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/
< huyssenz>
and I had git clone and compile the codebase. Could you tell me where is the existing neural network codebase? And what should I do next?
< huyssenz>
Thank you!
< rcurtin>
huyssenz: look in src/mlpack/methods/ann/, you can find it there---it's not fully documented yet, so you will have to dig a bit
< huyssenz>
rcurtin: great thank you! BTW, for that project, do I need to imply all those modules for just pick one to work with?
< huyssenz>
Actually, I only familiar with RBM and GAN
< rcurtin>
my understanding (I'm not the mentor) is that it would be up to you to decide how many to do, or how many is realistic for a summer
< huyssenz>
rcurtin: oh I see. Could you please tell me who is Marcus Edel on irc channel or how to contact with him? :)
< rcurtin>
je
< rcurtin>
*sorry, he's "zoq", but keep in mind, there are very many people asking a lot of questions
< rcurtin>
you should definitely search the mailing list for answers to ensure that your question has not already been answered
< huyssenz>
rcurtin: sure thing, I will go to check the mail list next. Thank you for your help
vinayakvivek has joined #mlpack
the-listener has joined #mlpack
the-listener has quit [Remote host closed the connection]
the-listener has joined #mlpack
keshav has joined #mlpack
govg has quit [Remote host closed the connection]
govg has joined #mlpack
keshav has quit [Ping timeout: 260 seconds]
Shihao has joined #mlpack
< Shihao>
Hello everyone, my name is Shihao. I am a student from Lehigh University.
< Shihao>
I am so excited that I can have a chance to participate in a open source community and learn some cool machine learning techniques.
< rcurtin>
Shihao: hello there, I almost went to lehigh! :)
< Shihao>
Really? Nice to meet you!
< rcurtin>
yeah, it was one of the schools I was considering, but I ended up going to Georgia Tech instead
cifar has joined #mlpack
< rcurtin>
there are lots of open issues on github, maybe you can find some interesting way to participate there
< Shihao>
I learned C++ when I was learning OOP and I learned some machine learning method like KNN, SVM from data mining course. I don't know whether I can contribute something to mlpack.
Sinjan has joined #mlpack
< Shihao>
I am really very interested in machine learning and recently I start reading "pattern recognition and machine learning".
Sinjan has quit [Client Quit]
flyingpot has joined #mlpack
< Shihao>
Can I build mlpack on my own Mac?
< Shihao>
It seems I need to install Armadillo and Boost first.
cifar has quit []
the-listener has quit [Ping timeout: 240 seconds]
mayank has joined #mlpack
the-listener has joined #mlpack
dhawalht has joined #mlpack
Shihao has left #mlpack []
dhawalht has quit [Quit: Page closed]
skk123 has quit [Ping timeout: 260 seconds]
mayank has quit [Ping timeout: 260 seconds]
the-listener has quit [Remote host closed the connection]
na01 has joined #mlpack
na01 has quit [Client Quit]
cifar has joined #mlpack
mayankkumar has joined #mlpack
< mayankkumar>
I am getting an error "undefined reference to `mlpack::util::CLIDeleter::CLIDeleter()"
< mayankkumar>
While compiling a sample program ! Can anybody help ?
flyingpot has quit [Ping timeout: 268 seconds]
mayankkumar has quit [Ping timeout: 260 seconds]
mayank has joined #mlpack
vikas has joined #mlpack
kris has joined #mlpack
har123 has joined #mlpack
< har123>
Hello
har123 has quit [Client Quit]
mikeling|afk is now known as mikeling
skk123 has joined #mlpack
florian_ has joined #mlpack
florian_ has quit [Client Quit]
witness_ has joined #mlpack
flyingpot has joined #mlpack
flyingpot has quit [Ping timeout: 260 seconds]
neo_ has joined #mlpack
neo_ has quit [Quit: Page closed]
n2 has joined #mlpack
n2 has left #mlpack []
n2 has joined #mlpack
n2 has left #mlpack []
mayank has quit [Ping timeout: 260 seconds]
Xin_ has joined #mlpack
Xin_ has quit [Ping timeout: 260 seconds]
flyingpot has joined #mlpack
flyingpot has quit [Ping timeout: 260 seconds]
dhawalht has joined #mlpack
< dhawalht>
join
< dhawalht>
Hi people. I am Dhawal Harkawat
Thyrix has joined #mlpack
shikhar has joined #mlpack
skk123 has quit [Ping timeout: 240 seconds]
dhawalht has quit [Quit: Page closed]
Sinjan has joined #mlpack
gk-1wm-su has joined #mlpack
gk-1wm-su has left #mlpack []
Sinjan has left #mlpack []
Sinjan_ has joined #mlpack
< Sinjan_>
rcurtin: I checked your reply in the logs. Shall I start working on the issues? Or do I need to select an idea before that?
sixpluszero has joined #mlpack
sixpluszero has quit [Quit: Page closed]
dracarys has joined #mlpack
dracarys has quit [Client Quit]
abhijeet has joined #mlpack
< abhijeet>
Hi, I'm interested in doing GSoC with mlpack...
< abhijeet>
I've gone through the project page and I'm interested in "Reinforcement Learning" project
< abhijeet>
I'm proficient in C++ and have experience in general machine learning and reinforcement learning
< abhijeet>
I've also done a project involving reinforcement learning... so basically my project was to design an agent that can play carrom...
< abhijeet>
It would be great if someone can provide me relevant links... and beginner level bugs... so that I can get myself familiar with the code base
flyingpot has joined #mlpack
flyingpot has quit [Ping timeout: 260 seconds]
vinayakvivek has quit [Quit: Connection closed for inactivity]
skk123 has joined #mlpack
shikhar has quit [Quit: Page closed]
skk123 has quit [Ping timeout: 264 seconds]
Sinjan_ has quit [Ping timeout: 260 seconds]
witness_ has quit [Read error: Connection reset by peer]
witness_ has joined #mlpack
< govg>
abhijeet: You can check the Github repository, there will be a lot of issues with specific tags on them (beginner, hard, etc)
bartek has quit [Remote host closed the connection]
bartek has joined #mlpack
bartek has left #mlpack []
ashk43712_ has quit [Ping timeout: 260 seconds]
kris1 has joined #mlpack
kris has quit [Quit: Leaving.]
mayank has joined #mlpack
kesslerfrost has joined #mlpack
skk123 has joined #mlpack
jacob_ has joined #mlpack
jacob_ is now known as Guest50655
Guest50655 has quit [Quit: Page closed]
witness_ has quit [Quit: Connection closed for inactivity]
< rcurtin>
Sinjan_: you're welcome to start working on the issues if you like or look more at the projects---I can't guide you there, the choice is yours\
< rcurtin>
ah wait oops I see that govg already answered :)
topology has joined #mlpack
< topology>
rcurtin: you there?
< rcurtin>
topology: I am, but I am working on other things
< topology>
i won't take much time. You mentioned 'implementing kernel density estimation as a dual-tree algorithm' as a potential project some time back.
< topology>
would you still be interested in mentoring it?
< rcurtin>
sure, I think that could be interesting
< rcurtin>
but we'll have to do something a little more complex than the simplest possible KDE algorithm, because that's really pretty simple :)
< rcurtin>
(although that would be a great start!)
< rcurtin>
Bill March published some papers recently about a KDE technique he was calling ASKIT... I think that there may be some interesting ideas in there
flyingpot has joined #mlpack
< rcurtin>
he was applying it to kernel methods (specifically, approximating a kernel matrix) but the same ideas could be applied to density estimation
< rcurtin>
sort by date, these series of papers were in 2015-2017 I think
< rcurtin>
probably "far-field compression for fast kernel summation methods in high dimensions" might be a place to start
< rcurtin>
but definitely, there will be a lot of reading and understanding and then "translating" to the tree-independent dual tree algorithm context that mlpack implements
< topology>
i understand that. i will go through Bill's papers first.
< rcurtin>
sure---keep in mind he may not be talking directly about kernel density estimation
< rcurtin>
but the problems he is solving can be easily adapted to the KDE problem
pvskand has joined #mlpack
flyingpot has quit [Ping timeout: 240 seconds]
< topology>
i will go through his papers and see if i can understand the idea. Thank you for the link :)
< topology>
i did read about dual-tree algorithms from your thesis. i think i have a fair grasp on the concept.
< topology>
i found the papers on "far-field compression" and ASKIT. i will read them and get back to you with questions. maybe we can discuss it further then?
shikhar has joined #mlpack
smeet has joined #mlpack
smeet has quit [Client Quit]
skinnypete has joined #mlpack
ashk43712_ has joined #mlpack
abhijeet has quit [Ping timeout: 260 seconds]
Shihao has joined #mlpack
Shihao has quit [Client Quit]
jarvis_ has joined #mlpack
mayank has quit [Ping timeout: 260 seconds]
jarvis_ has quit [Client Quit]
Smeet has joined #mlpack
mikeling has quit [Quit: Connection closed for inactivity]
< Shihao>
I am new to mlpack and I have some course experience with C++.
< Shihao>
mlpack is already complied and installed on my machine.
< pvskand>
@rcurtin - I wanted to ask if MLP has been implemented? I couldn't find it in the perceptron class.
< Shihao>
I guess I should go through naive_bayes_classifier.hpp and naive_bayes_classifier_impl.hpp first.
flyingpot has joined #mlpack
kris has joined #mlpack
kris1 has quit [Ping timeout: 240 seconds]
qber has joined #mlpack
qber has left #mlpack []
pvskand_ has joined #mlpack
flyingpot has quit [Ping timeout: 260 seconds]
pvskand has quit [Ping timeout: 260 seconds]
< zoq>
Shihao: Hello, going through the NBC code is definitely a good idea, the code has some really helpful comments too; let us know if you have any questions.
chvsp has joined #mlpack
< zoq>
pvskand: The ann code src/ann implements the FNN class, which is a container for MLP. feedforward_network_test.cpp has some examples.
< zoq>
I know he timed out but in case he reads the logs.
< shikhar>
If I include mlpack/core/arma_extend/arma_extend.hpp, the library compiles, but with pragma message that arma was included before mlpack
< zoq>
Have you tried "#include <mlpack/prereqs.hpp>"?
< chvsp>
Hi. @zoq, I have successfully changed the code for Kathir Subavathi Initialization. The Travis CI check failed due to a failing of some other test. Is there any way to restart the check without pushing in another commit?
< Shihao>
Is it correct that classify function receives a third parameter arma::Row<size_t>& prob and set it when calculating maximum probability ?
< Shihao>
I should take some time learning armo and mat type.
< shikhar>
zoq: Yes. The library fails to compile after that. "prefixedoutstream.hpp" also has std library headers included in it.
Lakshya1605 has joined #mlpack
< zoq>
chvsp: Great, I'll take a look at the PR tomorrow, as long as the test you implemented for the initialization method doesn't fail it's fine. We have some tests that e.g. don't converge in the specified number of iterations, so sometimes they fail. To answer your question, unfortunately, you can't just restart the build without committing something, but if you like I can do it for you. But as I said it doesn't
< zoq>
matter, as long as your test is fine.
< chvsp>
zoq: That would help. Thank you. :)
< Shihao>
when I compile from command line, like g++ nbc_main.cpp -o nbc_main -I/usr/local/include -std=c++11, it says error: use of undeclared identifier 'Log'.
< Shihao>
Is there anything wrong?
< zoq>
chvsp: Okay, restarted.
< zoq>
Shihao: You have to link against mlpack soemthing like: g++ nbc_main.cpp -o nbc_main -I/usr/local/include -std=c++11 -lmlpack
< zoq>
Shihao: But maybe you like to build the mlpack nbc executable?
< Shihao>
Yes, I want to test this program using some data.
< zoq>
Shihao: So you should be able to call 'mlpack_pca -h' if you used 'make install' if not from your build directory via 'bin/mlpack_nbc -h' right?
< zoq>
shikhar: Nice terminal: zsh?
< shikhar>
zoq: yes! :D
< Shihao>
zoq: Yes I can call 'mlpack_pca' directly.
< zoq>
Shihao: Okay, so I guess the next step is to go through the manpage (mlpack_nbc -h) and to figure out how to use the nbc executable.
< Shihao>
zoq: What if I want to recompile it when I changed some code there? Can I just compile 'nbc_main.cpp'?
< zoq>
shikhar: So, I guess, to figure out what happened here, I'll have to see the modified lines; maybe you already told me? I'm about to lose track ...
< shikhar>
zoq: Alright, I'll make a PR as soon as I finish writing tests for the fix.
< zoq>
Shihao: yes, 'make mlpack_nbc' followed by 'make install', I guess 'make install' clears the last build, so if it rebuilds everything skip the last step 'make install' and just call it from the build directory.
< zoq>
shikhar: Okay, sounds good, maybe rcurtin can tell you more, once he is back.
< zoq>
Out for dinner, back later.
< shikhar>
zoq: Thanks!
Lakshya1605 has quit [Quit: Page closed]
skinnypete has quit [Ping timeout: 260 seconds]
< Shihao>
Oh, I see it. Thank you very much!
Shihao has quit [Quit: Page closed]
kesslerfrost has quit [Quit: kesslerfrost]
chvsp has quit [Quit: Page closed]
topology has quit [Quit: Page closed]
< kris1>
zoq: What to do if we want to create 6 dimensional vector in armadillo
< kris1>
i know field could help but i am not sure if its available in 4.2
< kris1>
could we do something like this arma::MAT<arma::MAT<arma::MAT>>> m6d;
< kris1>
yes you can't do that i think we would have to use std::vector
< zoq>
kris1: Two options 1. use arma::mat(d, m x n) and wrap the cols 2. use arma::cube(m, n, d).
shaunak has joined #mlpack
< shaunak>
hi I'm new here
< shaunak>
i wnt to contribute in mlpack
< shaunak>
*want
< shaunak>
please tell me from where to start
< kris1>
zoq: i want a 6 dimensional array woudn't cube give me a 3d array of size m*n*d
< kris1>
also i did not understand the first method
< kris1>
i want a 5d array actually
< zoq>
kris1: Maybe I misunderstood your question; I thought you would need something like std::vector<arma::mat> c; c.push_back(arma::mat(m, n)) push arma::mat(m, n) 6 times?
< kris1>
zoq:I did something similar but i was thinking about the indexing. c[0][1][2] would give you a element so c is actually a 3d array only
< kris1>
am i right
< kris1>
in c one dim would be for vector and other 2 would be for mat
< zoq>
but c[0][1][2] retuns a single element e.g. double?
shaunak has quit [Ping timeout: 260 seconds]
mayank has joined #mlpack
< mayank>
if i make changes to the cpp source codes, do i need to build the whole mlpack again ?
< zoq>
mayank: 'make' should only build modified files.
< kris1>
zoq: yes it does return a single element but we need 5d array and not a 3d dimensional array.
< kris1>
hence 5 indexes should be required right
< kris1>
basically my q table has 4 indexes for state values and 1 index for the action
< kris1>
state = (0,1,3,0) and action = (1)
shikhar has quit [Quit: Page closed]
< kris1>
zoq: i think it would be better if we implement this as a lookup table right
< kris1>
what do you think
< rcurtin>
pvskand: no, I think Udit only did a single-layer perceptron. honestly, it should be reimplemented with the ANN code
< rcurtin>
lots of activity today... I can't keep up!
< zoq>
kris1: So I think what you need is std::vector<arma::cube>, arma::cube is basically what arma::field is but in arma::cube you can't use different sizes.
< zoq>
kris1: Yeah, I think that's probably the best idea.
< rcurtin>
kris1: I guess, I have to ask, for what reason would you need a 6-dimensional vector?
< kris1>
rcurtin: so i implementing a q learning algorithm
< kris1>
tables based version
< kris1>
which needs to lookup the state and action pair
< rcurtin>
ok, right, but I would still strongly consider finding a way to "flatten" it into 2 dimensions
< zoq>
kris1: I guess, you can also store everything into a single matrix and map the index somehow.
< rcurtin>
right, exactly
< rcurtin>
I'd also be concerned, unless you know that most of those indices will generally be small, that the object will use a massive amount of memory
< rcurtin>
er, "dimensions" not "indices"
< kris1>
rcurtin: yes thats why i was thinking of using a dict as in python. where the keys could be a string."0103".
< zoq>
kris1: Maybe something like index = index1 * index2 * index3 * ... works.
< kris1>
zoq:yes that would work i would just have to work that out. it would be something like 1*index + alpha1 * index1 + alpha2*index2+....
< kris1>
i think that could be worked out
< kris1>
thanks
< zoq>
Okay, I have to give someone a ride, but it looks like rcurtin is back :)
arunreddy_ has joined #mlpack
< govg>
kris1: that's how indexing is done normally in a 2d array, right? i*c + j is location of (i,j) in an rxc array, and this generalizes to any arbitrary dimensions, so if you know the total dimension you can make it work even with a 1D array (at really high cost of acces though, I guess)
< rcurtin>
yeah---the armadillo matrix is just a 2-d mapping to a single memory block, just like that
< kris1>
Thanks i am doing that now.
ashk43712_ has quit [Quit: Page closed]
emay_ has joined #mlpack
Shihao has joined #mlpack
emay_ has quit [Client Quit]
< rcurtin>
Shihao: did you get an adequate answer about #593?
< Shihao>
Yes. zoq's suggestion is really helpful.
< Shihao>
I think I can work on that issue now :)
< rcurtin>
ok, great, just make sure you write tests for any changes :)
< Shihao>
Where can I find some test data?
< rcurtin>
either you can randomly generate it, or take a look in src/mlpack/tests/data/
< rcurtin>
ideally we don't want to add a big dataset to the repository, so any unit tests should be small-scale
< Shihao>
Ok, got it.
Shihao has quit [Quit: Page closed]
Shihao has joined #mlpack
arunreddy_ has quit [Ping timeout: 260 seconds]
flyingpot has joined #mlpack
arunreddy_ has joined #mlpack
flyingpot has quit [Ping timeout: 260 seconds]
skk123 has quit [Ping timeout: 240 seconds]
Gitesh has joined #mlpack
Gitesh has quit [Quit: Page closed]
mayank has quit [Ping timeout: 260 seconds]
< zoq>
rcurtin: I think it should also work on android, but just to be sure, can you test it for me?
< zoq>
rcurtin: I think I'll push the svg into the mlpack.org repo and use that as src.
< rcurtin>
sure, let me check
< rcurtin>
it looks good to me, but I notice that the background of the image is not exactly the same color as the background of the page
< rcurtin>
maybe the image should be transparent?
< zoq>
hm, I thought it's transparent ... let's see
< rcurtin>
it looks just fine on firefox
< rcurtin>
so I wonder, maybe android is not respecting the transparency and assuming 0xffffff background?
< rcurtin>
I'm using the default "browser" so I don't know if it's worth worrying about
< rcurtin>
let's try firefox mobile...
< rcurtin>
hmm, also happens with firefox mobile
< rcurtin>
it's pretty subtle though---noticeable if you look, but not horribly distracting (well, I guess that depends on the person)
< rcurtin>
the page layout still looks good though
Shihao has quit [Ping timeout: 260 seconds]
< zoq>
hm, I tested it on mlpack.org and it looks like it's transparent.
< zoq>
strange
< rcurtin>
yeah, this is something I only see on android
< rcurtin>
it looks just fine on the desktop
< rcurtin>
it sounds like it might be a browser bug or something
< zoq>
crazy same thing on iOS, tested on chrome and safari
< rcurtin>
strange!
< rcurtin>
I tried to see if I could load the mobile page on my desktop, but I don't seem to see an easy way to do that
< zoq>
ah good idea, I think you can set the user agent in chrome
< zoq>
yeah, if I set the user agent I can see the same effect, I have an idea let's see if that's the problem
< zoq>
hm, http://kurg.org/test.html same image and it works on my phone, maybe github does something strange ...
< rcurtin>
yeah, I was wondering if there was some bad CSS somewhere on the github mobile stylesheet
< rcurtin>
I guess one could always open an issue on github itself :)
< zoq>
Okay style="background-color:rgba(0,0,0,0);" which sets the opacity worked. I've checked some other repos that also use an svg; I couldn't find a single repo that hasn't the same effect as we, maybe they don't know :)
< rcurtin>
yeah, I think few people look at github on phones :)
flyingpot has joined #mlpack
flyingpot has quit [Ping timeout: 260 seconds]
< kris1>
zoq: i have a function like this int select_action(Environment& env, double epsilon);
< kris1>
but when i am calling it q.select_action(env.action_space, learning_rate);
< kris1>
i get a error like this /usr/include/armadillo_bits/Mat_bones.hpp:60:27: error: initializing argument 1 of ‘arma::Mat<eT>::Mat(const char*) [with eT = double]’ [-fpermissive]
< kris1>
inline Mat(const char* text);
< kris1>
i don't know why? i am using the same parameters in a constructor and it works fine there
vinayakvivek has quit [Quit: Connection closed for inactivity]
< kris1>
so i think i am asking what is the way to pass the env variable
< zoq>
You can e.g. use https://www.diffchecker.com/diff to take a look at the diff, but what I did is to use Environment& env instead of const Environment& env and return an int.
< zoq>
Maybe you can send me the complete error message, if that's not working for you?
< zoq>
Btw. respect for starting with QLearning, eager to see the first results.