02:12 < Kirizaki> hello

02:12 < Kirizaki> I'm have this problem to solve and just started some reserach

02:13 < Kirizaki> maybe some of you guys could give me some keywords which could fit some good algorithms / patterns

02:13 < Kirizaki> so the problem is:

02:15 < Kirizaki> I have clients, they belong to groups and they know to which group belong to; if there will be any change in group (add client, remove client, custom event..), all clients from that group must be notified

02:15 < Kirizaki> assumptions:

02:15 < Kirizaki> 1. clients can belong to multiple groups

02:16 < Kirizaki> 2. there is ~1 000 000 clients and let's say ~100 groups

05:41 < rcurtin> Kirizaki: long time no see! or ひさしぶり I guess :)

05:42 < rcurtin> I'm assuming that for your problem you're looking to automatically guess which group the clients are in based on some information about the clients

05:42 < rcurtin> otherwise maybe that is not a machine learning problem I guess

05:43 < rcurtin> for guessing which groups the clients are in, you might consider, perhaps, some clustering algorithms where a point can be a part of multiple clusters

05:47 < Kirizaki> ひさしぶり! :)

05:48 < Kirizaki> yes, it's not a ML kind of problem, but I suck at engineering and I've assumed that some of you guys will give me some hints :P

05:49 < Kirizaki> *clustering algorithms* - I will look on that

05:49 < Kirizaki> thanks Ryan

05:50 < rcurtin> ah, if it's not an ML problem, clustering would probably not be the right way to go

05:51 < rcurtin> if the goal is just to be able to notify all clients in a group, it seems like you could just hold a list of client IDs for each group and loop over that

05:51 < rcurtin> it's late here... I'm going to head to bed. hope that is helpful :)

06:38 < akhandait> sreenik[m]: How's it going?

07:01 < zoq> moraleja: Sorry for the slow response, yes in this case you have to write your own linear solver.

07:06 < zoq> akfluffy: You can just use the 2D conv layer, it does work on each channel but it will not run another conv op over the channels.

07:45 < jenkins-mlpack2> Project docker mlpack nightly build build #355: STILL UNSTABLE in 3 hr 31 min: http://ci.mlpack.org/job/docker%20mlpack%20nightly%20build/355/

10:33 < sreenik[m]> akhandait: I have put together the code. But I am yet to test it against different ONNX models and see if it is getting converted alright. One assumption I have made is that the the biases are stored *before* the weights in the parameter variable for mlpack (I will verify this by tonight). Other than that there are no issues "on the surface". After testing it might need some debugging.

10:35 < sreenik[m]> Before I begin testing I shall upload it to github for you to see (tonight probably)

10:40 < sreenik[m]> The only thing I dread is the possibility of the dimensions of the weights of the original and converted models not matching.

11:56 < jeffin143> lozhnikov : Thanks for pointing out styling issue, I have made required changes and pushed it

11:56 < jeffin143> Also next time I will take care of these small things :)

13:18 < Toshal> rcurtin: zoq: Is there a method to know which seed value is set by FixedRandomSeed() method?

13:58 < zoq> Toshal: FixedRandomSeed does internally use a random number as seed, so no; but FixedRandomSeed is only for the bindings, for the rest you could set a specific seed before running the test.

14:00 < zoq> I don't think there is a method to get the seed from the C++ pseudo-random generator either

14:00 < zoq> in which case you could reuse that function

15:12 < akhandait> sreenik[m]: I don't think that should be a big problem. I will be busy till tomorrow night. So, if you upload it, I will take a look then.

15:13 < akhandait> sreenik[m]: Good post on the blog btw!

15:23 < Toshal> zoq: Thanks for the info. Actually I was just getting familiar radical test which sometimes get failed. I earlier thought that it could be the issue. But yes later on realized that that it's not the issue. Sorry for the inconvenience.

15:38 < favre49> Not sure what to do about this error

15:38 < favre49> error: invalid types ‘double[size_t {aka long unsigned int}]’ for array subscript data[innovID][i] = genomeList[i].connectionGeneList[j].getWeight();

15:38 < favre49> https://pastebin.com/1GVu9cVT

15:38 < favre49> ^ Pastebin with the code in question

15:39 < favre49> I hav multiple places where using size_t for element access in armadillo matrices are throwing these errors

15:53 < Toshal> favre49: I think you should use data(innovID, i). Let me know if this is incorrect.

16:21 < akfluffy> I can't find an example of how CNNs would work across color channels; the only one I can find is grayscale. Do I just pass my image (256x256x3) as an input?

16:21 < akfluffy> I noticed a PR on loading images from disk but it's still open

16:31 < lozhnikov> jeffin143: Yes, I saw the commit. I'll try to look through it today.

16:52 < favre49> Toshal: Thanks, that fixed it :)

17:57 < favre49> zoq: I've finished coding NEAT, I'm going to start debugging it and finding issues by testing it on the XOR test soon.

18:12 < sreenik[m]> akfluffy: Thanks. Okay, sounds good

18:13 < sreenik[m]> Oops, I meant akhandait instead of akfluffy :)

18:38 < zoq> favre49: Great, I'll take a look at the code and leave some comments; excited to see the first results.

18:41 < zoq> akfluffy: Yes, you can just pass the images with > 1 channel.

18:52 < rcurtin> robertohueso: how are things going? did my writeup on the theory error make sense?

18:52 < rcurtin> or maybe I made an error in my own attempt at error correction? :)

19:27 < robertohueso> rcurtin: I have been working all thay on this, the thing is your derivation gives very good results (maybe it's because of what you said about KDE bounds being loose)

19:28 < robertohueso> I just wrote an answer to your comment on the branch because I can't clearly see how did you get to the last derivation

19:28 < robertohueso> but appart from that derivation the whole process makes sense and seems correct to me :)

19:45 < jenkins-mlpack2> Yippee, build fixed!

19:45 < jenkins-mlpack2> Project mlpack - git commit test build #186: FIXED in 45 min: http://ci.mlpack.org/job/mlpack%20-%20git%20commit%20test/186/

21:35 < rcurtin> KimSangYeon-DGU_: hey, I just met Sumedh and he said you had some really cool plots and results, but I don't see any blog post

21:36 < rcurtin> were you planning to make a post, or did you have any problems with the blog repo or anything?