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/
togo has quit [Quit: Leaving]
< jeffin143[m]>
It's time , Google should keep Google winter of code too :)
< jeffin143[m]>
Mlpack will be growing at 10x seed 🤪
< jeffin143[m]>
Speed*
< rcurtin>
okay, very nearly got all the memory leaks... just two left, but I think I'll call it a night for now
< hemal[m]>
Does MLpack support hdf5 files ?
< hemal[m]>
Could a model and weights be loaded from hdf5 files and stored to that format ?
< Saksham[m]>
HDF5 is a supported by armadillo
< Saksham[m]>
(edited) ... supported by armadillo => ... supported file type of armadillo
< Saksham[m]>
So it can be used to load matrices, etc
< hemal[m]>
Thanks
< hemal[m]>
Also, I was wondering,
< hemal[m]>
Could mlpack be used as a backend to keras ?
< SriramSKGitter[m>
@hemal : I assume custom python bindings would have to be added. It does sound like a great idea though :)
< hemal[m]>
Sriram S K (Gitter): Thanks!
< hemal[m]>
Any more opinions/ideas on this?
< SriramSKGitter[m>
I'm afraid not hemal, I'm not too familiar with the bindings system, @rcurtin would probably know the right way to go about it. In addition this might be of some help: https://www.mlpack.org/doc/mlpack-3.2.2/doxygen/bindings.html
noaima has quit [Remote host closed the connection]
AbhiSaphire has quit [Remote host closed the connection]
JatoJoe has joined #mlpack
< JatoJoe>
Hello. Glad to be here. Prospective Gsoc intern
JatoJoe has quit [Remote host closed the connection]
< zoq>
himanshu_pathak[: You can just start with a single layer, don't think there is anything I can add, Ryam already pointed out what has to be changed.
< AbishaiEbenezerG>
hi @zoq! Got a query here
< hemal[m]>
Same as zoq said.
< zoq>
hemal[m]: I think this is pretty straightforward, so I don't think we have to open a PR for each layer.
< hemal[m]>
Ok cool
< zoq>
Don't want to end up with 20 PR :)
< hemal[m]>
Right !
< himanshu_pathak[>
zoq (@freenode_zoq:matrix.org): I have started working on it I will raise a pr soon thanks for helping hemal: I agree with zoq we should do this in a single pr not multiple pr
< hemal[m]>
An initial feedback would be of much help.
< hemal[m]>
Also,
< hemal[m]>
We could discuss on why we need arma::sp_mat as an input in the first place.
< hemal[m]>
jeffin143: yes, armadillo has sp_mat.
< hemal[m]>
My point is data::load() data::save() do not have support for sp_mat. They work on arma::mat, arma::col etc.
< hemal[m]>
So,
< hemal[m]>
1. Do we require sp_mat in data::load(), data::save()
< hemal[m]>
2. If yes, how to go about doing that?
< jeffin143[m]>
Definitely , that would help , what I would suggest is add a template option for one of load function and use that in arma::mat and arma::spmat to load a really big csv file or something , like really big and see relatively speed bump or may be a space benchmark to
< jeffin143[m]>
I am not sure if that would give speed bump, but we can surely check*
< hemal[m]>
Thanks!
< hemal[m]>
I was editing the existing data::load(),
< hemal[m]>
Totally forgot overloading,
< hemal[m]>
Thanks.
< AbishaiEbenezerG>
hi mlpack
< AbishaiEbenezerG>
I have a problem compiling a test file
< AbishaiEbenezerG>
@joeljosephjin was able to compile and execute the exact same code without any problems..
< AbishaiEbenezerG>
i've been trying to get past this for a week and i still don't know what i'm doing wrong..
< SriramSKGitter[m>
Are you sure mlpack is installed? Is `libmlpack.so` present in /usr/local/lib or elsewhere on your system?
< AbishaiEbenezerG>
its in /build/lib
< AbishaiEbenezerG>
mlpack/build/lib
< SriramSKGitter[m>
Did you run the `make install` command?
< AbishaiEbenezerG>
build doesn't show up unless i run that cmd right..?
< AbishaiEbenezerG>
ok i think i messed it up. i'm so sorry
< SriramSKGitter[m>
The `make` command compiles the source code and the compiled library is placed in `build/lib`. To complete the process (installing command line executables etc., add mlpack to system libraries in `/usr/local/lib`) , you should run `make install`. :)
< AbishaiEbenezerG>
I hadn't run make install in this directory. I'm not sure whether to cry or laugh at this - it took me a week!!
< SriramSKGitter[m>
We all started from somewhere :)
< AbishaiEbenezerG>
I'll get back when its done @sriramsk1999 .
< AbishaiEbenezerG>
CMake Error at cmake_install.cmake:41 (file):
< AbishaiEbenezerG>
file INSTALL cannot make directory "/usr/local/include/ensmallen_bits":
< AbishaiEbenezerG>
Permission denied.
< AbishaiEbenezerG>
this was the last line line of the make install
< PrinceGuptaGitte>
did you try `sudo make install`
< AbishaiEbenezerG>
Doing...
< AbishaiEbenezerG>
@prince776 @sriramsk1999 Thank you so much for your help!!!
< AbishaiEbenezerG>
I'm sorry . This was terribly dumb of me...
< AbishaiEbenezerG>
Thank you for your patience!
< PrinceGuptaGitte>
welcome. Also if you see permission denied in linux, sudo is mostly the answer(but also keep in mind, what you're doing can potentially change your system settings)
< PrinceGuptaGitte>
(edited) ... linux, sudo is ... => ... linux, `sudo` is ...
< PrinceGuptaGitte>
I had to reinstall Ubuntu for the same reason :)
< AbishaiEbenezerG>
yes yes. i had done similar things before...
< AbishaiEbenezerG>
but it beats me that this took me a week
< AbishaiEbenezerG>
i'm literally just sitting in shock..
< AbishaiEbenezerG>
Are these stuff documented somewhere?
< AbishaiEbenezerG>
you had to reinstall ubuntu ? Why did that happen?
< AbishaiEbenezerG>
ok i think i know why i didn't run the make install in the first place.
< PrinceGuptaGitte>
I don't remember what I did, but essentially i encountered some problems with apt-get, then copy pasted some commands and my system was in shambles
< AbishaiEbenezerG>
make -j4 literally took the entire day. I even discussed it here..
< AbishaiEbenezerG>
and i was so impatient to run the tests..
< AbishaiEbenezerG>
> Even with just 1 core make-ing mlpack , it shouldn't take this long.
< AbishaiEbenezerG>
remember this? I've been ruined since that day
< AbishaiEbenezerG>
you helped me even then
< PrinceGuptaGitte>
Oh yes I remember you were having a hard time with multiple cores
< AbishaiEbenezerG>
Yes
< PrinceGuptaGitte>
does it work fine now? or are you avoiding mutiple cores
< AbishaiEbenezerG>
Multiple core works fine now since the initial build is done
< jeffin143[m]>
hemal (@hemal:matrix.org): you there ?
< AbishaiEbenezerG>
but i'm not planning on trying that out all over again. One day is too much to give up. But i should try it sometime later ...
< AbishaiEbenezerG>
Ok. time to actually start learning the stuff that matter..
< AbishaiEbenezerG>
Thank you so much for the help!
< PrinceGuptaGitte>
good luck, also if you're going to deal with ANN codebase, learning how FFN<> class works is crucial
< AbishaiEbenezerG>
Yeah, i"ve gone through that code . I have the workflow all in pencil and paper here - now i'll actually try stuff out
< hemal[m]>
jeffin143: had gone for a walk, now i'm back,
< LakshyaOjhaGitte>
Isn't your city in lock down too?
< hemal[m]>
Curfew was till 9 pm. I went down at 9:05 pm to get some fresh air for 5 mins.
< hemal[m]>
Thanks for the concern.
< LakshyaOjhaGitte>
No worries
< LakshyaOjhaGitte>
My city got locked down now till 31st. :(
< PrinceGuptaGitte>
same
< hemal[m]>
just pray that the lockdown works and it is not extended any further.
< LakshyaOjhaGitte>
Lets hope so.
< PrinceGuptaGitte>
How do I run my MLPack application with BLAS?
< PrinceGuptaGitte>
also one more question, when using batch norm after convolution network, I just do it with `model.Add<BatchNorm<>>(channels*width*height)` right?
< PrinceGuptaGitte>
(edited) ... convolution network, I ... => ... convolution layer, I ...
< PrinceGuptaGitte>
> How do I run my MLPack application with BLAS?
< PrinceGuptaGitte>
I tried -lblas but no performace improvement was seen.
< PrinceGuptaGitte>
> How do I run my MLPack application with BLAS?
< PrinceGuptaGitte>
I tried -lblas but no performace improvement was seen.
< PrinceGuptaGitte>
> How do I run my MLPack application with BLAS?
< PrinceGuptaGitte>
I tried -lblas but no improvement in performance is seen.
< Param-29Gitter[m>
try installing OpenBLAS and running it with lopenblas
< Param-29Gitter[m>
-lopenblas*
< Param-29Gitter[m>
Hey @rcurtin ,@zoq how do we test if program is performing better when it uses OpenMP?
< Param-29Gitter[m>
because I tired doing profiling but some algorithms perform better in serial than it did in parallel
< Param-29Gitter[m>
I can upload a repository would you like to have a look on those results?
< Param-29Gitter[m>
(I ran my test on thyroid dataset.)
< jeffin143[m]>
hemal (@hemal:matrix.org): there are chances that sp_mat will not work for load and save
< jeffin143[m]>
Because there are many functions which are only supported for matrix and not for sp_mat and hence if some of these are used them it will throw error
< PrinceGuptaGitte>
> -lopenblas*
< PrinceGuptaGitte>
works, but no performace improvement :(
< Param-29Gitter[m>
Which algorithm are you working on?
favre49 has quit [Quit: leaving]
< PrinceGuptaGitte>
yes
< jeffin143[m]>
Not sure go ahead and experiment
< hemal[m]>
Ok, I would simply overload it for a while.
< hemal[m]>
With minimal changes, and test.
< jeffin143[m]>
Yrs
< jeffin143[m]>
Yes
ImQ009 has quit [Quit: Leaving]
< PrinceGuptaGitte>
Hello @zoq, did you get a chance to look at my gsoc proposal draft?
< rcurtin>
ok, just opened #2326 to fix all the memory leak issues that I could find... took a while to handle all of those