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/
Artea_ has joined #mlpack
ImQ009 has joined #mlpack
< jonpsy[m]> Hi ! So I've made a PR in which the tests pass locally but fail on AppVeyor.
< jeffin143[m]> jonpsy: that's a randome failure
< jeffin143[m]> Don't worry about that
< jeffin143[m]> Since the test has some randomness they sometimes fail
< jonpsy[m]> I see, will this failure inhibit testing other testcases (which come after this)?
< jonpsy[m]> @jeffin143 phew, that's a relief! Appreciate the help!
< shrit[m]> rcurtin: would you recall me how to use ARMA_DONT_USE_WRAPPER?
< shrit[m]> I have defined an option and it is being passed into the linker, but still I am getting the error related to undefined reference to `wrapper2_` functions
< shrit[m]> when cross compiling, I never define anything related to armadillo, even though, I never get a similar error message
< rcurtin[m]1> if you define `ARMA_DONT_USE_WRAPPER` before Armadillo has been included from anywhere, then you'll need to link with, e.g., `-lopenblas` or `-llapack -lblas`, instead of `-larmadillo`; if you're getting strange results, make sure that that macro is defined before including armadillo or mlpack
< shrit[m]> Yeah I have already provided openblas and compiled gfortran. Ineeded as you said I was just defining the option after armadillo include
< shrit[m]> Let us see if this passes
< shrit[m]> still not working
< rcurtin[m]1> well without any error output I can't help further. check that all dependencies of what you're building have ARMA_DONT_USE_WRAPPER set
< shrit[m]> To complete the matrix of yesterday of mlpack_test:
< shrit[m]> Locally-compiled statically linked x86-64: passing
< shrit[m]> The issue of armadillo was related to the use of `ARAM_INCLUDE_DIRS`, and I was `ARMA_INCLUDE_DIR`the path for armadillo since it is only one directory
< shrit[m]> * The issue of armadillo was related to the use of `ARMA_INCLUDE_DIRS`, and I was `ARMA_INCLUDE_DIR`the path for armadillo since it is only one directory
< shrit[m]> Do you think we can keep only one, for instance `ARMA_INCLUDE_DIR` since it is more logical, and I think we can not have several directories for armadillo
< shrit[m]> Also, this mean that the issue of `mlpack_fastmks` is related to the cross compilation process
< shrit[m]> Try another toolchain? and see if this passes?
< rcurtin[m]1> `ARMA_INCLUDE_DIR` is what you would specify to CMake to use for `FindArmadillo.cmake`, but we still need `ARMA_INCLUDE_DIRS` since the CMake configuration script may need to find other include directories; for instance, if the Armadillo installation is configured to use HDF5, we must also add the HDF5 include directory (as well as the Armadillo include directory) to `ARMA_INCLUDE_DIRS`
< rcurtin[m]1> basically, `ARMA_INCLUDE_DIRS` is for internal CMake usage, and probably shouldn't be specified on the command line; if something in there was confusing, maybe do you want to open a PR that improves the documentation?
nidhi has joined #mlpack
< shrit[m]> Good point, because, a lot of users might specify one of them especially when cross compiling since we can not use the already installed armadillo.
nidhi has quit [Client Quit]
< rcurtin[m]1> I had kind of hoped that `ARMA_INCLUDE_DIRS` would be "hidden" and nobody would find it---actually I am surprised you know about it, unless you went digging through `FindArmadillo.cmake`
< rcurtin[m]1> I guess, to clarify, I mean that I hoped that mlpack users would not find it... totally okay if developers find it 😃
< rcurtin[m]1> also, dealgood is back online and working 👍️
< zoq> Nice!
< rcurtin[m]1> I think maybe we could bring back the nightly docker build
< zoq> Yes, looks like the build farm recovered.
< shrit[m]> Great. considering the build farm, do we need to keep the coverage script in CMakeLists: https://github.com/mlpack/mlpack/blob/57da5c44fbc898454aca9d5c87e6edf35bf53c9e/CMakeLists.txt#L200
< rcurtin[m]1> The coverage script hasn't had a maintainer and hasn't worked in a long time... I wouldn't be opposed to removing it
< rcurtin[m]1> (I also don't think test coverage is a particularly good metric)
< shrit[m]> I will open a pull request to remove it 👍️
< shrit[m]> I will try another toolchain in order to cross compile for arm64, if this passes then the issue is related to the mainstream gcc cross-compilers
< rcurtin[m]1> yeah, that seems like a good thing to try
Artea_ is now known as Artea
< M94KAAA9W2> Heyy, I've sent a mail regarding GSoC on mailing list, hope I can get a discussion going on regarding my idea and get some feedback...Thankyou in advance.
< jonpsy[m]> Hey, I read it. Its looks cool!
< jonpsy[m]> Actually, I was planning on sending my proposal as mail as well, but you beat me to it xD
< M94KAAA9W2> Ohh...
< M94KAAA9W2> Which idea are you planning to wokr on?
< jonpsy[m]> Multiobjective optimization algorithms
< jonpsy[m]> I like how you've decided to propose your own project. 👍
< M94KAAA9W2> Yes, actually I want a dataframe like class in C++ as well, which could make dataset pre-processing very very easier and smooth...
< jonpsy[m]> Yep and i concur! If you're successful mlpack would be one of the first libraries to do that
< jonpsy[m]> <jonpsy[m] "I like how you've decided to pro"> My idea is somewhat similar, I'm proposing (kinda) my own idea with a mix of theirs.
< M94KAAA9W2> Yup, I will send a more detailed mail in a day or two which will talk more about the implementation...
< jonpsy[m]> Look forward to it 😁
< jonpsy[m]> * Looking forward for it 😁
ImQ009 has quit [Read error: Connection reset by peer]
< rcurtin[m]1> Hey Sergio Morales E. , I think it should be possible but there might be some minor work to it. If you pickle an mlpack model in Python, this uses the same serialization code as the command-line bindings. let me check something really quick...\
< rcurtin[m]1> okay... I think that this is true, but you'll need to test it to be sure. I think that you can get the exact same binary representation of the model in Python as a string if you do, e.g., `model.__getstate__()`
< rcurtin[m]1> so, for instance, if you did something like `result = mlpack.random_forest(...)`, and then `model = result['output_model']`, that should give you back a big long byte array
< rcurtin[m]1> ok! it seems to work:
< rcurtin[m]1> then from the command-line:
< rcurtin[m]1> and you should see that the results are the same :)
< rcurtin[m]1> I'm glad you asked, I had never been sure if you could do that---now I am sure that you can :) please send a ping if it doesn't work as expected 👍️
bigbosskor has joined #mlpack