--- Log opened Tue Aug 15 00:00:39 2017 | ||
-!- olinguyen [81615ad9@gateway/web/freenode/ip.129.97.90.217] has quit [Quit: Page closed] | 00:34 | |
-!- Netsplit *.net <-> *.split quits: micmn | 01:54 | |
-!- Netsplit *.net <-> *.split quits: sonney2k, shogitter, zxtx, OXPHOS, ironstark, @ChanServ | 01:54 | |
-!- Netsplit over, joins: micmn | 01:57 | |
-!- Netsplit over, joins: OXPHOS, @ChanServ | 01:58 | |
-!- ironstark [sid221607@gateway/web/irccloud.com/session] has joined #shogun | 01:58 | |
-!- ironstark [sid221607@gateway/web/irccloud.com/session] has quit [Changing host] | 01:58 | |
-!- ironstark [sid221607@gateway/web/irccloud.com/x-aijyzoawwgwvgtlz] has joined #shogun | 01:58 | |
-!- Netsplit over, joins: shogitter, sonney2k, zxtx | 01:59 | |
-!- Netsplit *.net <-> *.split quits: OXPHOS, @ChanServ | 01:59 | |
-!- Netsplit *.net <-> *.split quits: micmn | 02:00 | |
-!- Netsplit over, joins: micmn | 02:02 | |
-!- Netsplit over, joins: OXPHOS, @ChanServ | 02:02 | |
@sukey | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3965 opened by Sahil333 | 07:27 |
---|---|---|
-!- geektoni [~geektoni@93-34-128-38.ip49.fastwebnet.it] has joined #shogun | 09:48 | |
@sukey | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3965 synchronized by Sahil333 | 10:12 |
@wiking | geektoni, ping? | 11:11 |
geektoni | wiking: pong | 11:11 |
@wiking | anything blocking you/ | 11:11 |
@wiking | ? | 11:11 |
@wiking | yesterday i've passed out | 11:11 |
@wiking | dunno if there's anything that i should help u with | 11:12 |
@wiking | or not | 11:12 |
geektoni | wiking: well, there is still the python interface thing | 11:12 |
@wiking | which? | 11:12 |
geektoni | so, yesterday I was able to make the get_observations() object iterable | 11:13 |
geektoni | I mean that this https://github.com/geektoni/shogun/blob/parameter_observer_cv_3/examples/undocumented/python/evaluation_cross_validation_mkl_weight_storage.py#L76 works now | 11:15 |
geektoni | wiking: however, another little problem arose here https://github.com/geektoni/shogun/blob/parameter_observer_cv_3/examples/undocumented/python/evaluation_cross_validation_mkl_weight_storage.py#L79 | 11:16 |
@wiking | aah you mean the casting :) | 11:18 |
geektoni | since I store the trained machines as CMachine*, I have to cast them to the proper type when I need to use them | 11:18 |
geektoni | yes yes | 11:18 |
@wiking | so afaik there's no obtain_from_generic :P | 11:18 |
@wiking | (see labels has that operator) | 11:18 |
geektoni | ok, so I'll need to add it | 11:22 |
@wiking | yep :D | 11:22 |
geektoni | wiking: is there a general rule where to place obtain_from_generic methods? because I see they're implemented here and there :P | 11:24 |
@wiking | no it's crazy :) | 11:24 |
@wiking | it was on demand basis | 11:25 |
@wiking | :D | 11:25 |
geektoni | I'm not surprised ;) | 11:28 |
@wiking | :D well enterprise design | 11:29 |
@wiking | :) | 11:29 |
-!- HeikoS [~heiko@host-92-0-169-11.as43234.net] has joined #shogun | 11:48 | |
-!- mode/#shogun [+o HeikoS] by ChanServ | 11:48 | |
@wiking | HeikoS, ping? | 12:08 |
@HeikoS | wiking: jo | 12:08 |
@wiking | so the only question now is here | 12:09 |
@wiking | *where | 12:09 |
@sukey | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3965 synchronized by Sahil333 | 12:11 |
@wiking | btw have u seen this ^ PR? | 12:12 |
@HeikoS | wiking: didnt see it yet | 12:16 |
@HeikoS | will check | 12:16 |
@HeikoS | wiking: yeah we can discuss this with the team | 12:16 |
@HeikoS | have a voting system or so | 12:16 |
@HeikoS | but I think two things make sense: zurich or hungatry | 12:16 |
@wiking | i've picked up the thread | 12:18 |
@wiking | on the mailinglist | 12:18 |
@sukey | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3965 | 12:33 |
@sukey | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3965 synchronized by Sahil333 | 13:02 |
-!- sheogorath27 [sheogorath@gateway/shell/firrre/x-rdgcmdvvjpibutry] has left #shogun [] | 13:27 | |
@sukey | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3960 synchronized by MikeLing | 14:11 |
-!- HeikoS [~heiko@host-92-0-169-11.as43234.net] has quit [Ping timeout: 240 seconds] | 14:31 | |
@sukey | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3953 synchronized by geektoni | 15:58 |
geektoni | wiking: ^ I've fixed the last things. It should be ready now (let's see what CIs say). | 16:06 |
@wiking | \o/ | 16:06 |
-!- olinguyen [81615ad9@gateway/web/freenode/ip.129.97.90.217] has joined #shogun | 16:30 | |
geektoni | wiking: is there any way to make the meta example generator produce c++ class instantiations without the "C" prefix? :P without touching the generator or the cpp.json file I mean :P | 16:47 |
geektoni | wiking: btw, https://github.com/shogun-toolbox/shogun/pull/3953 is all green | 16:49 |
-!- HeikoS [~heiko@eduroam-int-pat-8-18.ucl.ac.uk] has joined #shogun | 17:04 | |
-!- mode/#shogun [+o HeikoS] by ChanServ | 17:04 | |
olinguyen | HeikoS: Hey, wanna chat? | 17:05 |
-!- HeikoS [~heiko@eduroam-int-pat-8-18.ucl.ac.uk] has quit [Ping timeout: 248 seconds] | 17:08 | |
-!- OXPHOS_ [92bd305b@gateway/web/freenode/ip.146.189.48.91] has joined #shogun | 17:15 | |
-!- OXPHOS_ [92bd305b@gateway/web/freenode/ip.146.189.48.91] has quit [Client Quit] | 17:16 | |
-!- HeikoS [~heiko@untrust-out.swc.ucl.ac.uk] has joined #shogun | 17:23 | |
-!- mode/#shogun [+o HeikoS] by ChanServ | 17:23 | |
olinguyen | HeikoS: hey | 17:25 |
@HeikoS | olinguyen: hi! | 17:25 |
olinguyen | do you have some time to chat a little? | 17:25 |
@HeikoS | chat is OK | 17:26 |
olinguyen | cool | 17:26 |
@HeikoS | I cannot do a call, as I have some stuff to do, will have to do that tomorrow | 17:26 |
olinguyen | that's fine | 17:26 |
olinguyen | So about SQL vs Pandas: i found that it would end up with more computation with pandas | 17:27 |
olinguyen | cause for a single patient and single variable (heart rate) it amounts to ~1,600,000 data points | 17:27 |
olinguyen | sorry | 17:27 |
olinguyen | all patients* | 17:27 |
olinguyen | but for only heart rate | 17:28 |
olinguyen | expanding that to other vital signs, i'd have to do a separate query each time | 17:28 |
olinguyen | and then resample and combine through pandas | 17:28 |
@HeikoS | olinguyen: I see | 17:29 |
@HeikoS | so with SQL you never even touch the data you dont need | 17:29 |
olinguyen | exactly, and i can aggregate all vital signs within that hour timestamp | 17:29 |
@HeikoS | that is fine then | 17:29 |
@HeikoS | make sure to do some (reproducible) testing that pandas and SQL re-sampling gives you the same result | 17:29 |
@HeikoS | (on small time windows of course) | 17:29 |
olinguyen | yea, i briefly double checked. it looks consistent | 17:30 |
olinguyen | i think you had other questions. do you want to leave that for the call tomorrow or so? | 17:31 |
@HeikoS | olinguyen: yeah | 17:32 |
@HeikoS | olinguyen: so were the thing sin the plots for one patient or for all? | 17:33 |
olinguyen | a single patient | 17:33 |
@HeikoS | I guess for all of them | 17:33 |
@HeikoS | ah | 17:33 |
@HeikoS | ok and the evaluation as well only for one? | 17:33 |
olinguyen | evaluation is for all patients | 17:33 |
@HeikoS | I see | 17:33 |
@HeikoS | ok thats good then | 17:33 |
olinguyen | yea | 17:33 |
@HeikoS | you should try to experiment with different target labels | 17:34 |
olinguyen | i split the train/test to 50/50 | 17:34 |
@HeikoS | "dies within the next week" "dies within next month" | 17:34 |
@HeikoS | note that those might include each other | 17:34 |
@HeikoS | if you use "within" | 17:34 |
olinguyen | right | 17:34 |
@HeikoS | but "within" is better than "at" | 17:34 |
@HeikoS | 50/50 of all available data? | 17:34 |
olinguyen | correct | 17:34 |
@HeikoS | ok | 17:34 |
@HeikoS | and whats the class ratio for the labels you used? | 17:34 |
olinguyen | 14/1 | 17:35 |
olinguyen | i didn't do any undersamping (left it as is) | 17:35 |
olinguyen | i'll see how changing the ratio affects the results | 17:35 |
@HeikoS | in theory, you should only have a single "positive" point per patient at the moment no? | 17:36 |
@HeikoS | what exactly was your label? | 17:36 |
olinguyen | it depends where in time we're looking at | 17:36 |
olinguyen | starting at 00:00, if the patient dies at 23:00, and we look at the sample at say 8:00, then the label is alive in 6 hours | 17:37 |
olinguyen | changing to "dies in 1 day" would be different | 17:37 |
olinguyen | but it seems that the classifier score is pretty similar for both outcomes (dies in 6 hours or dies in 1 day) | 17:38 |
@HeikoS | olinguyen: what does your "in" mean? | 17:44 |
@HeikoS | within? | 17:44 |
@HeikoS | or at? | 17:44 |
olinguyen | every row, or sample has a timestamp. Looking whether the patient is alive/dead in 6 hours means i'm looking at 6 hours ahead of time, from the time of that timestamp | 17:46 |
@HeikoS | olinguyen: ok, so that is "within" | 17:46 |
@HeikoS | since a patient never is alive after once being dead | 17:47 |
olinguyen | right | 17:47 |
@HeikoS | so you get multiple positives for a single death event | 17:47 |
olinguyen | that's correct | 17:47 |
@HeikoS | that is good | 17:47 |
@HeikoS | makes the problem slightly easier | 17:47 |
@HeikoS | can you try multiple time horizonts? | 17:47 |
@HeikoS | like a day, a week, a month | 17:47 |
@HeikoS | a year? | 17:47 |
@HeikoS | and the same for the features | 17:48 |
olinguyen | yea, but i'd have to base it on a few days of vital signs | 17:48 |
olinguyen | because the majority of patients stay in the ICU for < 5 days | 17:48 |
@HeikoS | computing averaged for 6h, 1d, 1w, 1y | 17:48 |
@HeikoS | ah I see | 17:48 |
@HeikoS | olinguyen: ok let's talk tomorrow | 17:48 |
olinguyen | sure! | 17:48 |
@HeikoS | i have to finish some stuff here | 17:49 |
olinguyen | np | 17:49 |
OXPHOS | Just came back from seattle :) | 17:53 |
OXPHOS | There's a cppcon by the end of Sept there | 17:53 |
OXPHOS | thinking about going | 17:53 |
-!- mikeling [uid89706@gateway/web/irccloud.com/x-dxrcskczrhsxspus] has joined #shogun | 18:24 | |
mikeling | HeikoS: Hi, does gpl module is necessary to build every time? Could we avoid to build it by run 'make'? | 18:25 |
@HeikoS | mikeling: you can turn int off | 18:25 |
@HeikoS | OXPHOS: hello! | 18:25 |
mikeling | oh! how | 18:26 |
@HeikoS | LICENSE_GPL_SHOGUN=OFF | 18:26 |
@HeikoS | cmake option | 18:26 |
mikeling | ok thank you@ | 18:26 |
mikeling | ! | 18:26 |
@HeikoS | let me know if it worked | 18:26 |
mikeling | it looks like I should turn off the SVMLight also? | 18:31 |
mikeling | it says Can only use SVMLight when GPL codes are included | 18:31 |
@wiking | HeikoS, i just remember that i wanted to ask... what happened with src/shogun/statistical_testing/internals/FeaturesUtil.cpp ? | 18:35 |
@wiking | it is dropped for good? | 18:35 |
@HeikoS | mikeling: yes SVMLight has GPL dependencies | 18:42 |
@HeikoS | wiking: I don't know | 18:42 |
@wiking | HeikoS, just because it's not in gpl | 18:43 |
@wiking | but used to be in shogun | 18:43 |
@wiking | hence i'm just wondering what happened | 18:43 |
mikeling | HeikoS: yes, I just turn it off and everything works :) | 18:43 |
@HeikoS | oh | 18:43 |
@HeikoS | mmh | 18:43 |
@HeikoS | I don tknow | 18:43 |
@HeikoS | dont remember changing things | 18:44 |
@HeikoS | but maybe rahul did? | 18:44 |
-!- geektoni [~geektoni@93-34-128-38.ip49.fastwebnet.it] has quit [Remote host closed the connection] | 19:20 | |
@sukey | [https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3960 synchronized by MikeLing | 19:31 |
-!- HeikoS [~heiko@untrust-out.swc.ucl.ac.uk] has quit [Ping timeout: 255 seconds] | 19:35 | |
-!- Netsplit *.net <-> *.split quits: mikeling | 21:05 | |
-!- mikeling [uid89706@gateway/web/irccloud.com/x-rnpaizsgjirnevii] has joined #shogun | 21:06 | |
--- Log closed Wed Aug 16 00:00:41 2017 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!