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/
< TD>
I am unfamiliar with CMake. Is this the code
< TD>
and Do I need to add it to CMakeList at the root?
Cooler_ has quit [Ping timeout: 240 seconds]
marcosirc has quit [Quit: WeeChat 1.4]
nilay has quit [Ping timeout: 250 seconds]
nilay has joined #mlpack
nilay has quit [Quit: Page closed]
mentekid has joined #mlpack
Mathnerd314 has quit [Ping timeout: 246 seconds]
nilay has joined #mlpack
nilay has quit [Ping timeout: 250 seconds]
< zoq>
TD: It should be sufficient to set: set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /bigobj") inside the if (MSVC) block; line 224 (root CMakeLists.txt file)
nilay has joined #mlpack
< zoq>
nilay: Hello, so, I think the next steps are, writing tests for the feature extraction part; afterwards move on with the implementation of the mapping function (discretize) we need to train a single structured tree. Do you agree?
< nilay>
zoq: yes
< nilay>
so i have read the decision stump
< nilay>
and i think we can implement the decision tree, by using the decision stump recursively
< nilay>
i have still to read the hoeffdding tree part.. and i have been working on the test cases also
< zoq>
Okay sounds good, I would say, let's finish the test cases first and then move on to the tree part.
nilay_ has joined #mlpack
nilay has quit [Ping timeout: 250 seconds]
< nilay_>
For test cases I don't know how to go about it concretely. Right now I am only checking the size of the matrices..
< zoq>
Do you think, it would be possible to compare against the reference implementation?
< nilay_>
How do we check that this is indeed the matrix output that we want
< nilay_>
Yeah something of that kind, and then check whether SSE is below a limit
< nilay_>
But I guess for the example features we have used they are random
< nilay_>
So that can't be done. And that's why I think only after random forest is built we can check against reference
< zoq>
If we use the same input for e.g. rgb2luv the output should be similar. And I agree, we can't test the complete pipeline, but we can test single functions, if we set the same location.
< zoq>
We have to make the functions public, but we can test them.
< nilay_>
Yeah so I can check those computer vision functions that we have written
< zoq>
that sounds perfect
< nilay_>
OK and we wouldn't require an object instance to call those functions?
< zoq>
We do, but that shouldn't be a problem: StructuredForests sf; sf.RGB2LUV(input); should work
< nilay_>
Then why do we need to make them public
< zoq>
You can only call private functions inside the class definition.
< nilay_>
Oh Ok yes.
nilay_ has quit [Ping timeout: 250 seconds]
< keonkim>
TD, rcurtin: I came back to vim. Visual Studio isn't my thing. :p
< keonkim>
Clion also seems to be a good alternative.
marcosirc has joined #mlpack
< rcurtin>
looks like debugging symbols accelerated the appveyor build by about 3 minutes, I guess no need to change them back then
conrad1B39CC has joined #mlpack
< conrad1B39CC>
hi ryan, do you have 2 mins ?
< rcurtin>
conrad1B39CC: sure, I am preparing to present a paper at a reading group but I can spare two minutes
< conrad1B39CC>
great. I'm planning to submit the short paper on Armadillo to JOSS (journal of open source software). I'd like to have you as a co-author. What affiliation should I use?
< rcurtin>
Symantec Corporation... can you give me the rest of the day to respond to your email, or does it need to happen now?
< conrad1B39CC>
rest of day is ok :) feeding the baby here, so making use of time while i can
< rcurtin>
sounds good, hope the baby is doing well :)
< rcurtin>
you'll have an email from me within the next 8 hours then
< conrad1B39CC>
almost walking... thanks :)
< rcurtin>
wow, already? time moves fast
< rcurtin>
I guess calculus lessons start soon :)
< conrad1B39CC>
they grow like mushrooms, they say. skip calculus and go directly to quantum mechanics
< rcurtin>
haha
< conrad1B39CC>
gagggaggaga. eh. <-- the little one says
< rcurtin>
:)
< conrad1B39CC>
i'll send you the updated paper via email.
< rcurtin>
ok, thanks
< rcurtin>
luckily these appear to be quite short (just an abstract), so, a quick read :)
< conrad1B39CC>
yeah, I like them. quick to the point, no mucking about. the stated goal of the journal is to give academic credit to software used for research, which I think this does nicely (paper gets a DOI etc).
< rcurtin>
it seems like a good goal, getting academic credit for software is a difficult problem
conrad1B39CC has quit [Quit: Page closed]
Mathnerd314 has joined #mlpack
zootorg has joined #mlpack
zootorg has quit [Client Quit]
< mentekid>
quick question because I think I've been doing something very stupid very long
< mentekid>
if I change a file and recompile it, do I need to recompile mlpack_test as well?
< mentekid>
or just mlpack_headers?
tham has joined #mlpack
< tham>
nilay zoq : I think I can implement the decision tree and random forest
< tham>
About the test cases, could we use part of the python implementation as reference?
< tham>
Treat the output of python as golden model
< tham>
What do you think?
< tham>
About the encapsulation
< tham>
Maybe we can add one more layer?Like change the name of original class to StructuredForestsImpl
< tham>
put it in the namespace mlpack::structured_tree::details
< tham>
and use it to implement StructuredForests
< tham>
This way we could test the function works or not by testing StructuredForestsImpl, yet provide encapsulation(make StructuredForests as friend of StructuredForestsImpl)
< tham>
Sorry, don't need to make StructuredForests as friend of StructuredForestsImpl, just put StructuredForestsImpl in details namespace and make the api become public
< tham>
I would like to implement a simple random forests before, If you don't mind I would like to take care of this task
< zoq>
tham: I guess, using the python code as reference is the plan.
< zoq>
tham: I'm not sure it's a good idea if you implement the decision tree and the random forest. However, I guess there is no reason not to help nilay with the implementation.
< zoq>
tham: Also, I'm not sure what you like to achieve with StructuredForestsImpl.
< tham>
Well, then, I would leave decision tree and random forest for nilay
< tham>
StructuredForestsImpl, the idea is, use it to expose the api
< tham>
Since it is put in the details namespace, by default, this class should not be used by the users(the comments should say that too)
< tham>
After that we could use StructuredForestsImpl to implement StructuredForests
< tham>
But this solution need the users to follow the rules without the check of compiler
< zoq>
tham: I don't mind to make them public, in fact, this would give us the ability to reuse some of the functions. Also, you can use every function independently of the class, since it doesn't depend on some class member.
< zoq>
tham: Maybe we should do that, but I think there is no actual need to do so.
< tham>
The class do have some useful functions for image processing/computer vision
< zoq>
tham: yes, I agree
govg has joined #mlpack
tham has quit [Quit: Page closed]
tsathoggua has joined #mlpack
< rcurtin>
mentekid: yep, you need to recompile mlpack_test if you modified a source file that mlpack_test depends on and want to run an updated version of mlpack_test
travis-ci has joined #mlpack
< travis-ci>
mlpack/mlpack#957 (master - e8e2ff1 : Ryan Curtin): The build was fixed.
< TD>
KeonKim, you might want to add if while building mlpack in a windows Visual Studio environment and you receive error C1128 go to the the specific project in Solution Explorer highlight it > Next go to 'Project' > 'Properties' > 'C/C++' > 'Command Line' and add /bigobj to 'Additional Options. After that mlpack will build successfully on Visual Studios 15
< TD>
mlpack officially installed in a Windows 10 Visual Studios 15 environment. Thanks KeonKim and every one else for the help
< TD>
Looking forward to using this in my research