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/
< TD>
rcurtin: thanks for the help! I think I might have it.......
< rcurtin>
sure :)
< TD>
SeverityCodeDescriptionProjectFileLineSuppression State ErrorC1083Cannot open include file: 'mlpack/core.hpp': No such file or directoryPatternBotC:\projects\mlpack-2.0.1\src\mlpack\tests\sparse_autoencoder_test.cpp7
< TD>
hold on it's still loading the files
< TD>
SeverityCodeDescriptionProjectFileLineSuppression State ErrorC1083Cannot open include file: 'mlpack/core.hpp': No such file or directoryPatternBotC:\projects\mlpack-2.0.1\src\mlpack\tests\tree_traits_test.cpp11
< TD>
There's 253 Errors
< TD>
SeverityCodeDescriptionProjectFileLineSuppression State ErrorC1083Cannot open source file: 'packages\boost_serialization.1.60.0.0\lib\native\src\xml_oarchive.cpp': No such file or directoryPatternBotC:\projects\mlpack-2.0.1\c1xx1
< rcurtin>
I don't know enough about your environment, is this just to compile mlpack?
< rcurtin>
or are you trying to compile something against mlpack?
< TD>
compile mlpack
< TD>
I was attempting to build a solution in VS2015
< TD>
after compiling mlpack
< keonkim>
hmm did you download the latest mlpack in github?
< keonkim>
I tried with version 2.0.1 and it had some problems
< keonkim>
Fatal, Debug, Info all fails with the same problem, where ever I put it inside cli_impl.hpp or cli.cpp.
< rcurtin>
keonkim: ah, yeah, I remember this now... either we have to force Log::Fatal to initialize early, or you can just make a temporary PrefixedOutStream object with the "FATAL" prefix.
< rcurtin>
probably the latter thing is easier, if you want to make a call to Log::Fatal inside of CLI
< keonkim>
TD: then did you manually name the folder mlpack-2.0.1? the downloaded file should say mlpack-master
< keonkim>
rcurtin: ok, and as I read through the code I discovered some functions are left unimplemented
< rcurtin>
keonkim: yeah, unimplemented functions and unused functions can be removed (if they are clearly not useful)
< rcurtin>
I'm surprised those have not been caught by now
< keonkim>
TD: hmm, I am not sure of the problem here. But searching on Google tells me that the error messages are related to NuGet packages (or windows compiler)?
< keonkim>
can you check if all the specified files while configuring Cmake exist in the directories?
< TD>
Sounds good, I am going to run back through it again
< TD>
Thanks guys!
< TD>
zoq, rcurtin, keonkim
< marcosirc>
time to sleep!
marcosirc has quit [Quit: WeeChat 1.4]
Mathnerd314 has quit [Ping timeout: 276 seconds]
< mentekid>
rcurtin: Do you have any thoughts on how we could measure performance of multiprobe LSH? I'm going to start that today hopefully
< mentekid>
My issue is - if we run multiprobe with the same parameters as single-probe but different numProbes, it will obviously be slower since it does more work. At the same time it will produce better results
< mentekid>
my though was to find parameter sets (K, L, T) that have roughly the same recall and measure their time
< mentekid>
but this might take a lot of tuning to find, so if you have any input about speeding this up I'd be glad to hear it :)
< rcurtin>
mentekid: we are at least guaranteed that multiprobe LSH should give the same recall as regular LSH
< rcurtin>
so we can check that recall is greater than or equal to regular LSH
< rcurtin>
and it would probably be reasonable to expect that recall is increased by some amount, but setting a good bound on how much recall will increase might be hard
< rcurtin>
like maybe we run 10 trials and say, in at least one of those trials, recall must be increased by like 10% or something
< rcurtin>
picking the actual values will probably take some validation
< rcurtin>
measuring the actual runtime is not a great idea, because someone could be running the tests on a machine that is overloaded and situations could occur where multiprobe LSH actually runs faster
< rcurtin>
like if the system is swapping during the single-probe LSH test but not when the multiprobe LSH test happens
travis-ci has joined #mlpack
< travis-ci>
mlpack/mlpack#936 (master - 14aca37 : Marcus Edel): The build is still failing.
< nilay>
can anyone tell me what does appveyor check and what does travis check? thanks.
< rcurtin>
marcosirc: lozhnikov: keonkim: I'm going to add your names to the list of contributors, is that okay? all of you have committed code but I don't see your names in COPYRIGHT.txt :)
< rcurtin>
nilay: appveyor checks the windows build; travis checks the linux build
< rcurtin>
nilay: I have no idea why appveyor failed, I am not going to worry about it
< rcurtin>
when Travis is successful (or specifically when I see that it compiled mlpack okay and the ind2sub test did not fail), I'll merge your PR
< nilay>
rcurtin: ok
< marcosirc>
rcurtin: Great, Thanks!
< rcurtin>
but you should also add your name and email to the list of contributors too :)
< rcurtin>
in src/mlpack/core.hpp and COPYRIGHT.txt
< rcurtin>
nilay: the test looks good, thanks for updating it :)
< lozhnikov>
rcurtin: ok
< nilay>
rcurtin: you are welcome
< keonkim>
rcurtin: I am honored
srivatsa96 has joined #mlpack
< srivatsa96>
Hello, Respected Members. I am a research enthusiast and have recently started working machine learning, I have basic knowledge of algorithms and have been trying to implement them, i want to contribute to mlpack, but being newbie in open source i have little experience in same. can someone please guide me how to begin with?
< rcurtin>
nilay: I'll just add your name to the list of contributors while I am at it, unless you are about to open a PR for it :)
< nilay>
rcurtin: ok yeah you can do it. that would be great :)
mentekid has quit [Remote host closed the connection]
mentekid has joined #mlpack
< mentekid>
rcurtin: I meant more in terms of benchmarking/optimization, not testing. I've already written a test that does what you say (expects more recall as T increases)
< rcurtin>
oh ok lete reread what you wrote...
< rcurtin>
*let me
< rcurtin>
I think the approach of trying different parameters is the right one, I can't think of anything better
< rcurtin>
I think you could write a pretty decent script to run with a grid of parameters
< rcurtin>
once that is done you can scatterplot recall vs runtime for both regular and multiprobe LSH
< rcurtin>
I dunno, do you think that is a reasonable approach? I'm not so sure how we could fit that into the benchmarking system nicely though
< rcurtin>
I only know how to do that by hand
travis-ci has joined #mlpack
< travis-ci>
mlpack/mlpack#941 (master - b29bad9 : Ryan Curtin): The build passed.
mentekid has quit [Remote host closed the connection]
TD has joined #mlpack
< TD>
keonkim: I am almost there
< TD>
I followed the instructions on your blog but I ran into the following errors while buidling mlpack
< TD>
SeverityCodeDescriptionProjectFileLineSuppression State ErrorC1128number of sections exceeded object file format limit: compile with /bigobjmlpack_allkrannC:\projects\mlpack-2.0.1\src\mlpack\methods\rann\allkrann_main.cpp1
< TD>
SeverityCodeDescriptionProjectFileLineSuppression State ErrorC1128number of sections exceeded object file format limit: compile with /bigobjmlpack_fastmksC:\projects\mlpack-2.0.1\src\mlpack\methods\fastmks\fastmks_main.cpp1
< TD>
SeverityCodeDescriptionProjectFileLineSuppression State ErrorC1128number of sections exceeded object file format limit: compile with /bigobjmlpack_hoeffding_treeC:\projects\mlpack-2.0.1\src\mlpack\methods\hoeffding_trees\hoeffding_tree_main.cpp1
< TD>
SeverityCodeDescriptionProjectFileLineSuppression State ErrorC1128number of sections exceeded object file format limit: compile with /bigobjmlpack_kfnC:\projects\mlpack-2.0.1\src\mlpack\methods\neighbor_search\kfn_main.cpp1
< TD>
SeverityCodeDescriptionProjectFileLineSuppression State ErrorC1128number of sections exceeded object file format limit: compile with /bigobjmlpack_knnC:\projects\mlpack-2.0.1\src\mlpack\methods\neighbor_search\knn_main.cpp1
< rcurtin>
keonkim: do you think we need to modify the CMake configuration to enable /bigobj in general? we can do that pretty easily
< TD>
SeverityCodeDescriptionProjectFileLineSuppression State ErrorC1128number of sections exceeded object file format limit: compile with /bigobjmlpack_range_searchC:\projects\mlpack-2.0.1\src\mlpack\methods\range_search\range_search_main.cpp1
< rcurtin>
TD: I think as a quick workaround you can change the project properties for each of those targets to enable the /bigobj compilation option
< rcurtin>
it has been many years since I used visual studio though, so I am not 100% sure of exactly how to do that anymore
< rcurtin>
maybe keon has a better solution, I am not sure
< rcurtin>
another idea would be to disable debugging and profiling (set DEBUG to OFF and PROFILE to OFF in the CMake configuration), but I don't know if you have done that already
< zoq>
yes right, anyway adding /bigobj to CMAKE_CXX_FLAGS should fix the problem
< zoq>
hm, I thought appveyor builds with DEBUG=ON, maybe we should add /bigobj if debug is set
< rcurtin>
I think we did DEBUG=OFF and PROFILE=OFF on appveyor to accelerate the build for the test
< rcurtin>
but I am not sure if the appveyor build is faster with or without debugging symbols
< rcurtin>
one way to find out... :)
< zoq>
:)
< rcurtin>
ok, I guess we will probably find out in a few hours... it seems like AppVeyor is usually pretty backlogged
TD has quit [Ping timeout: 250 seconds]
travis-ci has joined #mlpack
< travis-ci>
mlpack/mlpack#944 (master - 8551a21 : Ryan Curtin): The build was broken.