--- Log opened Fri Apr 26 00:00:49 2019 | ||
-!- wiking [~wiking@c-185-45-237-122.customer.ggaweb.ch] has quit [Remote host closed the connection] | 02:58 | |
-!- wiking [~wiking@c-185-45-237-122.customer.ggaweb.ch] has joined #shogun | 03:04 | |
-!- wiking [~wiking@c-185-45-237-122.customer.ggaweb.ch] has quit [Ping timeout: 246 seconds] | 03:08 | |
-!- wiking [~wiking@c-185-45-237-122.customer.ggaweb.ch] has joined #shogun | 03:29 | |
-!- wiking [~wiking@c-185-45-237-122.customer.ggaweb.ch] has quit [Ping timeout: 258 seconds] | 04:33 | |
-!- wiking [~wiking@c-185-45-237-122.customer.ggaweb.ch] has joined #shogun | 05:31 | |
-!- wiking [~wiking@c-185-45-237-122.customer.ggaweb.ch] has quit [Remote host closed the connection] | 08:32 | |
-!- wiking [~wiking@c-185-45-237-122.customer.ggaweb.ch] has joined #shogun | 08:44 | |
-!- gf712 [90520850@gateway/web/freenode/ip.144.82.8.80] has joined #shogun | 09:38 | |
-!- HeikoS [~heiko@52.pool85-48-187.static.orange.es] has joined #shogun | 10:49 | |
-!- mode/#shogun [+o HeikoS] by ChanServ | 10:49 | |
-!- gf712 [90520850@gateway/web/freenode/ip.144.82.8.80] has quit [Quit: Page closed] | 11:00 | |
-!- gf712 [90520850@gateway/web/freenode/ip.144.82.8.80] has joined #shogun | 11:05 | |
gf712 | HeikoS: have you seen Scott's last email? | 11:12 |
---|---|---|
wiking | gf712: here? | 11:27 |
@HeikoS | yo | 11:28 |
@HeikoS | gf712: hi | 11:28 |
gf712 | wiking: here | 11:28 |
@HeikoS | es just saw it | 11:28 |
@HeikoS | the dask thing is interesting | 11:28 |
gf712 | yup | 11:28 |
@HeikoS | regarding wiking's plans for arrow | 11:28 |
gf712 | the idl is a no go | 11:28 |
@HeikoS | and also in connection with the actor stuff | 11:28 |
gf712 | it's not even open source | 11:28 |
@HeikoS | yeah i know | 11:28 |
@HeikoS | thats fine | 11:28 |
@HeikoS | it's really nice that he sent all this | 11:28 |
wiking | gf712: ok so q: function composition runtime? :) | 11:28 |
@HeikoS | not sure how to integrate this stuff into the proposal though | 11:29 |
wiking | btw do we have a definition of large? | 11:29 |
@HeikoS | wiking: regarding scotts comments? | 11:29 |
wiking | yes | 11:29 |
@HeikoS | petabytes | 11:29 |
wiking | btw xarray should be easy | 11:29 |
@HeikoS | I think we should definitely take these comments into account and add them to the proposal | 11:29 |
wiking | there's just a typemap need to be done that's all | 11:30 |
@HeikoS | just the question is what exactly | 11:30 |
@HeikoS | wiking: agreed | 11:30 |
gf712 | wiking: array is just python no? | 11:30 |
wiking | and actually as we have eigen wrappers | 11:30 |
wiking | for sgvector/matrix | 11:30 |
wiking | (mappers....) | 11:30 |
wiking | same can be done with xarray | 11:30 |
@HeikoS | but the main problem is that their stuff blows memory | 11:30 |
wiking | gf712: no... it's a python wrapper over xarray | 11:30 |
@HeikoS | wiking: do you have thoughts on dask-style things? | 11:30 |
wiking | HeikoS: lazy op? | 11:30 |
@HeikoS | like say | 11:31 |
wiking | or using dask objs | 11:31 |
@HeikoS | lazy load mostly | 11:31 |
@HeikoS | running a distriuted logistic regression | 11:31 |
@HeikoS | or svm | 11:31 |
@HeikoS | on a datafile that is too large to fit in mem | 11:31 |
gf712 | HeikoS: it would be mostly about writing online algos no? | 11:32 |
@HeikoS | lazy-loading-ifying some standard alogs in shogun would make a lot of sense for them | 11:32 |
wiking | gf712: ok shit | 11:32 |
wiking | actually xarray != quantstack xarray | 11:32 |
wiking | :D | 11:32 |
@HeikoS | gf712: yes or modify some we have | 11:32 |
gf712 | yea | 11:32 |
gf712 | its something else | 11:32 |
wiking | damn | 11:32 |
gf712 | its a pure python thing | 11:32 |
wiking | then needs haxoring | 11:32 |
gf712 | im guessing it uses iterators | 11:32 |
wiking | but you can get the underlying c struct | 11:32 |
wiking | HeikoS: dask has arrow | 11:32 |
wiking | mapping | 11:32 |
@HeikoS | yes | 11:33 |
@HeikoS | thats what I thought | 11:33 |
@HeikoS | that the whole dask/arrow thing might be a nice addition | 11:33 |
wiking | although Dask: Python library for parallel and distributed execution of dynamic task graphs. Dask supports using pyarrow for accessing Parquet files | 11:33 |
@HeikoS | I have to run now. But I will edit the proposal more tonight/tomorrow. | 11:35 |
@HeikoS | If you have any thoughts, please dump them in the proposal | 11:36 |
@HeikoS | or the email | 11:36 |
@HeikoS | and I will put them in | 11:36 |
@HeikoS | question is where to fit this in the project, as the current focus is modelselection | 11:36 |
gf712 | HeikoS: it would be something written alongside making it available to the community | 11:37 |
@HeikoS | gf712: yes | 11:37 |
@HeikoS | ah ! | 11:37 |
@HeikoS | so we could add the under "community" | 11:37 |
@HeikoS | and in the abstract make some space for "community needs" | 11:37 |
gf712 | yea, I think so | 11:37 |
@HeikoS | cool | 11:37 |
gf712 | but need to write it somewhere else | 11:37 |
gf712 | maybe abstract | 11:37 |
gf712 | because it is quite a large chunk of work | 11:38 |
@HeikoS | yes it should be mentioned there | 11:38 |
gf712 | ish | 11:38 |
gf712 | and don't want it to be overseen | 11:38 |
@HeikoS | not the concrete thing, but the fact that this kind of stuff is being done | 11:38 |
gf712 | yes | 11:38 |
@HeikoS | gf712: could you try to add/phrase some stuff? :D | 11:38 |
@HeikoS | I will then factor it in over the weekend | 11:38 |
@HeikoS | Just talked to the other guy from Turing, but I think it is not the best fit | 11:38 |
gf712 | yea, ill add it to abstract and community | 11:38 |
@HeikoS | he introduced me though to some peeps, so maybe something comes up | 11:38 |
gf712 | ok! | 11:39 |
gf712 | deadline is on Monday night right? | 11:39 |
@HeikoS | gf712: I think in the community, we can mention the "top-10" list (maybe different word), and that we want to make sure we develop things useful for the community. We can even list the points we want to work on high level | 11:39 |
@HeikoS | yes monday 23:59 | 11:39 |
@HeikoS | ok cool, thx! Looking better now, I think we have a decent chance with this | 11:39 |
wiking | gf712: ok so i'm at ATI around noon on monday btw | 11:40 |
@HeikoS | wiking: enjoy london, whether is good I hear :) | 11:40 |
@HeikoS | see you later! | 11:40 |
gf712 | HeikoS: hope so! the collab would be good | 11:40 |
gf712 | see you! | 11:40 |
gf712 | wiking: ok! | 11:40 |
gf712 | wiking: function composition runtime? | 11:40 |
wiking | yes | 11:40 |
wiking | so say i have fun1...n | 11:41 |
wiking | that does operation on the input | 11:41 |
wiking | and what i wanna do is basically | 11:41 |
wiking | funN(funN-1(...f1(input)))) | 11:42 |
wiking | the problem is that how and what actually funN is | 11:42 |
wiking | i know only runtime | 11:42 |
gf712 | ah right | 11:42 |
gf712 | so you want to check at runtime return value | 11:42 |
-!- HeikoS1 [~heiko@140.pool85-48-187.static.orange.es] has joined #shogun | 11:42 | |
gf712 | and const ness | 11:42 |
gf712 | ? | 11:42 |
wiking | no | 11:43 |
wiking | i just wanna do the composition of the function into one function | 11:43 |
wiking | but runtime.... compile time easy | 11:44 |
gf712 | but you can store in a dynamic container ? | 11:44 |
gf712 | and then go from there? | 11:44 |
-!- HeikoS [~heiko@52.pool85-48-187.static.orange.es] has quit [Ping timeout: 250 seconds] | 11:44 | |
gf712 | recursive calls? | 11:44 |
wiking | yeah i mean in worst case i could do call over calls | 11:44 |
gf712 | you can't optimise more | 11:45 |
wiking | just wanna avoid | 11:45 |
wiking | vtables | 11:45 |
gf712 | unless you use some runtime optmisation | 11:45 |
wiking | because i could create teh whole object and have the whole hierarchy and have a virtual method for things | 11:45 |
wiking | but | 11:45 |
wiking | that's not really necessary here | 11:45 |
wiking | but yeah | 11:46 |
wiking | i'm looking at this and cannot see a runtime version that is neat | 11:46 |
wiking | only thing is that instead of a virtual func | 11:46 |
wiking | i define operator() on every obj | 11:46 |
wiking | and that's it | 11:46 |
wiking | but for example | 11:46 |
wiking | so say there's I as input | 11:47 |
wiking | and O as out | 11:47 |
wiking | i would like to be able to pass O as rvalue | 11:47 |
wiking | for each operator | 11:47 |
wiking | as there's really no need for copy in any way | 11:47 |
wiking | see what i mean? | 11:47 |
wiking | class A { O&& operator(I&&) } | 11:48 |
gf712 | yea, but you can forward it to the next function no? | 11:48 |
wiking | and i guess then i guess i could do | 11:48 |
wiking | a(a(i)) | 11:48 |
wiking | etc | 11:48 |
wiking | or there's a better way? | 11:48 |
gf712 | well that or move? | 11:50 |
gf712 | if you don't want return by value/` | 11:50 |
gf712 | ? | 11:50 |
wiking | ah no it's not a simple value | 11:50 |
wiking | :) | 11:50 |
gf712 | ah ok | 11:51 |
gf712 | yea I guess you could do class A { O&& operator(I&&) } | 11:51 |
gf712 | wiking: but when you create a temp I would imagine the compiler optimises it to be created in the right place | 11:52 |
-!- HeikoS [~heiko@187.pool85-48-187.static.orange.es] has joined #shogun | 11:52 | |
-!- mode/#shogun [+o HeikoS] by ChanServ | 11:52 | |
gf712 | i.e. f1(f2(i)) | 11:52 |
gf712 | f2(i) creates a temp right? | 11:52 |
gf712 | which the compiler should optimised to be created in place to be consumed by f1 ? | 11:53 |
gf712 | but then again this is all runtime | 11:53 |
gf712 | so not sure how it works | 11:53 |
-!- HeikoS1 [~heiko@140.pool85-48-187.static.orange.es] has quit [Ping timeout: 258 seconds] | 11:54 | |
-!- HeikoS [~heiko@187.pool85-48-187.static.orange.es] has quit [Ping timeout: 245 seconds] | 11:56 | |
-!- wiking [~wiking@c-185-45-237-122.customer.ggaweb.ch] has quit [Remote host closed the connection] | 12:59 | |
-!- wiking [~wiking@c-185-45-237-122.customer.ggaweb.ch] has joined #shogun | 13:00 | |
-!- wiking [~wiking@c-185-45-237-122.customer.ggaweb.ch] has quit [Ping timeout: 255 seconds] | 13:06 | |
-!- wiking [~wiking@ict-networks-195-176-113-062.fwd-v4.ethz.ch] has joined #shogun | 13:16 | |
-!- wiking [~wiking@ict-networks-195-176-113-062.fwd-v4.ethz.ch] has quit [Ping timeout: 250 seconds] | 13:21 | |
-!- wiking [~wiking@2001:67c:10ec:5784:8000::a02] has joined #shogun | 13:49 | |
wiking | hohoho | 15:02 |
wiking | gf712: | 15:02 |
wiking | fwded an email | 15:07 |
gf712 | wiking: just saw thanks | 15:35 |
wiking | https://devblogs.microsoft.com/cppblog/clang-llvm-support-in-visual-studio/ | 15:36 |
wiking | crazy stuff | 15:36 |
wiking | :))) | 15:36 |
gf712 | ah very nice | 15:37 |
gf712 | no more unavailable llvm variables? :D | 15:37 |
wiking | hehehe yeah | 15:37 |
gf712 | but there's no visual studio for Mac right? | 15:38 |
wiking | i think yes | 15:42 |
wiking | https://visualstudio.microsoft.com/vs/mac/ | 15:42 |
wiking | voila | 15:42 |
gf712 | but that doesn't support c++ :/ | 15:45 |
-!- wiking [~wiking@2001:67c:10ec:5784:8000::a02] has quit [Remote host closed the connection] | 17:01 | |
-!- wiking [~wiking@c-185-45-237-122.customer.ggaweb.ch] has joined #shogun | 17:16 | |
-!- wiking [~wiking@c-185-45-237-122.customer.ggaweb.ch] has quit [Ping timeout: 250 seconds] | 17:20 | |
-!- gf712 [90520850@gateway/web/freenode/ip.144.82.8.80] has quit [Quit: Page closed] | 18:11 | |
-!- wiking [~wiking@c-185-45-237-122.customer.ggaweb.ch] has joined #shogun | 19:50 | |
-!- wiking [~wiking@c-185-45-237-122.customer.ggaweb.ch] has quit [Remote host closed the connection] | 21:57 | |
-!- wiking [~wiking@c-185-45-237-122.customer.ggaweb.ch] has joined #shogun | 22:03 | |
--- Log closed Sat Apr 27 00:00:50 2019 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!