ChanServ changed the topic of #mlpack to: "mlpack: a fast, flexible machine learning library :: We don't always respond instantly, but we will respond; please be patient :: Logs at http://www.mlpack.org/irc/
< rcurtin> I'd said I was going to release a version of mlpack before I left last week, but I didn't manage to get to it
< rcurtin> I'll do it soon... but I have to dig out from all the emails that piled up first, so it might be a handful of days :)
< chopper_inbound[> Hello everyone! This is my weekly updates blog. Sorry for the late update. Hope it is better than the previous one :) https://mrityunjay-tripathi.github.io/gsoc-with-mlpack/
< saksham189Gitter> @himanshupathak21061998 I have left a review on your PR. Can you also share the weekly blog post?
< jeffin143[m]> <RyanBirminghamGi "I think tensorflow installs prot"> Ryan Birmingham (Gitter): I did that in a virtual environment
tomsun has joined #mlpack
< tomsun> 'make test' under mac get a lot of failed cases, why? and how to fix?
ImQ009 has joined #mlpack
< kartikdutt18[m]> Hey @tomsun, Could you give an example of some failed test cases?
< tomsun> ok
tomsun has quit []
tomsun_ has joined #mlpack
< anjishnu[m]> > 'make test' under mac get a lot of failed cases, why? and how to fix?
< anjishnu[m]> I faced a similar issue some time back. First I installed mlpack, ensmallen and armadillo using brew and ensured they were linked correctly. Then for the cloned repository, I tried compiling with python bindings off and openMP off. That worked and I didn't have failed test cases afterwards. Maybe try this? @tomsun
< tomsun_> anjishnu: I use brew too, and I enabled both openMP and python bindings on, let me try your settings.
HeikoS has joined #mlpack
tomsun_ has quit [Read error: Connection reset by peer]
< rcurtin> chopper_inbound[: thanks for the link---I enjoyed reading it :)
< chopper_inbound[> Thanks rcurtin 😀
< HimanshuPathakGi> Hello, everyone Sorry for posting late https://medium.com/@hpathak336/gsoc-week-1-5179907edeb3 This is my weekly update :)
< HimanshuPathakGi> > `chopper_inbound` Hello everyone! This is my weekly updates blog. Sorry for the late update. Hope it is better than the previous one :) https://mrityunjay-tripathi.github.io/gsoc-with-mlpack/
< HimanshuPathakGi> nice :)
< rcurtin> HimanshuPathakGi: awesome, thanks! I like the chicken image :)
< rcurtin> sometimes that's how I feel when debugging...
< rcurtin> shrit[m]: I got range_search_low_resource_main.cpp to build, but I had to do three things to make it work:
< rcurtin> 1. had to add build/include/ to include_directories() in CMakeLists.txt (so that mlpack headers were available)
< rcurtin> 2. removed 'const unsigned int' parameter from RSModel::serialize()
< rcurtin> 3. didn't use a shared_ptr<RSModel> for serialization; I used an RSModel directly. with the shared_ptr<>, there was this error:
< rcurtin> /usr/include/cereal/cereal.hpp:954:86: error: static assertion failed: cereal could not find any input serialization functions for the provided type and archive combination
< rcurtin> but, I'm not knowledgeable enough about cereal right now to know how to correctly serialize shared_ptrs
< shrit[m]> @ryan thanks, I will apply these modifications and see what is going to happen. If it is woking so there is no reason not work from my side.
< shrit[m]> considering the header, that means that CMake was looking on the already installed header on my machine instead of the new modified header
< HimanshuPathakGi> Thanks @rcurtin
< rcurtin> shrit[m]: yeah, weird things can happen if including a system mlpack header when that's not up to date with git master :)
< chopper_inbound[> @himanshu_pathak:matrix.org: thanks :)
< shrit[m]> @rcurtin exactly, when I used raw pointer, cereal asked for a smart pointer. When I added a smart pointer I got the same error as you. When removing pointers It as working as charm
< rcurtin> shrit[m]: yeah, I guess we will have to figure out the right way, since so many mlpack classes use pointers in one way or another
< shrit[m]> rcurtin We gained 1.3 MB by removing BOOST Serialization
< shrit[m]> 3.4M mlpack_knn
< shrit[m]> Before was 4.7
< jeffin143[m]> Yeyeye @shrit:matrix.org : a nice update
< shrit[m]> jeffin143 thanks
< nishantkr18[m]> <nishantkr18[m] "Hello everyone. here's the link "> Hey zoq favre49. Here's the link to the updates incase u missed. :)
< rcurtin> shrit[m]: nice!
< rcurtin> shrit[m]: does that include the replacement of program_options with CLI11? (I think program_options was pretty small anyway)
< jeffin143[m]> Himanshu Pathak (Gitter) : so you would be implementing kernel svm in mlpack
< jeffin143[m]> That would be awsm :)
< jeffin143[m]> Good to know I had rhe in my bucket list
< zoq> nishantkr18[m]: Thanks, I missed the post in the channel, yes.
< zoq> nishantkr18[m]: Nice read, I really like the problem faced section.
< zoq> nishantkr18[m]: Also First Man is a great movie.
< nishantkr18[m]> :)
< shrit[m]> Yes, CLI11 is pretty small also
< shrit[m]> However, this exclude serilization of arma::mat
< shrit[m]> rcurtin why there was a need to have access::rw() for armadillo paramters in here ? https://github.com/mlpack/mlpack/blob/7bd142d6f3f64e9712141d36b2a0f93a93982251/src/mlpack/core/arma_extend/SpMat_extra_meat.hpp#L23-L28
< shrit[m]> These are armadillo functions, sorry I confused with access from boost serialization
< rcurtin> shrit[m]: yeah, basically access::rw() is just a const_cast<> because the internal armadillo members are const
< rcurtin> so I have to use, e.g., access::rw(mat.mem_state) = X if I want to change mat.mem_state
< shrit[m]> By the way, cereal does not seems to have make_array, and in boost make_array is just casting for make_nvp
< rcurtin> yeah, there might need to be a little extra scaffolding we need to implement for cereal I guess?
< shrit[m]> I will do some test, and see if we need to implement that for cereal, is should be simple
< rcurtin> yeah, most of the boost serialization utilities were pretty simple too
< rcurtin> sigh, another day, another bug found in the bindings...
< rcurtin> I think the way I am binding them into my company's product exercises them in strange ways
< rcurtin> all of the bugs I'm finding are something weird I chose to do many years ago
< rcurtin> I need a time machine to go back in time and teach myself some things :)
< abernauer[m]> rcurtin: Is it a technical debt sitaution?
< rcurtin> abernauer[m]: nah, more of a "why in the world did I make that choice?" situation :)
< jeffin143[m]> @walragatver:matrix.org: sorry for making the git history dirty by adding last try as commit msg
< jeffin143[m]> Please squash and merge next time , so thst even if I forget to rebase it is handled in merge if possible
< walragatver[m]1> jeffin143: It's fine. I think no one would be looking at the commit history.
< walragatver[m]1> I will also take care of squashing in to single commit before merge
HeikoS has quit [Quit: Leaving.]
< abernauer[m]> rcurtin: Nice to see your avoiding some of the hiccups with the R bindings this go around.
< jeffin143[m]> @walragatver:matrix.org: are you still around
< jeffin143[m]> Taking about unit testing can you give an example of unit test
< jeffin143[m]> I am not sure what should I test , then function looks pretty naive to be tested
ImQ009 has quit [Quit: Leaving]
< walragatver[m]1> You can just check whether manual flush is working as expected or not
< walragatver[m]1> jeffin143: Sorry forgot to tag you
< walragatver[m]1> You can also test what happens when we log using multiple filewriter objects at the same time
< walragatver[m]1> In the same file I mean to say
< jeffin143[m]> Same file will not be created
< jeffin143[m]> Because there would be difference in time
< jeffin143[m]> Stamp
< jeffin143[m]> Like even a sec
< jeffin143[m]> I will test it :) and tell yoj
< walragatver[m]1> Also checking that the automatic flush is working two three times continuously correctly.
< jeffin143[m]> Oh I see :)
< walragatver[m]1> <jeffin143[m] "Because there would be differenc"> Yeah I know
< jeffin143[m]> walragatver: thanks for all the suggestions
< jeffin143[m]> These are good starting point
< walragatver[m]1> I was saying that multiple filewriter objects writing in multiple files
< walragatver[m]1> In one same file
< jeffin143[m]> Oh I see , yes that will be used in ensmallen to write
< jeffin143[m]> Nice suggestion
< walragatver[m]1> > Oh I see , yes that will be used in ensmallen to write
< walragatver[m]1> Hmm not sure about that use case
< jeffin143[m]> I mean ensmallen optimizer custom callback just like tensorflow
< walragatver[m]1> I said that thinking that some users might use multiple directories to inference in a better way.
< jeffin143[m]> Ok
< walragatver[m]1> > I mean ensmallen optimizer custom callback just like tensorflow
< walragatver[m]1> Okay, I will need to dig about it more.
< walragatver[m]1> Do callbacks use multiple directories? Have you made any research on it?
< jeffin143[m]> We have to come up with something for that :)
< jeffin143[m]> Tensorflow call back will create two dir
< jeffin143[m]> Named train and valid
< walragatver[m]1> Yeah we need to come up with it.
< walragatver[m]1> Thanks for the detail
< jeffin143[m]> No problem
< jeffin143[m]> Let's call it a night
< jeffin143[m]> Tmrw we have a meet , will discuss it then
< jeffin143[m]> Good night , thanks for your time
< walragatver[m]1> No problem. Goodnight.