02:01 < kris> lozhnikov: I see the advantages for your method. Though please have look at the comments here https://gist.github.com/kris-singh/cdd952a793b7b7dc932ddac83c579761

02:02 < kris> I think i would submit a pr by tommrow for Gibbs Sampling

05:12 < sgupta1> Hi Ryan. I have made another image mlpack: dependencies. I run that inside a container and then build mlpack-2.2.0 inside it. Also, ran the mlpack_test and got output as "no errors detected". The container's name is admiring_darwin. You can do a "docker logs admiring_darwin" to see it all. Alternatively, I will run in this command in the Jenkins so you can look at the console output from there as well.

05:16 < sgupta1> Have a look at the jenkins-docker job's console output on Jenkins server.

12:03 < kris1> Hi, lozhnikov were you able to look at the comment

12:03 < kris1> lozhnikov:

13:15 < rcurtin> sgupta: sounds good, I will take a look when I am not on a phone :)

13:15 < rcurtin> I think we shpuld put the Dockerfiles you are making in the jenkins-conf repository

13:15 < rcurtin> should*

13:16 < sgupta> On github?

13:17 < sgupta> Found it. Yes! This way people can see that yes I am doing something :)

13:18 < rcurtin> yeah and it also makes review / addition easier :)

13:19 < sgupta> Just let me know how to proceed with this? Should I send a PR? or what

13:20 < rcurtin> yeah a PR would be fine, I can review it later today

13:22 < sgupta> Okay. Sure.

13:26 < wiking> yoyoyo

13:26 < wiking> mlpack people

13:26 < wiking> what's the update frequency for http://www.mlpack.org/benchmark.html

13:26 < wiking> ?

13:26 < wiking> rcurtin,

13:48 < rcurtin> wiking: not very often... :(

13:48 < rcurtin> I am.running new benchmakrs now

13:48 < wiking> hehe

13:48 < wiking> do you know what is the shogun version that was used

13:48 < wiking> for the current numbers?

13:48 < rcurtin> hope to push new numbers soon

13:48 < rcurtin> not sure... 2.1.0?

13:49 < wiking> :DDDDDDDDDDDDDDD

13:50 < wiking> you install shogun with deb packages?

13:51 < wiking> or how do you take a shogun pkg?

13:51 < wiking> i mean lib?

13:53 < wiking> mmm

13:54 < wiking> rcurtin, i think you are using something that is 5.0.0 or later

13:54 < wiking> based on this https://github.com/mlpack/benchmarks/blob/master/libraries/shogun_install.sh#L36

13:54 < wiking> but then again i dont know what is on the website :P

13:55 < wiking> but then again

13:55 < wiking> https://github.com/mlpack/benchmarks/blob/master/config.yaml#L13

13:55 < wiking> :S

13:55 < wiking> 3.2.0 :D

13:55 < wiking> zoq, pingu

13:55 < wiking> zoq, bist du da?

13:56 < rcurtin> in transit now

13:56 < rcurtin> I will try and check when I get to where I am.going

13:56 < wiking> :>

13:56 < wiking> thnx heaps

13:56 < wiking> drive safely

14:12 < zoq> once I get back, I can dig up the version

14:13 < zoq> probably 50 min

14:15 < rcurtin> wiking: I think you are right, maybe it is 3.2.0

14:15 < rcurtin> I don't have the system that I ran those benchmarks on anymore

14:16 < rcurtin> the new benchmarks will definitely run against 5.0.0

14:16 < rcurtin> ah I see now 6.0 is out

14:17 < rcurtin> so we can update and use that (the shogun benchmarks haven't started running yet)

14:22 < sgupta> Hi Ryan!. I just created a PR. :) yay!

14:31 < rcurtin> sweet, thank you

14:32 < rcurtin> I am debugging why a system keeps powering off during the night at my work, but once that is done (and maybe a few other things) I will take a look

14:49 < zoq> wiking: Shogun 4.1.0 ; there is an open issue to intergrate the version numbers, I think Dewang will fix this issue over the summer. Also, we have install scripts for each lib, https://github.com/mlpack/benchmarks/tree/master/libraries maybe you can take a look at the shogun script? We will do our best to update all results over the summer maybe not only for the latest version.

14:57 < shikhar> Hi everybody!

14:57 < shikhar> Working on integrating CTest with the existing tests, just noticed that 2 tests are excluded from the list of sources in CMakeLists.txt in the test folder

14:58 < shikhar> lstm_peephole_test.cpp and layer_traits_test.cpp

14:58 < shikhar> Is there some problem with these tests?

15:00 < rcurtin> looks like that might be an oversight, but I'm not sure---I didn't write those two

15:03 < shikhar> They were not listed by ./mlpack_test --list_content, I'll try running them now to see if they were failing

15:04 < zoq> shikhar: ah I know I missed some tests as I refactored the code

15:04 < zoq> shikhar: we can delete those two

15:04 < shikhar> Alright, great

15:04 < zoq> do you like to do it or should I?

15:05 < shikhar> I'll be creating a PR soon

15:05 < zoq> okay, great, thanks!

15:10 < kris1> zoq: were you able to look at my comments on the gist.

15:12 < zoq> kris1: yes, I left a comment, at least that's what I thought

15:16 < zoq> maybe I don't get the idea behind creating the distribution inside GibbsSampler, so I think the user could pass the distribution if the default constructor isn't enough.

15:24 < wiking> zoq, oh i see

15:25 < wiking> yeah i saw the isntall script

15:25 < wiking> that looks fine

15:25 < wiking> the only thing it needs is bumping the version :)

15:26 < zoq> wiking: okay, that's easy

15:43 < lozhnikov> kris1: yes, I saw your comment and I replied here https://gist.github.com/kris-singh/cdd952a793b7b7dc932ddac83c579761

16:13 < kris1> Lozhnikov oh i see your point...... the inputs of say mean, variance for P(y|x) are coming from a layer previous to the gibbs sampling layer

16:17 < kris1> With that the user can use any distribution as long as the previous layer output's provides the parameters for the conditional distribution used in gibbs sampling

16:17 < kris1> Lozhnikov:

16:17 < lozhnikov> kris1: right

16:21 < kris1> okay just one more thing i, so in the case of sigmoid the hidden layers are sampled as follows h_i = sigmoid(W_i *x + b) but suppose i want to normal distribution h_i = N(Output1 of prev layer, Output of previous layer) so i would sample n(# of hidden units) times from this Normal distribution if the output > 0.5 then h_i =1 else 0

16:21 < kris1> is this interpretation correct

16:22 < kris1> or did you have somthing else in mind lozhnikov:

16:22 < lozhnikov> I think you have to pass 2*n variables (n for the mean and n for the diagonal variance) to the next layer (sampler)

16:27 < kris1> Hmmm okay, and Gibbs layer would be a base policy layer where we could implement several policy now how to combine the 2n variables for mean and variance

16:29 < kris1> lozhnikov:

16:35 < lozhnikov> kris1: yes, the layer can be implemented like BaseLayer (methods/ann/layer/base_layer.hpp)

16:38 < kris1> Ohkkkay thanks for the help ... i will try to implement the base gibbs layer first and try to create some policies for the layer. For testing i think we could actually try to try to approximate some standard distribution. Whaat do you think ?

16:40 < lozhnikov> well, I think you could use functions from core/math/random. The implementation is straightforward so I don't think that the tests are needed for that.

16:41 < kris1> Ahh okay i will get to it then

18:02 < rcurtin> chenzhe: my OS X setup is too built on top of MacPorts for me to reproduce your issue with homebrew...

18:02 < rcurtin> could I ask you to try to reproduce the situation again with my branch 'include-ordering'? https://github.com/rcurtin/mlpack

18:37 < chenzhe> rcurtin: yes, I reproduced it

18:39 < rcurtin> ok, did the changes in the include-ordering branch fix the issue?

18:40 < chenzhe> i just reproduced the make error with include-ordering branch

18:40 < rcurtin> hm, strange that the change did not fix it

18:40 < rcurtin> could I ask you to run 'VERBOSE=1 make' and put the output on pastebin or something so I can look through it?

18:54 < chenzhe> https://gist.github.com/czdiao/7caf730ed12e43bcc65ee2061cff1d1f

18:54 < chenzhe> I added to the same gist

18:56 < chenzhe> no, wait, it is not complete output

18:56 < rcurtin> yeah the verbose output is huge :)

18:57 < chenzhe> weird, i will email you the output file

19:00 < chenzhe> did you get it?

19:00 < rcurtin> yep, thank you

19:02 < rcurtin> ah, I may have misunderstood the BEFORE directive in CMake... can you try 'git pull' then clean, reconfigure, and rebuild?

19:02 < rcurtin> -I/Users/chenzhe/mlpack_tmp -I/usr/local/include -I/Users/chenzhe/mlpack_tmp/src/mlpack/.. -I/Users/chenzhe/mlpack_tmp/build/src/mlpack/..

19:02 < rcurtin> but the /usr/local/include should come after the /Users/chenzhe/mlpack_tmp/src/mlpack/..

19:06 < chenzhe> not solved...

19:09 < rcurtin> hm, interesting, can I get the output now?

19:09 < rcurtin> with that output I can probably figure it out, but I don't think I have more time to look into it today :(

19:09 < rcurtin> thanks again for the help

19:16 < chenzhe> sure. I just sent it to you

20:23 < rcurtin> chenzhe: willing to give it one more try? I think I got it this tine

20:24 < rcurtin> *time... :)

22:40 < chenzhe> rcurtin: Yes, it finally works!!!

23:24 < rcurtin> chenzhe: great, I will merge in the changes tomorrow then

23:41 < kris1> Where is the sigmoidLayer that i being used

23:41 < kris1> Which folder i mean

23:42 < kris1> it does not reside in ann/layer/

23:44 < kris1> Got it sorry