verne.freenode.net changed the topic of #mlpack to: http://www.mlpack.org/ -- We don't respond instantly... but we will respond. Give it a few minutes. Or hours. -- Channel logs: http://www.mlpack.org/irc/
vivekp has quit [Ping timeout: 248 seconds]
vivekp has joined #mlpack
jjmachan has joined #mlpack
jjmachan has quit [Ping timeout: 252 seconds]
jjmachan has joined #mlpack
deepanshu_ has joined #mlpack
vivekp has quit [Ping timeout: 248 seconds]
vivekp has joined #mlpack
jjmachan_ has joined #mlpack
jjmachan has quit [Ping timeout: 256 seconds]
jjmachan_ has quit [Ping timeout: 252 seconds]
pvskand has joined #mlpack
pvskand has quit [Client Quit]
pvskand has joined #mlpack
pvskand has quit [Client Quit]
pvskand has joined #mlpack
< pvskand>
Hi. So when is the latest release of mlpack scheduled any idea? Like when will the modules of GAN, rbm be released?
deepanshu_ has quit [Quit: Connection closed for inactivity]
sourabhvarshney1 has joined #mlpack
< sourabhvarshney1>
Hello everyone
< sourabhvarshney1>
@zoq @rcurtin I have implemented optimizer Nadamax in PR #1177. Would love to get another task.
pvskand has quit [Ping timeout: 240 seconds]
sourabhvarshney1 has quit [Ping timeout: 260 seconds]
bat_ has joined #mlpack
< bat_>
hey all , i just installed ml-pack on my machine but can someone help me understanding the directory structure of the repo ? i mean which folder contains which kinda files. New to open-source :)
pvskand has joined #mlpack
jjmachan_ has joined #mlpack
bat_ has quit [Ping timeout: 260 seconds]
jas__ has joined #mlpack
faizi has joined #mlpack
faizi has quit [Client Quit]
jas__ is now known as preet
faizi has joined #mlpack
faizi has left #mlpack []
pvskand has quit [Read error: Connection reset by peer]
jjmachan_ has quit [Ping timeout: 248 seconds]
dan_ has joined #mlpack
dan_ is now known as Guest52899
< Guest52899>
hey guys and happy holidays! So, I'm compiling mlpack from source on Fedora 27 and explicitly passing the Python bindings flag to CMAKE. I don't get any errors besides a warning regarding text2man (I think). Make and make install run correctly, but I cannot import the mlpack package. In the build folder, there's no bindings folder anywhere. Any ideas? Thanks!
< rcurtin>
Guest52899: hello there!
< rcurtin>
if you want to run the python bindings directly from the build folder there is a little bit of trickery needed...
< Guest52899>
rcurtin: I'd like to have them available system-wide, but there's no bindings folder anywhere. In the src/mlpack/bindings/python folder you mention
< rcurtin>
then you should be able to start python and run 'import mlpack'
< rcurtin>
hmmm, maybe the python bindings did not build
< rcurtin>
you are using mlpack 2.2.4 but the python bindings aren't available there
< rcurtin>
you'll need to use the git master branch
< preet>
ok so there is no use of cmake after building mlpack.So all the codes are in src folder ?
< Guest52899>
rcurtin: Makes sense. I'll try again and report back
< rcurtin>
preet: yes, once you configure with cmake you don't need to use it again
< rcurtin>
Guest52899: sounds good, let us know if you have any issues
< Guest52899>
rcurtin: Quick question: are the bindings for Python2 or 3? Can I force Python3 in any way?
< rcurtin>
yes, let me remember the name of the CMake variable to set...
< rcurtin>
I think actually it should just be '-DPYTHON=/usr/bin/python3' (or whatever the path you want to use is)
< rcurtin>
I will update the README.md with this information---sorry that it wasn't already available
< rcurtin>
I see we have a section on how to use them once they're compiled, but not a section on how to configure and compile the Python bindings specifically
< Guest52899>
rcurtin: Yeah, I think that woud be helpful. Currently under a whole lot of projects workload, otherwise I would submit a pull request with an edited README.md. Added to my to-do list, but I guess I'll only reach that task in 1.5 months or so :'(
< rcurtin>
hehe, I know the feeling
pvskand has joined #mlpack
pvskand has quit [Client Quit]
pvskand has joined #mlpack
pvskand has quit [Client Quit]
pvskand has joined #mlpack
< rcurtin>
'zax
< rcurtin>
oops, wrong window
< pvskand>
Hey<rcurtin> can you tell when will be the next release including the features of GAN and RBM?
< rcurtin>
pvskand: not sure, we still have a handful of things to do before then
< rcurtin>
right now the GAN and RBM are not merged, I am not sure if that will happen before 3.0.0
< rcurtin>
if you'd like to help, there are some open PRs that maybe you could help polish; but I don't know how easy it would be to jump into someone else's code for that
< rcurtin>
(open PRs about GANs and RBMs that is)
< pvskand>
Okay. I have been following kris's work. I will see if I can help.
< pvskand>
Thabka fk
< pvskand>
*thanks for the update
< rcurtin>
yeah, I think kris became very busy with classes, I haven't heard anything from him on github or in here in a while
< rcurtin>
are you using the same python that you compiled against? i.e. whatever you set -DPYTHON=... to
< rcurtin>
and what did you set PYTHONPATH and LD_LIBRARY_PATH to?
< Guest52899>
rcurtin: Yap, using the same Python. I ran export LD_LIBRARY_PATH="/usr/local/lib/:$LD_LIBRARY_PATH" as per the Github page and set the Python path to the folder where the bindings were compiled, under build/ (as suggested in the bug report)
< Guest52899>
Not sure if it's the correct thing to do
< rcurtin>
did you do 'make install'?
< rcurtin>
if not, LD_LIBRARY_PATH should be set to /path/to/mlpack/build/ (whatever that may be in your case)
< rcurtin>
and PYTHONPATH set to /path/to/mlpack/build/src/mlpack/bindings/python/
< rcurtin>
if you did install, then make sure /usr/local/lib/ (or wherever you installed to) is on LD_LIBRARY_PATH and I *think* you should not need to set PYTHONPATH
< rcurtin>
assuming that /usr/local/lib/python3.x/.../ is already on the PYTHONPATH
< Guest52899>
Oops, should have pasted this somewhere rcurtin. This is the output of Python's sys.path
< rcurtin>
no problem, looks like it installed as an egg into /usr/local/lib64/
< rcurtin>
what's in /usr/local/lib? is there a libmlpack.so there?
< rcurtin>
or is libmlpack.so in /usr/local/lib64?
< Guest52899>
In /usr/local/lib: libmlpack.so. and libmlpack.so.2 libmlpack.so.2.0 and libmlpack.so.2.2
< Guest52899>
No .so iles in /usr/local/lib64
< rcurtin>
I'm very confused why your system has /usr/lib64/, is this an unusual distribution or something that's a 32-bit architecture?
< rcurtin>
in any case I don't think that is part of the problem, but clearly something has installed wrong, and unfortunately python doesn't give very good debugging output when an import fails
< Guest52899>
:\ A standard Fedora 27 installation, I believe. 64bit
< rcurtin>
huh, ok, maybe Fedora is still doing /usr/lib64/ and I didn't realize that
< rcurtin>
I typically use debian-based systems even though I am a fedora packager :)
< rcurtin>
the error you get... is it 'cannot import name test_python_binding' or 'No module named mlpack'?
< rcurtin>
I have a fedora system, I'm going to try the same steps you just did to see if I can reproduce the issue
< rcurtin>
for what it's worth, the python bindings seem to still have a few little installation issues to be worked out :)
< rcurtin>
that is why they haven't been released yet... :)
< Guest52899>
I understand perfectly, happy to help debugging
< Guest52899>
I get"ModuleNotFoundError: No module named 'test_python_binding'"
< rcurtin>
and that is caused by 'import mlpack', right?
< Guest52899>
exactly
pvskand_ has left #mlpack []
pvskand_ has joined #mlpack
< rcurtin>
ok, I am trying to reproduce on a fedora system... give me a few minutes
< rcurtin>
if it is not too much trouble (I don't know your setup) you could also try building against python2 and see if this is successful
< rcurtin>
I am wondering if this is a versioning issue between py2/py3 and how they handle .sos
pvskand_ has left #mlpack []
< Guest52899>
rcurtin: On my system build took 1h or so, not sure if you're gonna get there faster than me
< Guest52899>
But I can get it started
< rcurtin>
yeah, the system I am using is not particularly powerful, so it will take a little while
< rcurtin>
but I'm trying to just reproduce the python3 problem that you're having for now
< zoq>
Guest52899: You could speed up the build by building in parallel like 'make -j4' to use four-cores.
< Guest52899>
Compiling against Python2
< Guest52899>
Compiled. python2 > import mlpack gives me 'no module named mlpack' Not sure which paths I should update, though, because of the differences between Python2 and 3
< rcurtin>
right, I think you would need to set PYTHONPATH differently
< rcurtin>
I'm still waiting on the compile to finish here, I think the system is very underpowered..
miqlas has joined #mlpack
< miqlas>
Hi Guys!
< miqlas>
rcurtin! I found possible problems in the current mlpack.
< miqlas>
The cflags line shouldn't have line breaks.
< miqlas>
The .pc file is completly broken because this
< miqlas>
In mlpack.pc.in it is: "Cflags: @MLPACK_INCLUDE_DIRS_STRING@"
< rcurtin>
you're right, it should be all on one line
< rcurtin>
on my system it does generate on only one line
< Guest52899>
rcurtin: Okay. I'll be around for quite a while, so any additional debugging steps just let me know. Really need to get a nice HMM implementation
< miqlas>
yes, because your include folder is /usr/include
< miqlas>
But for me this codepath is active: if (NOT "${incldir}" STREQUAL "/usr/include")
< rcurtin>
Guest52899: no problem, I am still digging (or really just waiting on the compilation to finish...)
< miqlas>
i suppose this codepath doesn't get too much testing :)
< rcurtin>
miqlas: ohhhh, the line wrap at line 535 is inserting the newline
< rcurtin>
let me double-check and reproduce that
< miqlas>
rcurtin: can i give you my postal address here? I mean because the sticker.
< miqlas>
rcurtin: wich is the line 535 for you?
< rcurtin>
miqlas: sure, however you want to get me the address is fine; email works too if you prefer---ryan@ratml.org
< miqlas>
Bt absolutely no idea where does that "-I/packages/mlpack-2.2.5-1/.self/include/" comes from as we got no folder like that
< miqlas>
and the header folder is duplicated.
< miqlas>
but they will generate only warnings, so it should stay so
< miqlas>
ouch, cmake tries to rebuild everything because this small change....
< miqlas>
rcurtin: please, review this policy, as this small patch should NOT require complete rebuild as nothing else changed
< miqlas>
ok, the main cmake file changed, and it means, my whole point irrelevant, because it is hard to decide, wich part did changed and if a complete rebuild required or not, and i suppose a complete rebuild is always a bright and safe site of the life, but hey, it takes hours for me :D
< rcurtin>
miqlas: I agree, ideally CMake should not need to rebuild everything for that, but because of the complexity of the mlpack setup I think it would be very difficult to isolate exactly why everything is being rebuilt
< rcurtin>
I can remove duplicates, hang on, let me push a further patch
< rcurtin>
but I also don't know where /packages/mlpack-2.2.5-1/.self/include/ would come from---maybe that is a default CMake include directory?
< miqlas>
possible.
< miqlas>
but it doesn't makes any harm, the compiler will emmit a warning as this folder not existing and will go further. nothing serious. I've seen worse.
< rcurtin>
I pushed another commit that will remove duplicates from the include directory list
< miqlas>
thanks.
Guest52899 has quit [Ping timeout: 260 seconds]
< miqlas>
rcurtin: if you got some minutes, can you check please the list of the available sci-libs for haiku, and comment if something important missing? Thanks! I give you the link to the list right now.
< rcurtin>
Guest52899: ok, I can reproduce the problem... I will get to work resolving it
< rcurtin>
miqlas: you might consider eigen and shogun, and if you are also interested in python packages there are many there, I'd start with scikit-learn
dan_dan has joined #mlpack
< miqlas>
we got eigen
< miqlas>
but it is maybe not in sci-libs
< dan_dan>
rcurtin: I am the Guest2383 from before, my connection dropped.
< miqlas>
the python stuff have also own folder, we got numpy and scipy and so on, but no scikit
< miqlas>
yet
< rcurtin>
dan_dan: sounds good, I think I have a solution I just need to understand *why*
< miqlas>
but man, how much i hate port python stuff...
< rcurtin>
if you find the mlpack/__init__.py file, it's full of lines like 'from test_python_binding import test_python_binding'
< rcurtin>
I've changed each one of those to 'from .test_python_binding import test_python_binding' (i.e. added a '.' before the package name) and now it imports okay
< rcurtin>
next I need to understand why that works so that I know the solution will be robust
< dan_dan>
rcurtin: I love computers, but sometimes I think there's someone, somewhere, enjoying being devilish like that
< rcurtin>
dan_dan: agreed, they must take pleasure in our suffering
< rcurtin>
miqlas: agreed, python can be such a pain to deal with sometimes, especially when you have to dig into its guts
< dan_dan>
rcurtin: Did that as well and yey, it works!
< preet>
rcurtin: btw what kinda tests does test folder contains ? I mean tests for what ?
< rcurtin>
dan_dan: great, let me work up a patch and a reason for what's wrong
< rcurtin>
preet: it's just tests for the various machine learning algorithms, to make sure they work (or at least to make sure they aren't totally wrong)
< rcurtin>
some of the tests are randomized tests of the algorithms
< dan_dan>
rcurtin: Let me know, I'm curious as well. I'll stay around
< rcurtin>
I have to get dinner shortly but I'll look into it after that