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/
< HimanshuPathakGi> Hey @zoq @saksham189 can you suggest a dataset already in mlpack which is not linearly separable to test it with Gaussian Kernel
ImQ009 has joined #mlpack
< sakshamb189[m]> Can you try to use the mnist dataset for 4's and 9's? Let me know what you think.
< HimanshuPathakGi> Yup will try it today
pie3 has joined #mlpack
< pie3> Can i share tensorflow related queries here?
< birm[m]> pie3: Some people here may have some TF experience, but this isn't a tensorflow channel.
< pie3> OK birm[m]
< pie3> So to use mlpack, one must have C++, right?
< jeffin143[m]> > So to use mlpack, one must have C++, right?
< jeffin143[m]> You can install python binding and julia or go binding to use it in other languages
< pie3> ok
< pie3> mlpack is redudant of scikit or which other unique features it offer?
< jeffin143[m]> There is a benchmark repository where you could find the benchmarks between different libraries
< jeffin143[m]> > There is a benchmark repository where you could find the benchmarks between different libraries
< jeffin143[m]> Under Mlpack org
< pie3> ok
pie3 has quit [Ping timeout: 246 seconds]
< rcurtin> HimanshuPathakGi: another idea would be a "concentric circles" dataset; you could probably make it pretty easily with some tool
< rcurtin> basically put points on the unit ball for class 0, and put points on 2*the unit ball for class 1
< rcurtin> this is linearly non-separable but a kernel SVM with appropriate kernel should be able to separate those classes easily
pie3 has joined #mlpack
< pie3> <pie3> "The only way to get a constant, variable (any result) from TenorFlow is the session" is this true?
< pie3> <pie3> so whcih activties, can be performed outside session, generally?
< rcurtin> pie3: as far as I know that is true, but this really isn't the best place to ask tensorflow questions, this is the mlpack library discussion channel
< rcurtin> mlpack is a completely separate library and doesn't really have any relation to tensorflow
< pie3> i understand
< pie3> tf.compat.v1.disable_eager_execution()
< pie3> sess = tf.compat.v1.Session()
< pie3> when this is used?
< pie3> to run TF old version inside new version?
< pie3> what is equivalent code for new version?
< rcurtin> personally, I have no idea---I'm not a tensorflow user
< rcurtin> have you tried asking in a tensorflow support forum or something? I'm certain you'll have better success there
< pie3> yes, i did there also
< pie3> but that is quite silent
< pie3> mlpack can replace/resemble tensorflow/keras/pytorch as well?
< pie3> or it is not designed for deep learning?
< kartikdutt18[m]> Hey @pie3, We do support deep-learning. You can take a look at our examples repo for some tutorials or src/methods/ann to see various layers implemented in mlpack.
< rcurtin> zoq: I had a breakthrough for the FastMKS test memory issue!
< kartikdutt18[m]> [examples_repo](https://github.com/mlpack/examples)
< rcurtin> I noticed this in the logs:
< rcurtin> /home/jenkins-mlpack/workspace/pull-requests mlpack memory@2/src/mlpack/tests/main_tests/fastmks_test.cpp(510): fatal error: in "FastMKSMainTest/FastMKSOffsetTest": Cannot load test dataset data_3d_ind.txt!
< rcurtin> so I ran locally and was able to reproduce it... tons of memory errors
< rcurtin> then I moved data_3d_ind.txt to the working directory... no memory errors
< rcurtin> so I think this really just boils down to the datasets not being in the right place for the memcheck job
< rcurtin> however, I noticed that even with these test failures, they actually aren't reported in the job... so I'll look into setting up the memcheck job to also report failed tests
< zoq> rcurtin: Ohh, that makes sense, not sure why we missed the message.
< zoq> rcurtin: I think I configured the memory check, sorry for the trouble.
< rcurtin> well part of it is that the logs of the memory check are... in the GBs :-D
< rcurtin> no worries, just happy to have it figured out (I think) :)
< rcurtin> I know that if I encounter the issue, someone else will too
< zoq> yes
< rcurtin> maybe it is a variant of Murphy's Law but it seems like every possible un-debugged weird issue will come back and have to be debugged... so there is no hiding from it :)
< zoq> haha, yes you are probably right
< rcurtin> I'm just elated to have actually figured it out
< rcurtin> I was getting really unhappy... "how could there be a problem I can't solve? why can't I figure it out?"
< pie3> a.numpy() & a.numpy gives different results, what is difference between them? First one shows only data, later shows complete information
< rcurtin> but really I just got lucky while watching the build go
< rcurtin> in any case I have learned a new trick for debugging, so, hopefully it will help in the future regardless :)
< zoq> what is the new trick - watching the build log? :)
< rcurtin> yeah, basically, the thing that I didn't think of was "are the tests passing?"---I had just assumed that they were, and that's how I overlooked the problem
< rcurtin> that's also why I couldn't reproduce it---every time I tried, I was running from the "correct" working directory with the test data available
< rcurtin> but the issue can only be reproduced when running from a "wrong" working directory :)
< zoq> Great that you figured it out, so much time 'lost' on this.
< rcurtin> the time would have only been "lost" if I hadn't learned something in the end :)
< zoq> I guess to save some time you can enable the other build nodes?
< HimanshuPathakGi> Sure @rcurtin I will try "concentric circles" dataset also thanks for suggestion :)
< abernauer[m]> Debugging and reading logs can be time consuming and attention driven so nice catch rcurtin
< rcurtin> zoq: yeah, actually today I was going to add another build system from my house (this one less underpowered)
< rcurtin> I think we blew up "scott", it went offline
< rcurtin> I'll reboot it today and we can see what happens :)
ImQ009 has quit [Quit: Leaving]
< shrit[m]> @rcurtin: I will do a quick test with your proposition
< rcurtin> sounds good---let me know if I should go get my headphones if you want to hop on a call
< shrit[m]> Yes
< rcurtin> ok, hang on
< shrit[m]> I am here
k3nz0 has quit [Ping timeout: 256 seconds]
k3nz0_ has quit [Ping timeout: 265 seconds]
k3nz0 has joined #mlpack
k3nz0_ has joined #mlpack
pie3 has quit [Read error: Connection reset by peer]