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/
ironstark has quit [Ping timeout: 240 seconds]
ironstark has joined #mlpack
sagarbhathwar has joined #mlpack
mikeling has joined #mlpack
sagarbhathwar has quit [Ping timeout: 260 seconds]
ironstark has quit [Ping timeout: 256 seconds]
ironstark has joined #mlpack
pengpeng has quit [Quit: pengpeng]
vinayakvivek has joined #mlpack
etotientz has joined #mlpack
etotientz has quit [Read error: Connection reset by peer]
Akash__ has joined #mlpack
Akash__ has quit [Client Quit]
govg has quit [Ping timeout: 260 seconds]
Proxy_ has quit [Quit: Leaving]
diehumblex has joined #mlpack
Aditya has joined #mlpack
Aditya_ has joined #mlpack
vss has joined #mlpack
Trion has joined #mlpack
< vss>
anyone applying for mlpack through gsoc ?
< Trion>
yeep! :D
< vss>
What are you planning to work on ?
< Trion>
Reinforcement Learning project, more precisely the algorithms
< vss>
cool , i want to work on augmented rnn , more precisely the neural turing machine :D
Aditya_ has quit [Ping timeout: 260 seconds]
Aditya has quit [Ping timeout: 260 seconds]
< Trion>
nice! :-)
< vss>
wonder if ill get in tho :/
< Trion>
tay motivated to learn and give a nice honest proposal :-) And ofcourse interact with community
< Trion>
stay*
< diehumblex>
hi i want to create sentiment analyser using nlp. can mlpack help here?
Trion has quit [Ping timeout: 240 seconds]
vinayakvivek has quit [Quit: Connection closed for inactivity]
< rcurtin>
diehumblex: I am not sure that mlpack implements many components that would be useful for NLP, so I don't know if mlpack would be the best choice
govg has joined #mlpack
< rcurtin>
but I am also not too familiar with NLP or sentiment analysis, so maybe it is possible that there are some parts of mlpack could be useful
< diehumblex>
rcurtin : Thanx
< mikeling>
rcurtin: hey
< mikeling>
thank you for the review :)!
< rcurtin>
sure, it isn't comprehensive
< rcurtin>
many more reviews to do today...
vinayakvivek has joined #mlpack
shikhar has joined #mlpack
thyrix has joined #mlpack
thyrix has quit [Ping timeout: 260 seconds]
lol_ has joined #mlpack
lol_ has left #mlpack []
rock has joined #mlpack
rock has quit [Client Quit]
sagarbhathwar has joined #mlpack
< sagarbhathwar>
Hi, I had a small doubt. When two functions share some common functionality, is it a good idea to write a private method to abstract that functionality from the two methods?
< rcurtin>
sagarbhathwar: that depends a lot; what is the functionality in question?
< diehumblex>
I would like to propose sentence structure analyser as a gsoc project. it would extend mlpack's support towards language processing. would it be of interest to mlpack ?
< sagarbhathwar>
Say in softmax regression, I am writing an overload for Classify which computes probabilities for input dataset. Then again, the other overload of Classify Computes probabilities + uses it to classify the data points. So, a private method ComputeProbabilities sounds like an abstractions for the two methods.
< rcurtin>
sagarbhathwar: why not just call the other overload of Classify()?
< rcurtin>
diehumblex: you are welcome to submit a proposal about that, but be careful that you have fully thought through what the API will be and how it will fit in mlpack
< rcurtin>
we need to make sure that new support to the library is well-integrated with other support, so that it doesn't feel like mlpack is many separate libraries that are all included together :)
< sagarbhathwar>
rcurtin: Sure, I could do that. Just wanted to make sure it is okay to call Classify from within Classify?
< rcurtin>
yes, I'm not sure why that wouldn't be okay
< sagarbhathwar>
Great then :)
< diehumblex>
rcurtin: who could possibly mentor this proposal ? i would like to thoroughly discuss it.
Trion has joined #mlpack
< rcurtin>
diehumblex: that would be the next problem, you would have to find a mentor. I certainly don't have time; I am not sure if anyone else does
< diehumblex>
rcurtin: so all present mentors are already occupied and no new mentor can join. guess should have come earlier.
< rcurtin>
diehumblex: something as different as 'adding NLP support' would have always been a long shot anyway, since it requires a huge amount of time figuring out how its API can match the rest of mlpack
< rcurtin>
it's possible there may be some possible mlpack contributor would would be interested in mentoring this, but like I said I personally don't have time to mentor that
< diehumblex>
Hi if anyone got interested in my proposal and would like to be mentor please let me know.
< Trion>
diehumblex, have you looked into cltk organisation?
< Trion>
They are based on nlp, so they will be able to provide a mentor much easily
< diehumblex>
yes actually they are in python and what i proposed they already have it
< Trion>
hmm.. I see, the parts of speech code shows they have support for just greek and latin at the moment and one of their project idea is to extend CLTK core to new languages.
< Trion>
For mlpack, you can leave a mail in mailing list and mentors can discuss if there is a possibility for nlp in mlpack/ nlpack! :D
mlpack has joined #mlpack
mlpack has left #mlpack []
Nax has joined #mlpack
< Nax>
zoq: Hi, I have researched on nuget package,
< Nax>
I found that, we ll need one more file (mlpack.nuspec) if we want to add automatic integration with appveyor
< Nax>
It would be quite easy to integrate mlpack with nuget using appveyor
Yunjian has joined #mlpack
< zoq>
Nax: Yeah, it's a pretty streamlined process, I think we can still use the jenkins repo to provide the all files including the nuspec file and load all files in the build process. There is one problem I see, if we use appveyor for the package creation, how can we controll that it's not creating a new package for every PR?
< diehumblex>
Trion : my bad i was talking about nltk :)
< Nax>
zoq: I guess, standard method would publishing only specific branch instead of master branch
< diehumblex>
cltk also looks like is in python
< Nax>
besides we can also provide nightly build (pre-release) and stable release would be from specific branch
< zoq>
Nax: Okay, if that works, that would be really nice.
< Nax>
zoq: but still as mentioned in #919, we still need nuget for armadillo, or can we include its files in nuget for mlpack?
< Trion>
diehumblex: yes it is :-) and quite similar
< zoq>
Nax: I would prefer to create an independent nuget package for armadillo and use that as dependency for the mlpack package.
< Nax>
zoq: ok got it, Thanks for the help!
mverma has joined #mlpack
mverma has quit [Client Quit]
vss has quit [Quit: Page closed]
Nax has quit [Ping timeout: 260 seconds]
vss has joined #mlpack
< vss>
hi ! does anyone know what are the necessary fields for the proposal draft (Gsoc) ?
< yaing>
Hey, guys. I'm just getting started with mlpack. Can someone tell me how can I work on mlpack codebase in codeblocks ide?? I like to do c++ development in codeblocks. Thanks
Trion has quit [Client Quit]
Trion has joined #mlpack
Trion has left #mlpack []
Trion has joined #mlpack
< Trion>
yaing: are you on linux or windows?
shikhar has quit [Quit: Page closed]
sagarbhathwar has quit [Ping timeout: 260 seconds]
< yaing>
Thank you but its for windows. i have installed mlpack on my ubuntu desktop
mikeling has quit [Quit: Connection closed for inactivity]
Trion2 has joined #mlpack
< Trion2>
damn my internet.. crashes more than it works :P
< Trion2>
is codeblocks interface very different in windows and ubuntu?
< yaing>
interface not so much but may be paths will be different
< yaing>
and i don't know how to load mlpack codebase in codeblocks :(
Trion3 has joined #mlpack
Trion has quit [Ping timeout: 260 seconds]
< Trion3>
The article seems to cover linux part too
< Trion3>
the path will be something like /usr/local/include
< yaing>
okay, i will try that
< yaing>
can also tell me how to load the mlpack codebase in it? I want to make changes to some files and then push them in my fork on github. Is that possible?
Trion2 has quit [Ping timeout: 260 seconds]
Yunjian has joined #mlpack
< yaing>
does anyone use codeblocks here? Please help
< zoq>
yaing: I'm pretty sure if you search for codeblocks and probably cmake you will find something helpful,; mlpack doesn't do anything special, when it comes to the project configuration.
< yaing>
okay thank you zoq. i'll try to find something
Trion3 has left #mlpack []
Trion has joined #mlpack
Yunjian has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
< pvskand>
Does that mean when you run bin/mlpack_test -t ConvolutionalNetworkTest, you dont pass the test?
< zoq>
pvskand: Yes, since we use random parameters and a fixed number of iterations it's possible that the network does not converge for the current setting, so sometime the test fails.
< pvskand>
I have ran the test 25 times and I find it to pass all the time! Since in cnn the weights are randomly generated, I do agree that the results might sometime fail. But it should not fail frequently right?
< zoq>
pvskand: Right, not frequently, also remember to set math::RandomSeed(std::time(NULL)); if you test it multile times, so that it uses a new seed for each run: https://github.com/mlpack/mlpack/issues/922
< pvskand>
Yeah! I am running the test again via shell script, so since the initial seed it set to NULL, every time I run the script it is set to null!
< chvsp>
Hi zoq: I have completed the implementation of the BatchNorm Layer. I was writing tests for the same. Can I hardcode the input and the expected output values? I have got a python implementation of this layer and I gave both, random inputs and have compared the various parameters. Would this be a valid test?
rzhu has quit [Quit: Page closed]
< zoq>
chvsp: Sounds like a reasonable test as long as the python code is correct; we should also think about some other tests, but in this case it's not that easy to come up with something really useful.
aashay has joined #mlpack
< chvsp>
zoq: I got the implementation here - http://cthorey.github.io./backpropagation/ . I have crossed checked it with the theoretical formulae. I agree that we need to have some other tests but I guess these will suffice for the interim.
< zoq>
chvsp: I agree, that is a good start, if you like open a PR and we can look over the code.
< chvsp>
Writing the tests will take some time? Would you like me to open the PR with the implementation?
< chvsp>
*time.
< zoq>
Take your time to write the test, no need to open the PR without the test.
< chvsp>
Okay :)
< chvsp>
zoq: To check the equality of two matrices, is it sufficient to check if the sum of all elements are equal in both? This was used in almost all of the ann layer tests.
< zoq>
chvsp: Depending on the matrix that is something you could do, probably a better way is to use CheckMatrices from the test_tools file.
< chvsp>
Oh nice function. I will try to use that :)
vinayakvivek has quit [Quit: Connection closed for inactivity]
notarobot has joined #mlpack
chvsp has quit [Changing host]
chvsp has joined #mlpack
aashay has quit [Changing host]
aashay has joined #mlpack
notarobot has quit [Changing host]
notarobot has joined #mlpack
notarobot has quit [Quit: Page closed]
diehumblex has quit [Quit: Connection closed for inactivity]