byte, that is probably coming from the bot but we don't really know that for sure 2011-09-20T14:05:50 yea--a 0xff byte without a 0x70 is illegal iirc 2011-09-20T14:06:20 0xffff is also illegal... but 0x70ff is fine IIRC 2011-09-20T14:06:45 can't you just catch the exception? 2011-09-20T14:06:51 that is what i am doing 2011-09-20T14:07:03 what we should do is encode the bot output to unicode using a replacement strategy so illegal bytes are replaced with the unicode replacement character 2011-09-20T14:07:07 any excpetion in io/parsing gets the bot kicked 2011-09-20T14:08:06 the best would be to restrict it to utf8 the first 128 codes 2011-09-20T14:08:09 0x70ffffff is nan isnt it? 2011-09-20T14:08:11 aka ascii 2011-09-20T14:08:14 i dealt with something similar 2011-09-20T14:08:22 its the easiest... 2011-09-20T14:08:42 mcstar: you may be right 2011-09-20T14:08:50 we allow real unicode most places already 2011-09-20T14:08:51 there are 3 byte codes in utf8 2011-09-20T14:09:04 right... 2011-09-20T14:10:05 I was just looking it up... other than catch the exception, you have to check the byte stream legal ness 2011-09-20T14:10:17 Loool, it's my friend. I can bring him here. 2011-09-20T14:10:39 Imean user=22 2011-09-20T14:10:44 ok 2011-09-20T14:10:45 which can include all 255 bytes... 2011-09-20T14:10:55 maybe he can explain why his bot would do that? 2011-09-20T14:12:06 He will come here in few minutes. 2011-09-20T14:13:06 janzert: we also have a sandbox error come up occasionally. It's at line 290 of engine.py, during the cleanup. I suspect we ought to just catch that error? 2011-09-20T14:13:22 no 2011-09-20T14:13:42 I saw that but haven't looked into it yet 2011-09-20T14:13:45 but it prevents the game from being posted, and that is after the game was played 2011-09-20T14:13:53 *** Dlayne has joined #aichallenge 2011-09-20T14:13:56 it means the sandbox is being used wrong somewhere 2011-09-20T14:14:11 it's a sanity check 2011-09-20T14:14:27 the code is trying to kill bots still alive just before that 2011-09-20T14:14:28 and better to have sandbox errors crash out 2011-09-20T14:14:51 somehow it's missing killing one or more 2011-09-20T14:16:26 hey guys 2011-09-20T14:16:34 janzert: maybe my headtail class is choping unicode in half? 2011-09-20T14:16:37 aichallenge: janzert epsilon * rcea40d1 / worker/worker.py : Before crashing try and save a result with bad data in it - http://git.io/fMFEIA 2011-09-20T14:17:18 I think it's most likely the bot is accidentally writing raw binary data to stdio 2011-09-20T14:17:39 stderr 2011-09-20T14:17:47 most likely 2011-09-20T14:18:40 I'm InsaneMalkavian. Did you discuss my bot? 2011-09-20T14:18:51 *** Dlayne is now known as InsaneMalkavian 2011-09-20T14:19:02 server sql/2_generate_matchup.sql has local changes, can I blow them away? 2011-09-20T14:19:50 looks like it is an added "delimiter ;" line at the end 2011-09-20T14:20:05 really? 2011-09-20T14:20:05 hold on 2011-09-20T14:20:06 or actually maybe just an added newline 2011-09-20T14:20:38 yes 2011-09-20T14:20:54 What's wrong with it? 2011-09-20T14:21:07 I renamed the procedure so I could test first. 2011-09-20T14:21:23 InsaneMalkavian: I looks like sometimes your bot outputs binary data on stderr 2011-09-20T14:22:10 java automagically replaces illegal sequences of bytes with a ? 2011-09-20T14:22:27 (i was wrong, nan is 0x7fff.....) 2011-09-20T14:22:29 was that yes I can blow away the changes? 2011-09-20T14:22:56 delt0r_: python can do it too if you tell it to ;) 2011-09-20T14:23:01 janzert: is self._signal("KILL") synchronous? 2011-09-20T14:23:09 I am sure it could 2011-09-20T14:23:12 janzert: yes 2011-09-20T14:23:20 I would expect it too in fact 2011-09-20T14:23:36 but at least my engine is "safe" ;) 2011-09-20T14:23:55 It's very unlikely. I didn't use it afaik 2011-09-20T14:27:38 *** cody__ has joined #aichallenge 2011-09-20T14:27:43 well my engine was safe *while* the default encoding is utf-8. It is not on mac... so that is fixed 2011-09-20T14:29:10 janzert: I'm going to set InsaneMalkavian as the seed player for every game, so we can test this 2011-09-20T14:29:11 oh, how are "finished" ants dealt with --ie all their ants are dead.. I keep giving them turn X;go and waiting for go 2011-09-20T14:29:41 finished bots i mean 2011-09-20T14:29:44 we give them a full state, then we kill the process 2011-09-20T14:29:57 I used cpp starter package and just slightly modified bot::makemovea 2011-09-20T14:31:29 ok, next time we get a utf traceback there should also be a corresponding pickle of the result object on the worker 2011-09-20T14:31:30 McLeopold: the full state at the time their last ant dies? 2011-09-20T14:32:04 *** cody__ has quit IRC (Ping timeout: 252 seconds) 2011-09-20T14:33:49