IRC logs of #shogun for Monday, 2017-07-10

--- Log opened Mon Jul 10 00:00:49 2017
sukey1	[https://github.com/shogun-toolbox/shogun] New commit https://github.com/shogun-toolbox/shogun/commit/943d622bc73c78094e65680cad8bbfcfd4782454 by vigsterkr	03:55
sukey1	[https://github.com/shogun-toolbox/shogun] New commit https://github.com/shogun-toolbox/shogun/commit/cc5628d262fb09eac56b6bd8fa5d82df3776886f by vigsterkr	04:48
-!- mikeling [uid89706@gateway/web/irccloud.com/x-jathdrlvlcdzbkyj] has joined #shogun		06:07
-!- travis-ci [~travis-ci@ec2-54-211-136-94.compute-1.amazonaws.com] has joined #shogun		06:28
travis-ci	it's Viktor Gal's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: https://travis-ci.org/shogun-toolbox/shogun/builds/251854970	06:28
-!- travis-ci [~travis-ci@ec2-54-211-136-94.compute-1.amazonaws.com] has left #shogun []		06:28
@wiking	mikeling, here?	07:07
-!- iglesiasg [~iglesiasg@217.119.234.214] has joined #shogun		07:22
-!- mode/#shogun [+o iglesiasg] by ChanServ		07:22
mikeling	wiking: ping	07:36
mikeling	hey	07:36
@wiking	ok	07:37
@wiking	so i'm wondering how come the sample_null is giving that weird vector back	07:38
@wiking	mikeling, btw the vector of the same element	07:38
@wiking	is always the same?	07:38
mikeling	yes	07:38
@wiking	meaning the content of it is always the same when you run the unit test?	07:38
mikeling	always	07:38
@wiking	it's always a vector of -0.0188990416936576366	07:39
@wiking	right?	07:39
mikeling	yes	07:39
@wiking	i guess in the code there a set_global_seed in the unit test?	07:39
mikeling	yes, we have	07:39
mikeling	set_global_seed(12345);	07:39
@wiking	what happens if you change it to set_global_seed(1)	07:40
@wiking	?	07:40
mikeling	let me have a try	07:40
@wiking	the vector in compute_p_value	07:40
@wiking	is again the same?	07:40
mikeling	?????????? I don't understand now.	07:41
mikeling	as you said, the result stay the same	07:42
mikeling	still -0.0188990416936576366 after I change the seed to 1	07:42
mikeling	by set_global_seed(1)	07:42
mikeling	wiking: ^	07:42
@wiking	oh	07:45
@wiking	that is most weird	07:45
@wiking	so setting the seed does not change anything of that unit test?	07:45
mikeling	wiking: sorry, sorry, just a second	07:45
mikeling	I address the right test	07:45
@wiking	okey	07:47
mikeling	wiking: ok, it changed to 0.260531611740589142	07:47
@wiking	ok	07:47
@wiking	but again	07:47
@wiking	it's a vector of	07:47
@wiking	0.260531611740589142	07:47
@wiking	the elements are the same	07:47
mikeling	yes [0.260531611740589142,0.260531611740589142,0.260531611740589142,0.260531611740589142,0.260531611740589142,0.260531611740589142,0.260531611740589142,0.260531611740589142,0.260531611740589142,0.260531611740589142]	07:47
@wiking	ok	07:47
@wiking	that is strange	07:48
@wiking	i wonder what's happening	07:52
@wiking	in statistical_testing there's literally no prn	07:52
@wiking	*prng call right?	07:52
mikeling	wiking: here is the lldb debug process of mine https://gist.github.com/MikeLing/81d827cdbd811acbec54b0cbbefb6ea5	07:53
mikeling	I hind some output and some detail function call of compute_p_value	07:54
mikeling	* hide	07:54
@wiking	yeo	07:54
@wiking	CMeanShiftDataGenerator	07:54
@wiking	is using prng i guess	07:54
mikeling	yes, and that's the only place I found related to prng within QuadraticTimeMMD.perform_test_permutation_unbiased_full test	07:55
@wiking	result[i]=CMath::randn_double	07:55
@wiking	that's in	07:55
mikeling	it been changed to m_rng->std_normal_distrib();	07:56
@wiking	features/streaming/generators/MeanShiftDataGenerator.cpp	07:56
@wiking	yes	07:56
mikeling	in this line of pr https://github.com/shogun-toolbox/shogun/pull/3906/files#diff-0b5192ff8ca1dff8cad9d9d6384da391L69	07:57
@wiking	yes	07:57
mikeling	mmm, it should be fine. Since the randn_double just return what std_normal_distrib() get https://github.com/shogun-toolbox/shogun/blob/be8e005da18d5bb6f6383013158dd0fc4d5d20ba/src/shogun/mathematics/Math.h#L1127-L1130	07:59
@wiking	yep	08:00
@wiking	i'm really puzzled	08:00
@wiking	mikeling, okey	08:12
@wiking	can i ask you to debug something	08:12
@wiking	pl	08:12
@wiking	plz	08:12
mikeling	sure	08:12
@wiking	so this is running right the code of CQuadraticTimeMMD	08:13
@wiking	can you do a breakline here	08:13
@wiking	https://github.com/shogun-toolbox/shogun/blob/develop/src/shogun/statistical_testing/QuadraticTimeMMD.cpp#L573	08:14
@wiking	and tell me which of the two methods are being called?	08:14
@wiking	NAM_MMD2_SPECTRUM or NAM_PERMUTATION	08:14
mikeling	NAM_PERMUTATION	08:15
@wiking	ok	08:15
@wiking	just a sec	08:15
mikeling	wiking: defined in here https://github.com/shogun-toolbox/shogun/blob/develop/tests/unit/statistical_testing/QuadraticTimeMMD_unittest.cc#L390	08:16
mikeling	the NAM_PERMUTATION	08:16
@wiking	ah ok	08:16
@wiking	cool	08:16
@wiking	just a sec	08:16
@wiking	breakline here https://github.com/shogun-toolbox/shogun/blob/feature/random-refactor/src/shogun/statistical_testing/QuadraticTimeMMD.cpp#L364	08:17
@wiking	and print the content of result.vector	08:17
mikeling	oh, I already do that in the morning, I believe it will be a vector contained same elements also	08:19
mikeling	leet me verify it	08:19
@wiking	ok	08:20
@wiking	mikeling, okey.... so another story	08:24
@wiking	if you put this on break does it call it	08:24
@wiking	https://github.com/shogun-toolbox/shogun/blob/develop/src/shogun/statistical_testing/internals/mmd/PermutationMMD.h#L201	08:25
mikeling	let me see, just a second	08:25
@wiking	k	08:26
mikeling	BTW, the result vector is vector=[-0.00157492,-0.00157492,-0.00157492,-0.00157492,-0.00157492,-0.00157492,-0.00157492,-0.00157492,-0.00157492,-0.00157492]	08:26
@wiking	ok so same elements	08:27
@wiking	ok	08:27
mikeling	yes	08:27
mikeling	wiking: precompute_permutation_inds() of PermutationMMD has been called	08:28
@wiking	k	08:28
@wiking	now my question is	08:28
@wiking	if you step till https://github.com/shogun-toolbox/shogun/blob/develop/src/shogun/statistical_testing/internals/mmd/PermutationMMD.h#L203	08:28
@wiking	can you print m_permuted_inds.vector	08:29
@wiking	?	08:29
mikeling	sure, a second	08:30
@wiking	then step 2 lines down into till https://github.com/shogun-toolbox/shogun/blob/develop/src/shogun/statistical_testing/internals/mmd/PermutationMMD.h#L206	08:30
@wiking	and print again m_permuted_inds.vector	08:30
mikeling	in the beginning, the m_permuted_inds is vector=[0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0]	08:34
@wiking	ok	08:34
mikeling	and afte cacluated	08:34
mikeling	it's ector=[12,45,25,11,21,49,40,2,36,1,19,42,39,31,6,13,32,29,16,47,5,10,14,3,15,26,22,17,24,20,33,7,37,46,30,48,34,18,41,23,0,43,9,44,8,38,4,27,28,35]	08:34
@wiking	after std::iota	08:34
@wiking	or after permutation?	08:34
mikeling	ok	08:34
@wiking	after iota it should be a that	08:34
@wiking	what you have showed	08:34
mikeling	wiking: after the whole loop https://github.com/shogun-toolbox/shogun/blob/develop/src/shogun/statistical_testing/internals/mmd/PermutationMMD.h#L214	08:34
mikeling	it's [12,45,25,11,21,49,40,2,36,1,19,42,39,31,6,13,32,29,16,47,5,10,14,3,15,26,22,17,24,20,33,7,37,46,30,48,34,18,41,23,0,43,9,44,8,38,4,27,28,35]	08:35
@wiking	what's the value of m_save_inds	08:35
@wiking	?	08:35
@wiking	and what's the value of m_inverted_permuted_inds	08:36
@wiking	ok can you do one thing	08:37
@wiking	after allocate_permutation_inds();	08:37
mikeling	sfter iota, it's [0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49]	08:37
mikeling	after	08:37
mikeling	after permute it's [12,45,25,11,21,49,40,2,36,1,19,42,39,31,6,13,32,29,16,47,5,10,14,3,15,26,22,17,24,20,33,7,37,46,30,48,34,18,41,23,0,43,9,44,8,38,4,27,28,35]	08:38
@wiking	before the for loop create a local prng: auto prng = std::unique_ptr....	08:38
@wiking	and pass it to https://github.com/shogun-toolbox/shogun/blob/develop/src/shogun/statistical_testing/internals/mmd/PermutationMMD.h#L206	08:38
@wiking	as a second argument	08:38
@wiking	basically instead of CMath::permute(m_permuted_inds);	08:38
@wiking	do CMath::permute(m_permuted_inds, prng.get());	08:38
@wiking	because with the fixed seed CMath::permute without passing prng will always do the very same permutation	08:41
@wiking	because inside CMath::permute you always create a new prng	08:41
@wiking	and that prng has always the same seed	08:41
@wiking	if the seed is set globally	08:41
@wiking	and i think here we want something else :)	08:42
@wiking	we want really random stuff	08:42
@wiking	after each call	08:42
@wiking	in this example m_num_null_samples =?	08:43
mikeling	ok, the m_num_null_samples == 10	08:44
mikeling	I create a prng by auto prng = std::unique_ptr<CRandom>(new CRandom());	08:44
mikeling	and after CMath::permute(m_permuted_inds, prng.get());	08:44
@wiking	yes	08:44
mikeling	m_permuted_inds.display_vector() give me	08:44
mikeling	vector=[12,45,25,11,21,49,40,2,36,1,19,42,39,31,6,13,32,29,16,47,5,10,14,3,15,26,22,17,24,20,33,7,37,46,30,48,34,18,41,23,0,43,9,44,8,38,4,27,28,35]	08:44
@wiking	yeah but after each loop	08:44
@wiking	they should be different	08:45
@wiking	right?	08:45
mikeling	mmm, let me see	08:45
@wiking	first one will be the same	08:45
@wiking	that you had previously	08:45
@wiking	when n = 0	08:45
@wiking	when when n =1 or the rest	08:46
@wiking	m_permuted_inds.display_vector() should give you different results	08:46
mikeling	wiking: it gives me same result for each loop	08:46
@wiking	mmm that should not be the case	08:47
@wiking	if you pass the prng	08:47
@wiking	this should be the case if you dont pass the prng after each loop	08:47
mikeling	like this https://pastebin.mozilla.org/9026742	08:47
@wiking	ok oh	08:48
@wiking	wrong place	08:48
@wiking	:)	08:48
@wiking	auto prng = std::unique_ptr<CRandom>(new CRandom());	08:48
@wiking	should be outside of the for loop	08:48
mikeling	oh, sorry	08:48
@wiking	just before https://github.com/shogun-toolbox/shogun/blob/develop/src/shogun/statistical_testing/internals/mmd/PermutationMMD.h#L203	08:48
@wiking	btw i hope you know that in lldb you can print an SGVector content like this	08:49
@wiking	say in case of m_permuted_inds	08:49
@wiking	jsut type: p m_permuted_inds.vector	08:49
@wiking	and it'll print the content of m_permuted_inds	08:49
@wiking	anyhow	08:49
@wiking	once you put the prng outside of the for loop	08:49
@wiking	and you pass that to permute	08:50
@wiking	then each loop you should have a different permutation	08:50
mikeling	wiking: yes, it is	08:53
mikeling	https://pastebin.mozilla.org/9026744	08:53
@wiking	ok good	08:54
mikeling	each loop will give me different result	08:54
@wiking	now if we go back to that vector	08:54
@wiking	https://github.com/shogun-toolbox/shogun/blob/develop/src/shogun/statistical_testing/HypothesisTest.cpp#L79	08:54
@wiking	that shouldnt be anymore like full of same elements	08:54
@wiking	right?	08:54
@wiking	it should be some random numbers	08:54
mikeling	yes! it's vector=[-0.0188990416936576366,0.289082631468772888,-0.000296307698590680957,0.0906120724976062775,-0.0715163871645927429,-0.0252181515097618103,-0.213874764740467072,0.412667155265808105,0.154181867837905884,-0.204684354364871979]	08:55
@wiking	cool	08:55
mikeling	without same element anymore	08:55
@wiking	perfect	08:55
@wiking	this for sure was a problem	08:55
@wiking	so now what you need to do is the following	08:56
@wiking	it's only CMath::permute that is still using random	08:56
@wiking	in a way right?	08:56
mikeling	yes	08:57
@wiking	ok so now you'll have to check everywhere	08:57
@wiking	in the code	08:57
mikeling	like https://pastebin.mozilla.org/9026745	08:57
@wiking	where CMath::permute is used	08:57
mikeling	ok	08:58
mikeling	and add a prng to them?	08:58
@wiking	so yeah i mean in case	08:58
@wiking	there's a for loop and not just 1 permutation	08:58
@wiking	then you'll have to do this	08:58
@wiking	but maybe the best is	08:59
@wiking	that you force	08:59
@wiking	CMath::permute(SGVector<T> v, CRandom* rand)	08:59
@wiking	so theres' no part where rand = NULL	09:00
mikeling	ok, I see	09:00
@wiking	and maybe define another function	09:00
@wiking	that is only	09:00
@wiking	CMath::permute(SGVector<T> v)	09:00
@wiking	but yeah	09:00
@wiking	we can leave it as it is now	09:00
@wiking	but you need to go through the code	09:00
@wiking	and see where CMath::permute is being used	09:00
@wiking	and where CMath::permute is called in a loop	09:01
@wiking	then you'll have to do the trick we did this time	09:01
@wiking	note!	09:01
@wiking	when you do this patch	09:01
@wiking	do note amend or squash	09:01
@wiking	put it on to of	09:01
@wiking	39920817b24cac3fb1c6098b8ce484f3a9d8de7d	09:01
@wiking	ok/	09:01
@wiking	so i can see separately this part of the code	09:01
mikeling	ok, I will keep that in mind	09:02
mikeling	thank you	09:02
@wiking	cool	09:02
@wiking	i dont know if the unit test now passes or not	09:02
@wiking	but this definitely was a problem	09:02
@wiking	but i would suppose that now after using the same seed	09:02
@wiking	it should actually has the same output	09:03
@wiking	unless of course prng was called before	09:03
@wiking	in some other places :S	09:03
-!- geektoni [~geektoni@93-34-128-38.ip49.fastwebnet.it] has joined #shogun		09:25
sukey1	[https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3919 opened by geektoni	09:47
sukey1	[https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3919 synchronized by geektoni	10:00
@wiking	gsocers is it an odd or even week/	10:47
@wiking	?	10:47
@wiking	i think an even :D	10:47
@wiking	so prepare the blog posts plz ^ geektoni micmn mikeling	10:48
@wiking	this was just a gentle reminder :P	10:48
@wiking	lisitsyn, btw i know it sounds at first a bit stupid but i'm wondering whether in case of the pip thingy (and as well conda) we shouldn't just fuzzball 1 .so	10:53
@wiking	meaning that we would take libshogun.a and put that into swig .so	10:53
@wiking	basically this is what tf does	10:54
@wiking	as i believe we have this layer of separation of libshogun.so and swig interfaces	10:56
@wiking	because we wanna actually be able to ship in one ssytem	10:56
@wiking	multiple interfaces	10:56
@wiking	but honestly	10:56
@wiking	i think the general user would like to use either R or python or whatever interface	10:57
@wiking	and have it as one	10:57
@wiking	without the hassle of having libshogun.so around	10:57
-!- HeikoS [~heiko@host-92-0-169-11.as43234.net] has joined #shogun		11:03
-!- mode/#shogun [+o HeikoS] by ChanServ		11:03
lisitsyn	wiking: .a into interface .so?	11:04
@wiking	yes	11:04
lisitsyn	I agree	11:04
lisitsyn	even if user wants multiple interfaces	11:05
lisitsyn	wiking: it also reduces the need to provide DYLD_LIBRARY_PATH	11:05
@wiking	hehehe indeed :)	11:05
@wiking	o fcourse one can lament	11:06
@wiking	about the story of why one uses shared libs	11:06
@wiking	and memory	11:06
@wiking	and lalalla	11:06
@wiking	but in the BULLSHIT age of docker	11:06
lisitsyn	shared libs were useful when we had 1mb	11:06
@wiking	NOBODY cares anymore of shared libraries and that stuff	11:06
@wiking	man this	11:06
@wiking	this	11:06
@wiking	http://money.cnn.com/2017/07/07/technology/google-dandelion/index.html	11:06
@wiking	THIS IS THE FUCKING BULLSHIT	11:06
@wiking	this technology is 40 years old at least	11:07
@wiking	like AC	11:07
@wiking	i have the very same system installed in serbia (NOTE SERBIA)	11:07
@wiking	for 4k EUR	11:07
@wiking	in my house	11:07
@wiking	when things are going out of hands....	11:08
@wiking	anyhow yeah	11:08
@wiking	maybe we should just forget the story	11:08
lisitsyn	hah	11:08
@wiking	and interface libs	11:08
@wiking	are actually the whole thing in one	11:08
lisitsyn	wiking: there is an idea from some system	11:09
lisitsyn	let me lookup	11:09
@wiking	anyhow the official packages	11:09
@wiking	if we would ever have one	11:09
@wiking	like R or pypi	11:09
@wiking	etc	11:09
@wiking	they would anyways would need to ship something like that	11:10
@wiking	if they are precompiled	11:10
lisitsyn	http://nixos.org	11:13
lisitsyn	wiking: huh found	11:13
@wiking	packaging?	11:13
@wiking	i think i saw this one	11:13
lisitsyn	wiking: so there is a graph	11:14
lisitsyn	of A -> B and so on	11:14
lisitsyn	but no versions	11:14
lisitsyn	so you fix the versions and try to test the whole	11:14
lisitsyn	and then deploy this fixed thing	11:14
lisitsyn	this reminds me that there are alternatives to using dynamic libs all around	11:15
lisitsyn	:)	11:15
@wiking	:)	11:15
lisitsyn	wiking: we link everything static way here at my job	11:16
lisitsyn	this makes you independent of system you're running at	11:16
@wiking	\o/	11:16
lisitsyn	it is just x86_64 for you	11:16
@wiking	yeah i mean	11:19
@wiking	when you told me	11:19
@wiking	that you static ompile your code	11:19
@wiking	into the interpreter	11:19
@wiking	taht's some heavy weight shit :D	11:19
@wiking	./bazdmeg	11:20
@wiking	and everything is running	11:20
@wiking	:D	11:20
@wiking	./shogun	11:20
@wiking	:DDDDDDDDDDDDd	11:20
@wiking	which is a python interpreter	11:20
@wiking	and all that in one	11:20
@wiking	HeikoS, what do you think &	11:21
@wiking	^	11:21
@wiking	shall we just ship a whole python interpreter with shogun	11:21
@HeikoS	wiking: think about?	11:21
@wiking	and then everybody is happy? :D	11:21
@HeikoS	wiking: Lol	11:21
@HeikoS	what?	11:21
@HeikoS	like the guys from graphlab?	11:21
@wiking	well just thinkering	11:21
@wiking	yeah	11:22
@wiking	those people	11:22
@wiking	the Tika	11:22
@wiking	tuka	11:22
@wiking	puri	11:22
@wiking	what was their last name before apple bought them?	11:22
@wiking	turi!	11:22
@HeikoS	I dont know	11:22
@HeikoS	whatever	11:22
@wiking	man that was even better	11:22
@wiking	c++ compiler with the whole code	11:22
@wiking	\o/	11:22
@HeikoS	but is there a specific reason why we want to join this shit-wave?	11:22
@wiking	HeikoS, anyhow rather about this	11:22
@HeikoS	if we integrate with conda, that is kind of that no?	11:23
@wiking	[16:51] <wiking> lisitsyn, btw i know it sounds at first a bit stupid but i'm wondering whether in case of the pip thingy (and as well conda) we shouldn't just fuzzball 1 .so	11:23
@wiking	[16:51] <wiking> meaning that we would take libshogun.a and put that into swig .so	11:23
@wiking	[16:51] <wiking> basically this is what tf does	11:23
@wiking	mmm Dougal does it a bit differently	11:23
@wiking	and he still uses the 2 layer thing	11:23
@wiking	meaning core libshogun.so and shogun.so interface lib	11:23
@wiking	but i'm saying that i dont see really nowadays a point	11:23
@wiking	into this nice engineering	11:23
@wiking	because NOBODY cares anymore about c++ api	11:24
@wiking	and shared libraries	11:24
@wiking	they just wanna have a module that they could	11:24
@wiking	conda install	11:24
@wiking	pip install	11:24
@wiking	R install	11:24
@wiking	whetever install	11:24
@wiking	and that magically works	11:24
@wiking	which is OF COURSE	11:24
@wiking	normal	11:24
@HeikoS	mmmh	11:24
@wiking	i mean check out TF	11:24
@HeikoS	and if you want to use shogun with your existing python installation?	11:24
@wiking	they dont fucking ship c++ lib anywhere	11:24
@HeikoS	all these libs that are so easy to install	11:25
@HeikoS	they always create such a mess downstream in your workflow	11:25
@wiking	nono forget the ship with an interpreter	11:25
@wiking	that was rather a joke	11:25
@wiking	i'm saying that maybe we should start shipping 1 shared lib	11:25
@wiking	per interface	11:25
@HeikoS	ah	11:25
@HeikoS	rather than c++ lib and then python lib	11:25
@wiking	yes	11:25
@HeikoS	got it	11:25
@wiking	we would just fuzzball everything	11:26
@wiking	into one .so	11:26
@wiking	and that's like python's shogun.so	11:26
@wiking	then R's shogun.so	11:26
@wiking	etc etc etc	11:26
@HeikoS	that would make running things a bit easier	11:26
@wiking	no separate libshogun.so	11:26
@wiking	that you need to LD_LIBRARY PATh	11:26
@HeikoS	no more ld path setting and so	11:26
@wiking	etc etc etc	11:26
@wiking	yeye	11:26
@HeikoS	but there would still be the stnadalone libshogun.so	11:26
@wiking	it's just one fuckign switch :)	11:26
@wiking	well you could have one	11:26
@HeikoS	but it just a separate thing from the python one	11:26
@wiking	but your python .so wouldn't use it	11:27
@wiking	your c++ code could of course use it	11:27
@wiking	i mean in the case	11:27
@wiking	when you wanted to have on one system	11:27
@HeikoS	this also makes packaging a bit easier I guess	11:27
@wiking	R python ruby and jvm	11:27
@HeikoS	eg for .deb	11:27
@wiking	and all those interfaces	11:27
@HeikoS	noi more depoendencies	11:27
@wiking	mmm	11:27
@wiking	on deb it's tricky	11:27
@wiking	:D	11:27
@wiking	but yeah	11:27
@wiking	lets rather concetrate	11:27
@wiking	on language packages	11:28
@HeikoS	yeah	11:28
@HeikoS	agreed on that	11:28
@wiking	like gem, maven, pypi, R	11:28
@wiking	etc	11:28
@HeikoS	we need more coverage for the languages	11:28
@wiking	so those things would anyways need a libshogun.so	11:28
@wiking	shipped somehow	11:28
@HeikoS	and ease of install	11:28
@wiking	because you cannot assume	11:28
@wiking	that they'll have it installed it somewhow	11:28
@wiking	so either you put a separate libshogun.so in there	11:28
@wiking	next to the language's swig interface so	11:28
@wiking	or just fuck it	11:29
@wiking	and put everything into one	11:29
@wiking	:)	11:29
lisitsyn	again, I think it is a good idea	11:29
@wiking	see tensorflow	11:29
@wiking	they do this	11:29
lisitsyn	we just change our targets a bit	11:29
lisitsyn	the worst thing that happens in the case	11:29
@wiking	pypi package of tensorflow	11:29
lisitsyn	is bloat of size	11:29
@wiking	ships one fuzzball of .so	11:29
lisitsyn	who cares	11:29
@wiking	that is the python interface including the c++ code	11:29
@wiking	*shared lib of tf	11:29
@wiking	and that's all	11:30
@wiking	on our end	11:30
@wiking	this is not such a horror job	11:30
@wiking	as we just switch the linking	11:30
lisitsyn	no it is just changing the linking	11:30
@wiking	of the swig interface	11:30
@HeikoS	DO IT	11:30
@wiking	with shogun	11:30
@HeikoS	DO IT	11:30
@HeikoS	DO IT	11:30
lisitsyn	hh	11:30
@wiking	:)	11:30
@HeikoS	:D	11:30
@wiking	muezli	11:31
@HeikoS	mikeling: I havent got the global function calls working yet, but it is close. Maybe later today	11:31
@wiking	that's all	11:31
@HeikoS	hey btw	11:31
@HeikoS	wiking, lisitsyn shall we release BSD soon?	11:31
@HeikoS	lisitsyn: you have some header magic?	11:31
@wiking	HeikoS, feature/eigen-lapack	11:31
lisitsyn	HeikoS: I am thinking of way to clean up licences	11:31
@HeikoS	remove all gpl headers	11:31
@wiking	HeikoS, if you use linux that should work already....	11:31
@HeikoS	put BSD :D	11:31
lisitsyn	I mean yes it is easy to get list of authors	11:31
lisitsyn	but some things would be strange	11:31
@HeikoS	wiking: ah cool	11:31
@HeikoS	wiking: will check it out	11:32
lisitsyn	there are some glitches	11:32
@wiking	HeikoS, problem is that there seems to be a bug with cmake's Lapack detector	11:32
@wiking	HeikoS, so i could make it detect (even with -DBVA_VENDOR=Intel)	11:32
@wiking	conda's mkl	11:32
@wiking	conda's openblas works...	11:32
lisitsyn	HeikoS: we have things like local email (master@mymachine) and stuff like that	11:32
@HeikoS	wiking: ah that is good, ok anything I can push dougal to help with?	11:32
@wiking	HeikoS, anyhow you can make eigne to use your lapakc backend	11:32
@HeikoS	lisitsyn: email not needed	11:33
@wiking	HeikoS, we need some windows help	11:33
@HeikoS	lisitsyn: just names	11:33
lisitsyn	HeikoS: same thing with names	11:33
@wiking	HeikoS, as there's a linking problem	11:33
@wiking	and i dont get what the fuck	11:33
lisitsyn	HeikoS: so I will come up with some cleaning	11:33
@HeikoS	wiking: dougal doesnt have a windows machine ...	11:33
@wiking	yeah	11:33
@wiking	i know	11:33
@wiking	but he did everything :))	11:33
@HeikoS	lisitsyn: doesnt have t obe perfect	11:33
@wiking	so it's cool	11:33
@HeikoS	just kill all those headers	11:33
@HeikoS	wiking: great	11:33
@HeikoS	should announce that and acknowledge him once it is working	11:34
@wiking	yes	11:34
@wiking	news :)	11:34
@wiking	man we didn't udpate news for ages	11:34
@wiking	:DDDD	11:34
@wiking	neither did our nice students	11:34
@wiking	:P	11:34
@HeikoS	wiking: yeah always good to do that straight away	11:34
@wiking	yep	11:34
@wiking	my fault	11:34
lisitsyn	wiking: FAKE_NEWS	11:34
@wiking	should have been warning people	11:34
@HeikoS	push the guys :)	11:34
@wiking	lisitsyn, FAAAAAAAAAAAAAAAAAAAAAKE	11:34
@HeikoS	wiking: so the bsd files	11:34
@HeikoS	they are all in this new folder now	11:35
@wiking	you mean gpl	11:35
@HeikoS	shall I move that in a separate repo?	11:35
@HeikoS	yeah	11:35
@wiking	mmm ah yeah	11:35
@wiking	unfortunately we'll hav eto do that	11:35
@wiking	to comply with gpl	11:35
@HeikoS	sure	11:35
@HeikoS	how to reference?	11:35
@HeikoS	cmake git clones it?	11:35
@wiking	submodule is fine	11:35
@wiking	afaik	11:35
lisitsyn	oh no submodules again	11:35
@HeikoS	ah no	11:35
@HeikoS	no submodule	11:35
@wiking	lisitsyn, we have to	11:35
lisitsyn	sheeat	11:35
@wiking	or i mean	11:35
@HeikoS	can't we just give a path?	11:35
@wiking	we can just not do anything	11:35
@wiking	:DDDDDDDDDDDD	11:36
@wiking	and fuck gpl	11:36
@wiking	:P	11:36
lisitsyn	submodule is better than path for sure	11:36
@wiking	but man i mean	11:36
lisitsyn	what do you mean	11:36
@HeikoS	but then we need to constantly update that thing	11:36
@wiking	we already have like n+1 request	11:36
@wiking	why MKL is not working	11:36
@wiking	on debian packages	11:36
@wiking	...	11:36
@wiking	(svmlight)	11:36
@wiking	HeikoS, yeah like with data	11:36
@wiking	i mean with cmake	11:36
@HeikoS	wiking, lisitsyn ok then I will submodule it	11:36
lisitsyn	but we stop developing gpl right?	11:36
@wiking	you'd have to do the same	11:36
@HeikoS	lisitsyn: no	11:36
lisitsyn	I mean we don't update it anyway	11:36
lisitsyn	I am not fancy of adding more gpl deps	11:37
lisitsyn	are you?	11:37
@HeikoS	lisitsyn: no	11:37
@HeikoS	but still maintain	11:37
lisitsyn	yes maintain is like one commit per year anyway	11:37
@HeikoS	sure	11:37
@wiking	:>	11:37
@HeikoS	so then	11:37
@HeikoS	submodule the gpl folder and that's it right?	11:37
@wiking	yeah that should do it	11:37
@wiking	and since we are lazy fucks	11:38
@wiking	we'll never update it	11:38
@wiking	:D	11:38
@HeikoS	haha	11:38
@wiking	HeikoS, and what by default bsd build?	11:38
lisitsyn	yeah that's what I mean	11:38
lisitsyn	hah	11:38
@wiking	HeikoS, btw what happened with unit tests? D:	11:38
@wiking	HeikoS, you'd move those as well :D	11:38
@HeikoS	wiking: no why?	11:38
@wiking	or fuck that shit	11:38
@wiking	?	11:38
lisitsyn	why?	11:38
@HeikoS	guarded (.)	11:38
lisitsyn	tests are not gpl	11:38
@wiking	:)	11:38
@HeikoS	8-)	11:38
lisitsyn	:P	11:38
@wiking	ok i mean i dont wanna go all lawyer here	11:38
@wiking	so yeah let's assume that it's fine	11:39
@wiking	:)	11:39
lisitsyn	I think it is fine	11:39
lisitsyn	oh that license is soooo	11:39
@HeikoS	they contain includes of non-existing files that are never given to the compiler	11:39
@wiking	i mean i dont know	11:39
@wiking	its just i guess	11:39
@HeikoS	GPL doesnt pass that barrier	11:39
@wiking	it's border line	11:39
lisitsyn	I don't believe GPL can touch this	11:39
@wiking	but who knows :)	11:39
@HeikoS	mc hammer agrees	11:39
lisitsyn	yes	11:39
@wiking	k	11:39
@HeikoS	good	11:40
lisitsyn	mcmc hammer!	11:40
@HeikoS	lisitsyn: man	11:40
@HeikoS	I wanted to call my algorithm like that	11:40
lisitsyn	good name	11:40
@HeikoS	but then I found out some physics guys had that already	11:40
lisitsyn	haha yes	11:40
@HeikoS	but my repo is still called like that :D	11:40
@HeikoS	ok then	11:41
@HeikoS	Ill cycle to work now	11:41
lisitsyn	HeikoS: I explained the concept of 'test' set with turning on that song	11:41
@HeikoS	talk later	11:41
lisitsyn	:D	11:41
@HeikoS	:D	11:41
lisitsyn	HeikoS: that's the only sunny day in moscow today so the same for me	11:42
lisitsyn	going to take some bike	11:42
lisitsyn	:P	11:42
-!- travis-ci [~travis-ci@ec2-54-81-148-90.compute-1.amazonaws.com] has joined #shogun		11:44
travis-ci	it's Heiko Strathmann's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: https://travis-ci.org/shogun-toolbox/shogun/builds/251772035	11:44
-!- travis-ci [~travis-ci@ec2-54-81-148-90.compute-1.amazonaws.com] has left #shogun []		11:44
@wiking	lisitsyn, :DDDDDDDDDDDDDDDDDDDDDDd	11:45
@wiking	lisitsyn, only sunny day :D	11:45
lisitsyn	wiking: it seems the US is testing climatic sanctions because we have no summer this time	11:47
@wiking	:))))	11:47
-!- HeikoS [~heiko@host-92-0-169-11.as43234.net] has quit [Quit: Leaving.]		11:53
-!- HeikoS [~heiko@untrust-out.swc.ucl.ac.uk] has joined #shogun		12:54
-!- mode/#shogun [+o HeikoS] by ChanServ		12:54
sukey1	[https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3920 opened by micmn	13:21
@wiking	micmn, how do u control the num of threads?	13:32
micmn	it's just linalg	13:35
micmn	wiking: ^	13:35
@wiking	micmn, -> eigen decides?	13:37
micmn	yep, standard matrix product	13:37
@wiking	k	13:37
@wiking	and i guess u have 4 vcores	13:37
@wiking	that's why 4 threadss?	13:37
micmn	yes	13:37
@wiking	cool	13:41
@wiking	the power of immutable structures :D	13:41
@wiking	now it's just time to get unlimited ram	13:42
@wiking	\o	13:42
micmn	XD	13:44
@wiking	ok so	13:51
@wiking	here	13:51
@wiking	HERE IS THE PURE GENIUS OF THE INTERNETZ	13:51
lisitsyn	oh	13:51
lisitsyn	:D	13:51
@wiking	there's just really nothing after this horizon	13:51
@wiking	lisitsyn, https://eros.vision/	13:51
@wiking	:DDDDDDDDDDDDDDDDDDDDDDDDDD	13:51
@wiking	i mean man	13:51
@wiking	GENIOOOOOOOOOOOUUUUUUUUUUUUS! :)	13:52
lisitsyn	hahah	13:52
@wiking	i mean fuck all the things	13:52
@wiking	ICO for biatch	13:52
@wiking	:D	13:52
@wiking	and THEY RAISED ALREADY 100k	13:52
lisitsyn	haha	13:53
@wiking	mmm no	13:53
@wiking	180k	13:53
@wiking	LOL	13:53
@wiking	i mean can you believe this	13:53
@wiking	this shows the infinite craziness of the internetz	13:54
@wiking	:D	13:54
@iglesiasg	Hola!	13:54
@iglesiasg	HeikoS: checking the gpl stuff: https://github.com/shogun-toolbox/shogun/blob/develop/src/shogun/kernel/string/DistantSegmentsKernel.cpp#L8	13:54
@wiking	iglesiasg, ^	13:54
@iglesiasg	wiking: about eros?	13:55
@wiking	yes	13:55
@iglesiasg	yeah, I just opened the website	13:55
@iglesiasg	still not fired fiuuu	13:55
@iglesiasg	xD	13:55
@wiking	:D	13:55
@HeikoS	iglesiasg: thanks	13:55
@HeikoS	iglesiasg: well spotted :)	13:56
@wiking	lisitsyn, best is	13:56
@wiking	lisitsyn, 2% revenue sharing	13:56
@wiking	:DDDDDDDD	13:56
@wiking	i mean essentialy if you invest	13:56
@wiking	you are a virtual pimp	13:56
@iglesiasg	HeikoS: there are a few more I think, let me dig into it	13:57
@iglesiasg	lisitsyn: can tapkee_shogun be updated to bsd? https://github.com/shogun-toolbox/shogun/blob/develop/src/shogun/lib/tapkee/tapkee_shogun.hpp#L3	14:00
lisitsyn	iglesiasg: for sure	14:00
@iglesiasg	great	14:00
@wiking	https://i.ytimg.com/vi/xLw5xDq4YfM/hqdefault.jpg	14:02
sukey1	[https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3921 opened by iglesias	14:09
sukey1	[https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3921	14:09
@wiking	micmn, C.o.V.E.R.I.D.G.E :)	14:12
@iglesiasg	HeikoS: brent slipped through too I think: https://github.com/shogun-toolbox/shogun/blob/develop/src/shogun/lib/external/brent.h	14:16
micmn	wiking: eh? :p	14:21
@wiking	micmn, nw... added the comment to the thing	14:22
@wiking	pr	14:22
@wiking	need some ++ on the codecov with some unit testz	14:22
@wiking	otherwise i'll send this cat your house https://i.ytimg.com/vi/xLw5xDq4YfM/hqdefault.jpg	14:22
micmn	ok sure xD	14:23
@wiking	thnx	14:23
-!- yamz [~myamada@2620:13:0:10:bdae:b37e:9286:1eb6] has quit [Read error: Connection reset by peer]		14:24
-!- yammer [~myamada@64.7.137.182] has joined #shogun		14:24
-!- yammer [~myamada@64.7.137.182] has quit [Read error: Connection reset by peer]		14:27
-!- yammer [~myamada@2620:13:0:10:ad90:f873:79fb:1bee] has joined #shogun		14:27
-!- geektoni [~geektoni@93-34-128-38.ip49.fastwebnet.it] has quit [Ping timeout: 260 seconds]		14:30
-!- yammerz [~myamada@64.7.137.182] has joined #shogun		14:31
-!- yammer [~myamada@2620:13:0:10:ad90:f873:79fb:1bee] has quit [Read error: Connection reset by peer]		14:32
-!- geektoni [~geektoni@93-34-128-38.ip49.fastwebnet.it] has joined #shogun		14:32
Trixis	just to check, if i need a custom kernel and im using java to interface with shogun, the only way to do it as of now is to compute the inner products in a separate function, and then set the kernel matrix in CustomKernel, right?	14:45
@wiking	Trixis, how's your fight?	14:45
@wiking	ok sooooo	14:45
@wiking	if you use java	14:45
@wiking	you actually would love to have this	14:45
@wiking	Trixis, https://github.com/shogun-toolbox/shogun/issues/3892	14:46
@wiking	this is possible	14:46
@wiking	and with python it work	14:46
@wiking	s	14:46
@wiking	and it is possible with swig	14:47
@wiking	it needs some patches around the swig code though	14:47
Trixis	wiking: otherwise, my fight is going pretty well so far, got the code that computes my features to interact with shogun properly	14:47
Trixis	wiking: i see	14:47
@wiking	but it is definitely something that should be possible	14:47
@wiking	once that'd be fixed	14:47
@wiking	you could derive a class from DirectorKernel	14:47
@wiking	in java	14:47
@wiking	and basically that'd act like any other kernel in shogun	14:48
@wiking	that's actually a great feature of swig	14:48
@wiking	unfortunately does not work yet with java :(	14:48
@wiking	i mean in shogun	14:48
@wiking	in general you can read about it on swig's website	14:49
@wiking	Trixis, but yeah	14:49
@wiking	otherwise until the magic thing happens	14:49
@wiking	you have no better choice atm	14:49
@iglesiasg	the LogPlusOne preprocessor is pretty great	14:49
@wiking	iglesiasg, KILL IT WITH FIRE	14:50
@wiking	or reimplement it	14:50
@wiking	:)	14:50
@wiking	you know you dictate to sergey what the thing should look like	14:51
@iglesiasg	wiking: I think we can drop this, right?	14:51
@wiking	sergey pretends he never saw the code :)	14:51
@wiking	and implements it	14:51
@wiking	done	14:51
@wiking	:)	14:51
@wiking	honestly	14:52
@wiking	this thing about gpl	14:52
@wiking	and being scared of	14:52
@wiking	is a bunch of lala land	14:52
@wiking	but yeah	14:52
@wiking	KILL IIIIIIIIIIIIIIIIIIIT WITH FIRE!	14:52
@wiking	look at linux kernel	14:52
@iglesiasg	mmm I suppose it might have some use in streaming features	14:52
@wiking	it's gpl :)	14:52
Trixis	wiking: kk, sounds good. Guess ill just wrap all kernels with a java interface, lol.	14:52
@iglesiasg	but it is a dense preprocessor so I am not sure it can be use in that case	14:52
@wiking	Trixis, :)	14:54
Trixis	at this stage, i just need it to work. This is science after all, software engineering can come later.	14:56
@wiking	hehehe	14:56
@wiking	what is your use case?	14:56
@wiking	nuclear reactor stabilizer?	14:56
Trixis	wiking: haha, no. Computational biology, compound property prediction	14:57
@wiking	\o/	14:57
@wiking	awesome	14:57
@wiking	dont forget we have a doi	14:57
@wiking	;)	14:57
Trixis	wiking: ofc, if i get something publishable out of this i sure will cite you	14:58
@wiking	:)	14:58
@wiking	lisitsyn, this is just for u "Get 10x Speedup in Tensorflow Multi-Task Learning using Python Multiprocessing	15:06
lisitsyn	wiking: LOL	15:06
@wiking	man	15:06
@wiking	the amoount of spam generated on internet	15:06
@wiking	is truly amazing	15:07
Trixis	the init method in CombinedKernel simply calls init, i.e. compute of each of the subkernels, right?	15:07
@wiking	you mean init(lhs, rhs)?	15:08
Trixis	yeah	15:08
@wiking	yeah should do that afaik	15:08
@wiking	http://shogun.ml/api/latest/CombinedKernel_8cpp_source.html	15:09
@wiking	but yeah	15:09
@wiking	init weights the subkernels	15:09
@wiking	and init's the kernel normalizers	15:10
@iglesiasg	if anybody can shed some light into how set_preprocessed works I am happy to hear :)	15:14
@wiking	que?	15:14
@wiking	it's magico	15:14
@wiking	dont ask questions that cannot be answered	15:14
@iglesiasg	examples/undocumented/python/serialization_complex_example.py	15:14
@wiking	:D	15:14
@iglesiasg	hahaha	15:14
@iglesiasg	feats.add_preprocessor(NormOne())	15:14
@iglesiasg	feats.add_preprocessor(LogPlusOne())	15:14
@iglesiasg	feats.set_preprocessed(1)	15:14
@iglesiasg	no idea what that is doing	15:15
@iglesiasg	if it is saying something like only apply the preprocessor in 1?!	15:15
* iglesiasg is afk for a few		15:16
@wiking	iglesiasg,	15:16
@wiking	/** i'th entry is true if features were already preprocessed with preproc i */	15:16
@wiking	CDynamicArray<bool>* preprocessed;	15:16
@wiking	so u can specify	15:16
@wiking	which preprocessors on the featues	15:16
@wiking	has already been applied	15:16
@wiking	i guess	15:16
@wiking	this is when you would extend your feature set?	15:16
@wiking	or i mean i dont get it :D)	15:16
@wiking	i dont understand why would you add a preprocessor	15:17
@wiking	that is already done on the data?	15:18
@wiking	especially that set_feature_matrix	15:18
@wiking	does not reset the preprocessed flag	15:18
@wiking	so say you take a densefeatures class	15:18
@wiking	do this magic limbo	15:18
@wiking	then you set_feature_matrix	15:19
@wiking	but you dont apply there that preprocessor	15:19
@wiking	then it's lalala	15:19
@wiking	i think only the masters can shed some light on this :D	15:19
@wiking	these is some heavy sorcery from mordor	15:20
@wiking	sonney2k, ^ :)	15:21
@wiking	iglesiasg, or maybe this is like a convenience function	15:33
@wiking	iglesiasg, where you can easily turn on and off preprocessors	15:33
@wiking	iglesiasg, cool stuff for parameter searching :D	15:33
@wiking	just flip a boolean	15:34
@wiking	right?	15:34
@wiking	instead of add_preprocessor	15:34
@wiking	remove_preprocessor	15:34
@wiking	lalalal	15:34
@wiking	i cannot think about anything else	15:34
* iglesiasg is back		15:35
@wiking	\/\o/\/	15:36
Trixis	wiking: hence i also have to also call init_normalizer to achieve an equivalent statement, right?	15:36
@wiking	Trixis, when?	15:37
@wiking	i mean if you call in combinedKernel.init	15:37
@wiking	then that's called	15:37
@wiking	after initing all the kernels	15:37
@wiking	of course in ccase you build yourself	15:37
Trixis	wiking: after i call init / set the kernel matrix for each of the subkernels	15:37
@wiking	the matrix	15:37
@iglesiasg	wiking: yeah, I agree with your thoughts on the magic limbo	15:37
@wiking	mmm idk Trixis	15:38
@wiking	i have to check the code	15:38
@wiking	but i mean you need to init the normalizers for sure	15:38
@wiking	but the normalizers later	15:38
@wiking	automagically applied	15:38
@wiking	ont the kernel matrix	15:38
@wiking	afaik	15:38
@wiking	iglesiasg, cool for param search really	15:38
@wiking	:)	15:38
@wiking	pretty styleish	15:39
@wiking	check git blame	15:39
@wiking	maybe htere's some meaningful log :D	15:39
Trixis	wiking: i mean, normalisation is it's own identity, hence calling it twice should still yield the same result, right?	15:43
@wiking	one would hope!	15:43
Trixis	hahaha	15:43
@wiking	but yeah i mean	15:43
@wiking	weird things happen osme time	15:44
@wiking	if u find a bug	15:44
@wiking	just report plz	15:44
Trixis	kk	15:44
Trixis	oh, actually, looks like i can just set the customkernel matrices beforehand, then call init in combined kernel, since it simply skips custom kernels	15:45
Trixis	well this simplifies it a lot	15:45
@wiking	man i wish there would be a way	15:47
@wiking	to put these conversations somehow	15:47
@wiking	easily to a structured data	15:47
@wiking	so that others later could understand this	15:47
Trixis	wiking: eh i might as well create an example from parts of the code, if you remind me in a couple of months	15:53
@wiking	Trixis, remind	15:57
@wiking	:)	15:57
-!- HeikoS [~heiko@untrust-out.swc.ucl.ac.uk] has quit [Ping timeout: 248 seconds]		16:06
sukey1	[https://github.com/shogun-toolbox/shogun] New commit https://github.com/shogun-toolbox/shogun/commit/6a21cf1cfd83f756ad07186401d4ceb93639b93a by karlnapf	16:11
-!- HeikoS [~heiko@eduroam-int-pat-8-96.ucl.ac.uk] has joined #shogun		16:23
-!- mode/#shogun [+o HeikoS] by ChanServ		16:23
@wiking	HeikoS, ye may	16:37
@wiking	or i meant hey man	16:37
@wiking	:)	16:37
@wiking	here?	16:37
@HeikoS	I am	16:37
@wiking	ok	16:37
@wiking	so apart from whether or not u wanna join to a reggee party	16:38
@wiking	about the iterators	16:38
@wiking	so i was thinking abou tthis	16:38
@wiking	a bit	16:38
@wiking	and looking into swig0	16:38
@HeikoS	so whats exactly up? not sure I follow :D	16:38
@wiking	so i mean are we sure that the whole story works nicely	16:39
@wiking	i mean i'm not sure why it wouldn't work right	16:39
@wiking	.begin()	16:39
@wiking	.end()	16:39
@wiking	.next()	16:39
@wiking	these are the stuff we wanna be able to expose to swig interfaces	16:40
sukey1	[https://github.com/shogun-toolbox/shogun] New commit https://github.com/shogun-toolbox/shogun/commit/36842e97c92e79ed4c1095ba66d6152a1d69f170 by karlnapf	16:40
@wiking	so this should work one would hope	16:40
@wiking	https://stackoverflow.com/a/27243461	16:40
@HeikoS	ah	16:40
@wiking	i just dont wanna end up where that we have	16:40
@wiking	all these amazing design	16:40
@wiking	and realise that swig is magico lando	16:40
@HeikoS	you think we are there yet .. thinking about exposingn this to swig	16:40
@wiking	and we cannot do it	16:40
@HeikoS	not sure we want to expose this to swig	16:41
@wiking	yeye but we would like to	16:41
@HeikoS	(iteartors)	16:41
@wiking	or?	16:41
@HeikoS	not so sure	16:41
@wiking	HeikoS, features?	16:41
@HeikoS	extract_feature_matrix()	16:41
@wiking	as we were talking about features as iterators	16:41
@wiking	yeye but that's shito	16:41
@HeikoS	I would expose conversion methods to the typemapped types	16:41
@wiking	if it's a streaming matrix	16:41
@wiking	*feature	16:41
@HeikoS	there can also be an API for that	16:41
@HeikoS	but internally	16:41
@HeikoS	we dont do that stuff	16:41
@wiking	i mean the good stuff with iterators	16:41
@wiking	that that allows us	16:41
@HeikoS	due to all the issues with types etc	16:42
@wiking	dontcare for sreaming or nonstreaming	16:42
@wiking	anyhow this is just my point	16:42
@wiking	that we should think a bit	16:42
@wiking	about swig in this context	16:42
@wiking	like for example	16:42
@wiking	some<>	16:42
@wiking	:)	16:42
@wiking	good stuff	16:42
@wiking	none usable as is atm from swig	16:42
@HeikoS	yeah ok	16:43
@HeikoS	I get the point	16:43
@HeikoS	whats up with some btw?	16:43
@wiking	well	16:43
@HeikoS	what about just typemapping it?	16:43
@wiking	mmm	16:43
@wiking	it's a template	16:43
@wiking	how do you typemap that	16:43
@wiking	?	16:43
@wiking	and anyways the problem there is that	16:43
@HeikoS	no you dont, you are right	16:43
@wiking	you would kind of start the work	16:43
@wiking	that siwg just partially done	16:44
@wiking	for std::shared_ptr	16:44
@wiking	i mean there's support for it	16:44
@wiking	but like not for all the languages etc	16:44
@HeikoS	mmh	16:44
@wiking	so if we go done on the path	16:44
@HeikoS	what about decoupling the SWIG api from the rest of shogun	16:44
@wiking	of Some<>	16:44
@wiking	ok but yeah i mean then you start down on the road of	16:44
@wiking	native *	16:44
@wiking	and some hax0r stuff internally	16:45
@wiking	prune to errors	16:45
@HeikoS	I mean like SWIG stays as it is (with reference counting etc)	16:45
@wiking	hax0r = c++Z feaures	16:45
@wiking	yeye i get it	16:45
@wiking	alhotugh	16:45
@wiking	i wanted to share THIS	16:45
@HeikoS	and internally we modify some to work with that?	16:45
@wiking	just a sec	16:45
-!- iglesiasg [~iglesiasg@217.119.234.214] has quit [Quit: leaving]		16:45
@HeikoS	wiking: I have another question btw: https://travis-ci.org/shogun-toolbox/shogun/jobs/252016173#L3503	16:46
@HeikoS	how to import global functions from java?	16:46
@wiking	HeikoS, https://github.com/shogun-toolbox/shogun/issues/2880	16:46
@wiking	so that talks about SGRefObject	16:47
@wiking	HeikoS, https://github.com/shogun-toolbox/shogun/issues/2880#issuecomment-127923640	16:47
@wiking	:)	16:47
@wiking	"did get removed for some reason by someone who knows why that	16:47
@wiking	was maybe not a good idea?"	16:47
@wiking	particularly like this one :D	16:47
@HeikoS	haha	16:48
@HeikoS	I think thoralf was that	16:48
@wiking	"Do not remember. Git logs say that Thorsten removed it. "Obsolete "."	16:48
@wiking	:D	16:48
@wiking	good shit	16:48
@HeikoS	haha :D	16:48
@wiking	:)	16:48
@HeikoS	easy to re-introduce	16:48
@HeikoS	so but I agree we have to think this through a bit more	16:48
@HeikoS	so whats the best approach as of nowß	16:49
@HeikoS	?	16:49
@HeikoS	and also see my question above .... any ideas?	16:49
@wiking	yes	16:49
@wiking	just trying to get u an answer	16:49
-!- HeikoS [~heiko@eduroam-int-pat-8-96.ucl.ac.uk] has quit [Quit: Leaving.]		16:50
-!- travis-ci [~travis-ci@ec2-54-204-59-182.compute-1.amazonaws.com] has joined #shogun		16:57
travis-ci	it's Heiko Strathmann's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: https://travis-ci.org/shogun-toolbox/shogun/builds/252016167	16:57
-!- travis-ci [~travis-ci@ec2-54-204-59-182.compute-1.amazonaws.com] has left #shogun []		16:57
-!- HeikoS [~heiko@untrust-out.swc.ucl.ac.uk] has joined #shogun		17:07
-!- mode/#shogun [+o HeikoS] by ChanServ		17:07
@HeikoS	wiking: sorry had to move place	17:08
@wiking	k	17:12
@wiking	HeikoS, so theres' this story	17:12
@wiking	%module (jniclassname="name") modulename	17:12
@wiking	If name is the same as modulename then the module class name gets changed from modulename to modulenameModule.	17:12
@wiking	so there should be a module shogun	17:13
@HeikoS	import org.shogun.MulticlassLabels;	17:13
@HeikoS	this works	17:13
@HeikoS	import org.shogun.get_global_io;	17:13
@HeikoS	this doesnt	17:13
@HeikoS	(it does though in python=	17:13
@HeikoS	)	17:13
@wiking	nono	17:13
@wiking	so there should be a clas	17:13
@wiking	shogun	17:13
@HeikoS	ah	17:14
@wiking	apparently	17:14
@HeikoS	import org.shogun.module.get_global_io;	17:14
@wiking	and that wraps the globals	17:14
@wiking	nono	17:14
@wiking	do not import	17:14
@HeikoS	"There is no such thing as a global Java function so global C functions are wrapped as static methods in the module class."	17:14
sukey1	[https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3920 synchronized by micmn	17:14
@wiking	get_global	17:14
@wiking	as u cannot import	17:14
@wiking	a method	17:14
@wiking	yes	17:14
@wiking	exactly	17:14
@wiking	so you wouold have	17:14
@wiking	org.shogun.shogun ?	17:14
@wiking	and then shogun.get_global_io()	17:14
@HeikoS	yes	17:14
@HeikoS	import org.shogun.shogun;	17:14
@HeikoS	that works	17:15
@HeikoS	ok let me try	17:15
@wiking	but i think	17:16
@wiking	nothing	17:17
@wiking	anyhow that should be it	17:17
@wiking	based on SWIG's documentation	17:17
@wiking	but maaaybe beaseu of that story in the	17:17
@HeikoS	http://www.swig.org/Doc2.0/SWIGDocumentation.html#Java_module_packages_classes	17:17
@wiking	docs	17:17
@HeikoS	24.3.2 Functions	17:17
@wiking	you might have shogunModule	17:17
@wiking	so org.shogun.shogunModule	17:17
@HeikoS	learning so much about shogun these days ;)	17:19
@wiking	:))))	17:20
@wiking	magical land	17:20
@wiking	ok so	17:21
@wiking	just to get everybody	17:21
@wiking	calm down and realise that yeah things are not perfect anywhere	17:21
@wiking	https://thepasteb.in/p/Q1hBZOjXKwrc8	17:21
@wiking	this is when trying to compile a 80billion dollar revenue's company sw on M$ windows	17:22
@wiking	i meant "company of a 80 billion USD revenue"	17:22
@wiking	...	17:22
@wiking	so if we look from this perspective	17:22
@wiking	we are doing pretty ok :D	17:23
sukey1	[https://github.com/shogun-toolbox/shogun] New commit https://github.com/shogun-toolbox/shogun/commit/92e3257ead97291e5e8c53aebdd8a16fbc490210 by karlnapf	17:23
@HeikoS	haha :D	17:23
@HeikoS	everyone has the same problems	17:24
@HeikoS	BUT WE FIX THEM	17:24
@wiking	indeed	17:24
@wiking	i mean its' funny i think	17:24
@wiking	to look at things this way	17:24
@wiking	man really that's a company of 80 billion USD	17:24
@wiking	with shitload of engineers	17:24
@wiking	and they fail on the same little shits	17:24
@wiking	distribution etc	17:24
@wiking	mmmm	17:28
-!- iglesias [~iglesias@wlan073045.mobiel.utwente.nl] has joined #shogun		17:28
@wiking	does anybody relised that TF	17:28
@wiking	actually cannot be built offline	17:28
@wiking	?	17:28
@wiking	muha :D	17:29
iglesias	without blaze?	17:29
@wiking	iglesias, what are u doing at the uni?	17:29
@wiking	iglesias, nono man ./configure script downloads a .js form the web	17:29
iglesias	wiking: yo soy un estudiante, de por vida	17:29
@wiking	and even if it downloaded it	17:29
@wiking	it would redonwload it again	17:29
@wiking	pura vida!	17:29
iglesias	aaah haha cool	17:30
@wiking	good shitz	17:30
@wiking	being forever student	17:30
iglesias	sí	17:30
iglesias	es bueno	17:30
@wiking	so yeah	17:31
@wiking	good shit	17:31
iglesias	wiking: I started a fun test with tf btw	17:31
@wiking	one could rest after seeing these stuff	17:31
iglesias	or challenge rather than test	17:31
@wiking	iglesias, tf is cool btw	17:32
iglesias	will I manage to build a hello world without blaze? :O	17:32
iglesias	just old plain g++ -I ... -l ...	17:32
@wiking	heheh	17:32
@wiking	have fun	17:32
@wiking	:)	17:32
iglesias	haha yeah	17:32
iglesias	I left it in blaze -s	17:32
@wiking	:>	17:32
iglesias	but it looked like it was building lot of stuff	17:33
@wiking	but i mean its good that we are struggling with the same shit :)	17:33
iglesias	not just the minimal example	17:33
iglesias	which I found funky btw since the last time I touched that directory I already did blaze sth	17:33
-!- HeikoS [~heiko@untrust-out.swc.ucl.ac.uk] has quit [Ping timeout: 258 seconds]		17:34
iglesias	anyhow, it is in the other machine, so I will check it tomorrow	17:34
sukey1	[https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3920 synchronized by micmn	17:34
iglesias	if you hear don't hear from me about it, I failed the challenge xD	17:34
@wiking	:>>>>>>>.	17:35
-!- HeikoS [~heiko@eduroam-int-pat-8-96.ucl.ac.uk] has joined #shogun		17:50
-!- mode/#shogun [+o HeikoS] by ChanServ		17:50
-!- iglesias [~iglesias@wlan073045.mobiel.utwente.nl] has quit [Quit: leaving]		18:07
-!- HeikoS [~heiko@eduroam-int-pat-8-96.ucl.ac.uk] has quit [Quit: Leaving.]		18:14
@wiking	lisitsyn, https://github.com/hioa-cs/IncludeOS	18:14
@wiking	:D	18:14
@wiking	lisitsyn, this is ++ version of your statically linking libraries together	18:15
@wiking	:D	18:15
@wiking	not only have the interpreter	18:15
@wiking	but the whole fukcing os into one thing	18:15
@wiking	:D	18:15
@wiking	"Extreme memory footprint: A minimal bootable image, including bootloader, operating system components and a complete C++ standard library is currently 707K when optimized for size."	18:16
@wiking	:)	18:16
@wiking	man this is crazy	18:17
@wiking	:))))))))))))))	18:17
@wiking	geektoni, killed it with fire?	18:21
-!- HeikoS [~heiko@untrust-out.swc.ucl.ac.uk] has joined #shogun		18:22
-!- mode/#shogun [+o HeikoS] by ChanServ		18:22
-!- travis-ci [~travis-ci@ec2-54-81-148-90.compute-1.amazonaws.com] has joined #shogun		18:23
travis-ci	it's Heiko Strathmann's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: https://travis-ci.org/shogun-toolbox/shogun/builds/252044729	18:23
-!- travis-ci [~travis-ci@ec2-54-81-148-90.compute-1.amazonaws.com] has left #shogun []		18:23
@wiking	HeikoS, one more thing	18:25
@wiking	we should start using variables for data path even in the xamples	18:26
@wiking	as the ../../	18:26
@wiking	is yeah mikulas :)	18:26
@wiking	and another story that is great http://jupyter-dashboards-layout.readthedocs.io/en/latest/index.html	18:30
@HeikoS	wiking: both good	18:35
@HeikoS	yeah the data	18:35
-!- geektoni [~geektoni@93-34-128-38.ip49.fastwebnet.it] has quit [Remote host closed the connection]		18:54
@HeikoS	micmn: you around?	19:01
micmn	HeikoS: here	19:05
@HeikoS	In file included from /home/heiko/git/shogun/build_bsd/tests/unit/trained_model_serialization_unittest.cc:171:0:	19:06
@HeikoS	/home/heiko/git/shogun/src/gpl/shogun/classifier/svm/SVMOcas.h:20:41: fatal error: shogun/lib/external/libocas.h: No such file or directory	19:06
@HeikoS	compilation terminated.	19:06
@HeikoS	micmn: I am getting that atm did you ever see it before?	19:06
micmn	mmm no	19:08
micmn	so that is guarded with USE_GPL_SHOGUN	19:08
sukey1	[https://github.com/shogun-toolbox/shogun] New commit https://github.com/shogun-toolbox/shogun/commit/c084a7631ae7590b7138176ab4e250238dec1133 by karlnapf	19:09
sukey1	[https://github.com/shogun-toolbox/shogun] New branch feature/BSD_refactor created	19:09
micmn	maybe it's not handling correctly the gpl folder	19:10
micmn	are you compiling with USE_GPL_SHOGUN right?	19:10
sukey1	[https://github.com/shogun-toolbox/shogun] New commit https://github.com/shogun-toolbox/shogun/commit/be24a46e71292479a92770581bf1d42a36dc8ebd by karlnapf	19:10
sukey1	[https://github.com/shogun-toolbox/shogun] New branch feature/BSD_refactor created	19:11
sukey1	[https://github.com/shogun-toolbox/shogun] New commit https://github.com/shogun-toolbox/shogun/commit/c084a7631ae7590b7138176ab4e250238dec1133 by karlnapf	19:11
@HeikoS	micmn: no I disabled it	19:13
@HeikoS	since I am working on the bSD release	19:13
@HeikoS	micmn: did you take the guards into account?	19:13
@HeikoS	mikeling: I just merged the global calls in meta lanugage	19:13
@HeikoS	mikeling: see meta_api/calls.sg for an example	19:13
micmn	yeah	19:13
@HeikoS	micmn: ok maybe it is something with the cache	19:14
micmn	I'll disable GPL_SHOGUN and see what's going on	19:15
@HeikoS	micmn: wait a sec	19:15
@HeikoS	yeah so here is what happens	19:15
@HeikoS	no actually I dont understand	19:16
@HeikoS	mmmg	19:16
@HeikoS	micmn: I wanted to ask something else	19:16
@HeikoS	what is the state of the linalg wihtout templated matrix?	19:16
micmn	mmm not sure how to continue, let me check the pr	19:18
micmn	so the thing about lazy-evaluation	19:19
micmn	what's your idea on it?	19:20
sukey1	[https://github.com/shogun-toolbox/shogun] Issue https://github.com/shogun-toolbox/shogun/issues/3922 opened by karlnapf	19:20
sukey1	[https://github.com/shogun-toolbox/shogun] Issue https://github.com/shogun-toolbox/shogun/issues/3922 karlnapf added label: "Cleanups"	19:20
sukey1	[https://github.com/shogun-toolbox/shogun] Issue https://github.com/shogun-toolbox/shogun/issues/3922 karlnapf added label: "cmake"	19:20
@HeikoS	micmn: I am not sure how to do it and not sure whether that even solves our problems	19:21
@HeikoS	micmn: any ideas on the type determination in linalg?	19:21
@HeikoS	I mean we can go with a simple thing for now, the API is independent of that	19:22
@HeikoS	and I am more afte rthe API for now	19:22
micmn	you mean to avoid the switch?	19:22
@HeikoS	i.e. that the classes exist and that all linalg methods take the un-templated arguments as well as templated ones	19:22
@HeikoS	micmn: yes	19:22
micmn	so for methods like matrix_prod it's negligible	19:23
@HeikoS	micmn: yep	19:23
@HeikoS	micmn: it is more like if you repeatedly call a cheap method	19:23
@HeikoS	micmn: but I am fine with a switch for now	19:23
@HeikoS	as said, this can be changed independently of the API	19:23
micmn	btw in get_cov() I tried to replace vector.vector[i] with vector[i]	19:24
@HeikoS	yes?	19:24
micmn	and the speed drawback was huge	19:24
micmn	I think because of the call check_on_cpu() or something like that	19:24
@HeikoS	thats just a boolean flag	19:25
@HeikoS	is that with compiler optimizations?	19:25
micmn	mmm I'm not sure I'll double check	19:25
@HeikoS	micmn: one thing that might help is to remove the ";" at the end of the error message (then they are faster)	19:25
@HeikoS	micmn: with optimizations i cant imagine that would be slower	19:25
micmn	anyway, the next thing for matrix class	19:27
micmn	could be a prototype for iterators	19:27
@HeikoS	micmn: good point	19:28
@HeikoS	micmn: I also thought that maybe it is a good idea to take a really simple algorithm	19:28
@HeikoS	like linear ridge regression	19:28
@HeikoS	and write that with an untemplate matrix	19:28
@HeikoS	so that there is no float_**** in the algorithm code	19:29
micmn	I could use a typical use-case for an iterator	19:29
micmn	linear ridge regression has it?	19:29
@HeikoS	micmn: no iterators needed	19:29
@HeikoS	just covariance	19:29
@HeikoS	and cholesky	19:29
@HeikoS	but would be a good start I think before digging into more complex terrain	19:29
micmn	ok	19:30
micmn	so the plan is: address the comments on the PR and add methods needed by linear ridge regression and rewrite it with the new matrix class	19:32
micmn	probably I'll need Vector too	19:33
@HeikoS	micmn: btw	19:33
@HeikoS	http://buildbot.shogun-toolbox.org:8080/#/builders/4	19:33
@HeikoS	micmn: http://buildbot.shogun-toolbox.org:8080/#/builders/4/builds/15/steps/5/logs/stdio	19:34
@HeikoS	it is this patch that broke the build	19:34
@HeikoS	micmn: easy to reproduce locally: delete the src/gpl directory, re-run cmake, and then make	19:35
micmn	thx I'm on it	19:40
-!- HeikoS [~heiko@untrust-out.swc.ucl.ac.uk] has quit [Ping timeout: 255 seconds]		19:42
-!- travis-ci [~travis-ci@ec2-54-81-148-90.compute-1.amazonaws.com] has joined #shogun		19:56
travis-ci	it's Heiko Strathmann's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: https://travis-ci.org/shogun-toolbox/shogun/builds/252085514	19:56
-!- travis-ci [~travis-ci@ec2-54-81-148-90.compute-1.amazonaws.com] has left #shogun []		19:56
@wiking	micmn,	23:09
@wiking	[05:03] <micmn> [19:22:01] and the speed drawback was huge	23:09
@wiking	[05:03] <micmn> [19:22:21] I think because of the call check_on_cpu() or something like that	23:09
@wiking	you can profile iy	23:09
@wiking	*it	23:09
@wiking	and maybe use google's benchmarking lib to get the runtimes	23:10
@wiking	btw lazy evaluation is in a way nothing else but you build a stack	23:10
@wiking	and the interface would look like a builder pattern	23:11
@wiking	each operation = push to the OP stack	23:11
@wiking	micmn, https://github.com/ericniebler/range-v3	23:28
lisitsyn	ranges!	23:28
@wiking	micmn, https://github.com/ericniebler/range-v3/blob/master/doc/index.md#views	23:39
@wiking	lisitsyn, this is a fucking nice lib	23:39
Trixis	"MKL.cpp line 901: Assertion R >= 0 failed!" any idea what the potential cause could be?	23:44
--- Log closed Tue Jul 11 00:00:50 2017

Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!