02:05 < favre49> manish7294: okay, thank you. Also, I live in Bangalore

12:26 < jenkins-mlpack2> Project docker mlpack nightly build build #327: FAILURE in 1 day 8 hr: http://ci.mlpack.org/job/docker%20mlpack%20nightly%20build/327/

12:29 < jenkins-mlpack2> Project docker mlpack weekly build build #48: STILL FAILING in 14 hr: http://ci.mlpack.org/job/docker%20mlpack%20weekly%20build/48/

12:30 < jenkins-mlpack2> Project docker mlpack nightly build build #328: STILL FAILING in 3 min 54 sec: http://ci.mlpack.org/job/docker%20mlpack%20nightly%20build/328/

13:36 < sumedhghaisas> Hey KimSangYeon

13:36 < KimSangYeon-DGU> Hi Sumedh!!

13:36 < KimSangYeon-DGU> I have a question

13:36 < sumedhghaisas> go ahead...

13:36 < KimSangYeon-DGU> What is a good research project you think?

13:37 < KimSangYeon-DGU> I'd like to propel our project to the direction.

13:38 < sumedhghaisas> hmm... I am not sure I get the question correctly. Do you mean what hypothesis will we be checking?

13:40 < KimSangYeon-DGU> Yeah :)

13:40 < KimSangYeon-DGU> We have a schedule to plot the 3D probability space of Quantum GMM

13:41 < sumedhghaisas> sure. I think a good hypothesis is 'does QGMM provide better representation power than GMM'

13:41 < sumedhghaisas> what do you think?

13:41 < KimSangYeon-DGU> Great!!

13:41 < KimSangYeon-DGU> Yes

13:41 < ShikharJ> sumedhghaisas: Sorry for disturbing, are you tackling an open problem with Quantum GMMs over the summer? I thought, the project was regarding implementation of QGMMs?

13:42 < sumedhghaisas> and regarding the 3D plot I think its a good stepping stone

13:42 < sumedhghaisas> once we have an idea how does the distribution look like compared to GMM we will have better idea about our path

13:42 < KimSangYeon-DGU> Yeah :)

13:43 < sumedhghaisas> SikharJ: Hey Sikhar. Yes, I think QuantumGMM is an open problem. We will be implementing QGMM but its not clear yet if it will work exactly as its described

13:43 < sumedhghaisas> thus we will model this project as a research question instead

13:44 < ShikharJ> sumedhghaisas: Super interesting. I'd love to see the progress on that.

13:44 < KimSangYeon-DGU> ShikharJ: I'll try to it :)

13:45 < sumedhghaisas> KimSangYeon-DGU: Do you have any idea how would you like to proceed with 3D modelling?

13:45 < KimSangYeon-DGU> with my mento, Sumedh

13:45 < sumedhghaisas> SikharJ: Indeed! I have his project at the back of my head for some time noew. Never got time around it.

13:46 < KimSangYeon-DGU> sumedhghaisas: Yes, Once we implement the QuantumGaussianDistribution and QuantumGMM, I get some results of it and plot them using Python libraries

13:47 < sumedhghaisas> SikharJ: You are more than welcome to join our weekly meets also. We would mostly be talking about the problem. Would love to hear external views as well.

13:47 < KimSangYeon-DGU> Is is a proper way??

13:47 < KimSangYeon-DGU> *it

13:47 < KimSangYeon-DGU> sumedhghaisas: Definitely :)

13:48 < sumedhghaisas> KimSangYeon-DGU: Given the nature of the probelm I would first recommend doing it on paper. I see couple of problems in getting the distribution.

13:48 < ShikharJ> KimSangYeon-DGU: I should probably mention, your proposal was simply amazing. All 46 pages of it, it was three times the size of my original proposal a year back (I think that's a new record for mlpack) :P So can't wait to see the implementation of it as well :)

13:48 < sumedhghaisas> if we have a way of getting the distribution on the paper, we can do quick plotting using matplotlib

13:50 < KimSangYeon-DGU> sumedhghaisas: Ah~ yes, I'll do that. Thanks for letting me know

13:50 < ShikharJ> sumedhghaisas: Thanks, though I can't say I'll be experienced with QGMMs, I'm barely discovering the math behind GMMs myself.

13:50 < KimSangYeon-DGU> ShikharJ: I think it was a great time to work with you. I'll do my best to make this project successful :) Thanks!!

13:51 < sumedhghaisas> KimSangYeon-DGU: I think we can concentrate on Equation 9 in the paper. We need to plot the same distribution first.

13:52 < KimSangYeon-DGU> sumedhghaisas: Yeah, I'll try to it using matplotlib

13:53 < sumedhghaisas> great. The real trouble is finding the normalization constants 'the alphas' in the equation

13:54 < KimSangYeon-DGU> sumedhghaisas: Agreed

13:54 < KimSangYeon-DGU> Well, before the coding period, I'll check the paper using matplotlib

13:55 < KimSangYeon-DGU> How can I show you the result??

13:55 < sumedhghaisas> KiSangYeon-DGU: Thats a good idea. Although I have some doubts in Equation 9. I still haven't understood why have they taken sum over all the data points

13:56 < sumedhghaisas> I think it should we integration over the distribution rather than sum

13:56 < sumedhghaisas> what do you think?

13:56 < KimSangYeon-DGU> Yes, actually, I worried about it

13:56 < KimSangYeon-DGU> Oh

13:56 < KimSangYeon-DGU> integration will be better

13:57 < KimSangYeon-DGU> Did you intend QGMM for n-classes??

13:57 < KimSangYeon-DGU> not 2-classes that original paper presents

13:57 < sumedhghaisas> Good. If both of us think that then there is some chance that we are right :P while plotting the distribution we have no data points ..., so we have to integrate over the whole distribution

13:58 < sumedhghaisas> for the plotting? we can plot for 2 classes... no worries. Any single interference pattern will do just fine

13:58 < KimSangYeon-DGU> Okay!

13:59 < sumedhghaisas> okay I need to logout for sometime. I will be back in 20 minutes. :)

13:59 < KimSangYeon-DGU> Yeah

13:59 < KimSangYeon-DGU> Thanks for the meeting!!

14:01 < KimSangYeon-DGU> sumedhghaisas: I missed one!

14:03 < KimSangYeon-DGU> I think the reason paper use sum is because it uses log-probability and took exp.

14:03 < KimSangYeon-DGU> *uses

14:06 < KimSangYeon-DGU> *used

14:12 < KimSangYeon-DGU> I'll give it a shot! Good luck with your class you teach :)

14:29 < sumedhghaisas> KimSangYeon-DGU: I dont think I understood that right. Where is the log involved?

14:31 < KimSangYeon-DGU> sumedhghaisas: Well, it's my guess but it's not mentioned directly.

14:36 < sumedhghaisas> ahh I see. But when we are not taking dataset into consideration, the plain distribution has to be integrated to find the total area under the curve

14:38 < KimSangYeon-DGU> Ahh, Yeah :)

15:18 < lozhnikov> jeffin143: Hi, I am still looking through #1814. I hope I'll finish in a couple of hours.

16:46 < jeffin143> lozhnikov : sure,

19:25 < jeffin143> lozhnikov : if I have to split a string into char , for exam jef into <j><e><f> , none of the standard tokenizer does it, how should I handle that case..?

19:30 < zoq> Couldn't you use the operator[]? Or maybe the string you like to parse is more complex?

19:33 < jeffin143> Zoq I am currently using boost::tokenizer

19:36 < jeffin143> Boost::char_separator<char>sep("@");

19:36 < zoq> Right, I guess you could use standard C++.

19:37 < zoq> @ is the delimiter?

19:37 < jeffin143> Boost:: tokenizer<boost::char_separor<char>>tokenizer (str,sep)

19:37 < jeffin143> Will tokenize using @

19:38 < jeffin143> Yes @ is the delimiter

19:38 < jeffin143> I want way out that if user passes @ , then it should tokenize using that , or else if user doesn't pass anything then tokenize to character or if user pass @-+ then tokenize using @-+

19:39 < jeffin143> Basically a template version

19:40 < zoq> What happens if you pass nothing as the delimiter?

19:41 < zoq> You could use boost for the user used a delemiter and use standard C++ for the other case.

19:41 < zoq> Something like: std::string myString = "Hello"; std::vector<char> myStringVector; myStringVector.push_back(myString[0]) ....

19:45 < jeffin143> If I pass nothing as the delimiter, It is taking all punctuation and space as default, that is the definition of boost

19:45 < jeffin143> I will go with your idea of doing 2 things, if user passes a delimiter then do that and if not then use STD::c++