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/
rf_sust2018 has joined #mlpack
rf_sust2018 has quit [Ping timeout: 250 seconds]
rf_sust2018 has joined #mlpack
rf_sust2018 has quit [Quit: Leaving.]
seewishnew has joined #mlpack
root has joined #mlpack
root is now known as Guest21391
Guest21391 has quit [Client Quit]
KimSangYeon-DGU has joined #mlpack
KimSangYeon-DGU has quit [Ping timeout: 256 seconds]
Jhar23 has joined #mlpack
Jhar23 has quit [Client Quit]
shine_ has joined #mlpack
travis-ci has joined #mlpack
< travis-ci>
mlpack/mlpack#6584 (master - febc8e3 : Ryan Curtin): The build has errored.
< mulx10>
rcurtin: Will you abe able to review my proposal in your spare time?
jeffin143 has quit [Ping timeout: 255 seconds]
< mulx10>
rcurtin: If yes, thanks it would be great help to me.
< mulx10>
rcurtin: If no, it's fine :).
< mulx10>
thanks
mulx10 has quit [Ping timeout: 256 seconds]
Shiv has joined #mlpack
Shiv has left #mlpack []
seewishnew has quit [Remote host closed the connection]
reborux has joined #mlpack
jeffin143 has joined #mlpack
reborux has quit [Ping timeout: 256 seconds]
KimSangYeon-DGU has joined #mlpack
shine_ has joined #mlpack
< shine_>
zoq: i had a little doubt about the reinforcement learning project idea in gsoc
< shine_>
The papers provided in the idea improve upon the q-learning based reinforcement learning implemented for atari using deep learning. So, do we just need to add the functionality that implements those improvements and can be used in building a reinforcement learning model or are we needed to implement the reinforcement learning algorithm along woth these improvements specifically with atari games in mind and show the full
< shine_>
implemented model learning and playing the game at the end of gsoc
mulx10 has joined #mlpack
mulx10 has quit [Ping timeout: 256 seconds]
favre49 has joined #mlpack
< favre49>
rcurtin: I'm sorry about this, but some stuff has come up and I won't have the time to work on PR #88 on ensmallen until April, probably. Is there anyway i can remove the "needs review" flag for the time being? Of course, anyone else is welcome to work on it.
favre49 has quit [Client Quit]
KimSangYeon-DGU has quit [Quit: Page closed]
vinayak has joined #mlpack
< vinayak>
hello sir.
< vinayak>
i had send my tentative proposal for GSOC 2019 on mlpack@lists.mlpack.org please have a look. Thank you
vinayak has quit [Client Quit]
shine_ has quit [Quit: Connection closed for inactivity]
KimSangYeon-DGU has joined #mlpack
KimSangYeon-DGU has quit [Ping timeout: 256 seconds]
ark has joined #mlpack
ark has quit [Ping timeout: 256 seconds]
< rcurtin>
mulx10: I'll see if I can take a look, sure
< rcurtin>
favre49: no problem! we can handle it in April when you're free or someone else can
< rcurtin>
I'll remove the 'needs review' tag
Toshal has joined #mlpack
jeffin143 has quit [Read error: Connection reset by peer]
Suryo has joined #mlpack
< Suryo>
rcurtin, zoq: I had a few questions about Armadillo.
< Suryo>
I've tried to trace the code where matrix/cube additions happen and I think that I need to be looking at arrayops_meat.hpp
< Suryo>
And the base addition operation is done from line 392
< Suryo>
And over here, basically, all the elements in an array are looped over and added one-by-one.
< Suryo>
rcurtin: if you can take a look at this, kindly correct me if I'm wrong.
< rcurtin>
Suryo: I think I know what you are talking about, yeah; you mean the code that gets run when you do, e.g., 'a + b' and both a and b are Armadillo objects, right?
< rcurtin>
if I remember right code for the elementwise operations like this is actually generated by a kind of big ugly macro
< rcurtin>
and the specific loops that are used are tuned so that they take advantage of SIMD instructions and do some amount of loop unrolling
< rcurtin>
ah, sorry, that is something else, what I was referring to is code that's run for, e.g., elementwise log(), exp(), etc.
mulx10 has joined #mlpack
< rcurtin>
in any case, yeah, this is basically the same; the loops are written such that the compiler should be able to take advantage of SIMD
< Suryo>
"you mean the code that gets run when you do, e.g., 'a + b' and both a and b are Armadillo objects, right?" : yes
< Suryo>
SIMD. Okay, I'll take a look. Thank you tons for the prompt reply!
< Suryo>
zoq: Thank you for the review. I'll fix everything that you've pointed out over this week and the next.
richard has joined #mlpack
richard is now known as Guest4742
Guest4742 has quit [Client Quit]
< Suryo>
rcurtin, zoq: In the best interest of time, I'll leave my study of armadillo for later. For now, I'll focus on implementing some degree of parallelism within ensmallen, though it will still remain a secondary objective.
Suryo has quit [Quit: Page closed]
< rcurtin>
Suryo: yeah, so what you can assume about Armadillo with respect to parallelism is that Armadillo itself does not really do any internal parallelism (there are a few exceptions)
< rcurtin>
but you should expect that if a lower-level LAPACK or BLAS function gets called, the user will typically be using OpenBLAS, which internally is parallel
< rcurtin>
so, e.g., calls to solve() or chol() or matrix multiplications will already be parallel
< rcurtin>
but that code that you just looked at is not parallelized, it is only SIMD
KimSangYeon-DGU has joined #mlpack
Toshal has quit [Read error: Connection reset by peer]
Suryo has joined #mlpack
< Suryo>
rcurtin: that clarifies things a bit. But I'll read up more on these.
Suryo has quit [Remote host closed the connection]
jeffinsam has joined #mlpack
sdjfhnfdf has joined #mlpack
sdjfhnfdf has quit [Client Quit]
mulx10 has quit [Ping timeout: 256 seconds]
mulx10 has joined #mlpack
< jeffinsam>
zoq, shikhar j, rcurtin : would it be apt to keep a chek on the dimension passed in binarize function in binarize.hpp, so that user doesn't pass out of a dimension more than the size of matrix
Lorcan has joined #mlpack
< rcurtin>
jeffinsam: yeah, that would be nice, you can throw a std::invalid_argument if the user screwed up :)
< rcurtin>
that's better than invalid memory accesses and crashes :)
< jeffinsam>
also , may be we could overload the function to pass , a vector of rows to binarize, presently it is binarizing the whole matrix or a particular row
< jeffinsam>
i would like to hear your take on the above idea
< rcurtin>
yeah, that would be nice for sure
< jeffinsam>
ok so i would raise a pr for that :) if you are okay with it
< rcurtin>
yeah, sure, please go ahead
< rcurtin>
I will try and find time to review it :)
< rcurtin>
yesterday I went through everything because I wanted to make sure that everyone preparing an application for GSoC had at least some feedback, but I won't be able to keep up that pace for sure
< rcurtin>
za
< rcurtin>
oops, ignore the 'za', typo :)
< jeffinsam>
yeah, i can understand, btw thanks for the review on pr
< jeffinsam>
i had that bigger picture of one hot encoding, but thought of writing in gsoc application
jeffin143 has joined #mlpack
robb8 has joined #mlpack
< robb8>
how can I calculate my neural network's mean squared error _after_ it's done training? the double that model.Train() returns does not seem to ever change as I add new layers, etc
< robb8>
also, you guys rewrote the docs didn't you? I can't google search for this stuff anymore lol
Toshal has joined #mlpack
jeffinsam has quit [Ping timeout: 256 seconds]
i8hantanu has quit [Quit: Connection closed for inactivity]
Dibyanshu has joined #mlpack
< Dibyanshu>
Hi , I am going through GSOC Organizations, and i found out mlpack , I have taken up the machine learning course from Standford Online Courses by Prof Andrew Ng . I am proficient in data stuructures and algorithms , since i am doing Compeditive programming for more than a year . So , I am very intrested in this project . Can i get further details , how can i proceed ?
< rcurtin>
I noticed that the Travis CI builds aren't actually testing the Python bindings! but it should be a relatively simple fix
< rcurtin>
I'll open a PR shortly
< zoq>
it does run e.g. NMFMainTest, dosn't that include the python bindings?
shine_ has joined #mlpack
Lorcan has quit [Ping timeout: 256 seconds]
robb9 has joined #mlpack
Toshal has quit [Ping timeout: 246 seconds]
AnirudhYadav has joined #mlpack
Toshal has joined #mlpack
AnirudhYadav has quit [Ping timeout: 256 seconds]
rajiv_ has joined #mlpack
Dibyanshu has quit [Quit: Page closed]
< rcurtin>
zoq: the main tests actually just run the special "test" binding type, not the Python bindings
< rcurtin>
so there is this language-agnostic test type that basically is just a wrapper that lets you pass things in through CLI and then run mlpackMain()
< rcurtin>
since we have Python bindings (and other language bindings) and tests for the specific functionality there, we don't test each Python binding directly
< rcurtin>
the idea is this: we test that we can pass different Python objects to C++ and back in python_bindings_test (and cli_bindings_test, and soon julia_bindings_test)
< rcurtin>
but once we have that tested, we assume that part works, and then the 'main_tests/' directory assumes that mlpackMain() will receive things properly through CLI, and the idea is to test that each binding does reasonable things with its input (especially when the input is invalid)
mulx10 has quit [Ping timeout: 256 seconds]
< robb9>
is RMSProp random?
robb9 has quit [Ping timeout: 256 seconds]
< rajiv_>
zoq, rcurtin: Is there a method to parse xml annotation in arma?
< rcurtin>
rajiv_: not that I know of
< rcurtin>
robb9: what do you mean by random? do you mean the order in which it visits points? if you're using it from ensmallen, then it visits points in a random order if `shuffle = true`
< rcurtin>
if you're using rmsprop as an optimizer in some mlpack command-line or Python binding, it'll visit points after shuffling the data unless you specify "linear_scan" as "true" (which is not the default)
mulx10 has joined #mlpack
< rcurtin>
by the way the default `shuffle` value for ensmallen is `true`
< rcurtin>
right, change 'docs/' to 'doc/' and you should get it
< rcurtin>
all the paths for the documentation just changed with the website being redone... I wonder if I should set up an apache rewrite rule or something? or just wait until Google updates it cache
< rajiv_>
Yeah google still hasn't updated xD Do you suggest any method to parse XML annotations? I am preparing a proposal for YOLO architecture which involves this data... It would be great if you could suggest something :)
< rajiv_>
Adding additional library would add unnecessary dependancy... Hence I thought of regex... But I'm not sure
mulx10 has quit [Ping timeout: 256 seconds]
< zoq>
rcurtin: Thanks for the clarification.
rajiv_ has quit [Quit: Page closed]
< rcurtin>
rajiv_: I am not sure, maybe something with boost::spirit which we are already using?
< rcurtin>
boost::spirit is pretty ugly though...
< rcurtin>
or, well, maybe not ugly. but weird and hard to learn for sure :)
divyansh997 has joined #mlpack
Toshal has quit [Ping timeout: 245 seconds]
sreenik has joined #mlpack
sreenik has quit [Client Quit]
ranjan has joined #mlpack
ranjan has quit [Client Quit]
sreenik has joined #mlpack
< sreenik>
rajiv: you could check out boost's xml reader.
< sreenik>
zoq: I had a flight today, I could just skim through what you had suggested. I should be able to look further into the kcpa solution tomorrow. I shall let you know if I need to know anything in specific. Thanks :)
< gauravcr7rm>
zoq: Hello marcus , I have sent you my tentative proposal for the NEAT project on your personal mail id.
< gauravcr7rm>
zoq: I have incorporated GRAPH X crossover and am also have included MM-NEAT in the proposal. Please review it whenever you have time and do suggest some changes. Thanks
arisat has joined #mlpack
< arisat>
is this the right place to ask queries regarding gsoc?
< zoq>
gauravcr7rm: Actually do you mind to upload the draft to via the gsoc dashboard, makes it easier to keep track of what is there?
< zoq>
arisat: Hello, yes it is.
gauravcr7rm has quit [Ping timeout: 256 seconds]
< arisat>
zoq : Thanks, I am interested in working on the PSO project
< arisat>
I wanted to know where should I start from?
< arisat>
I am Arijeet Satapathy, Senior Year CS Major Student at IIIT Bhubaneswar, India
< zoq>
arisat: I see, there are some discussions about the project on the mailing list, note take a look at the archive from last year as well: http://mlpack.org/pipermail/mlpack/
< arisat>
zoq : Ok, will go through them. Thanks a lot!!
< zoq>
arisat: Happy to clarify anything if you have any question.
favre49 has joined #mlpack
shine_ has quit [Quit: Connection closed for inactivity]
< favre49>
zoq: Would you like me to also resubmit my draft proposal through the GSoC dashboard? I've already mailed it to you
favre49 has quit [Client Quit]
arisat has quit [Quit: Page closed]
AlforML has joined #mlpack
< zoq>
favre49: That would be great, I was about to write you a mail to suggest the same.
< zoq>
favre49: Getting feedback is part of the application process, so Google added a "feature" for that.
divyansh997 has quit [Quit: Going offline, see ya! (www.adiirc.com)]
syko has joined #mlpack
< syko>
if i only know cs50 and a little bit about object oriented programming on c++ , should i submit or i am not good enough?
< rcurtin>
syko: I don't know what cs50 is, probably the course number of an introductory programming class?
< rcurtin>
I would say, take a look through the list of projects, and if you comfortably feel like you can do one of them, then it would be reasonable to submit a proposal
< syko>
yea introductory programming class from harvard its free on edx
divyansh997 has joined #mlpack
divi has joined #mlpack
syko has quit [Quit: Page closed]
divi_ has joined #mlpack
divyansh997 has quit [Ping timeout: 255 seconds]
divi_ is now known as divyansh997
divi has quit [Ping timeout: 250 seconds]
som has joined #mlpack
som has quit [Client Quit]
Suryo has joined #mlpack
< Suryo>
zoq: I'll be putting my proposal up really soon, too. I sure hope that you'll get a chance to look at it!
Suryo has quit [Client Quit]
< zoq>
Suryo: Great :)
< zoq>
shine_: Sorry for the slow response, I used Atari just as an example, if you have an interesting RL related method in mind that please feel free to propose one; the methods listed are just ideas.