08:54 < rcurtin> sgupta: I am still waiting to hear from you so that I can help out...

08:56 < sgupta> rcurtin: yes, I'll mail it to you in some time

09:10 < rcurtin> sgupta: if it is what you are currently held up on, why the delay?

09:19 < sgupta> rcurtin: sent you an email.

09:26 < rcurtin> sumedhghaisas_: please don't forget your status update this week, I didn't see one last week... it is always nice to read about how things have been going :)

16:31 < rcurtin> amusing tweet I found: https://twitter.com/arialdomartini/status/745476407531167744

16:34 < zoq> ... is this true, or fake news :)

16:35 < rcurtin> I think it is true... but I guess I did not verify it

16:35 < rcurtin> I am sitting in a keynote talk right now and the speaker presented the slide, so it must be true I guess :)

16:36 < zoq> sitting in a keynote and merging PR's :)

16:36 < rcurtin> yeah I guess so, there is too much work to do I guess :(

16:38 < rcurtin> the primary topic of the keynote hits a lot of themes I have been hearing in the machine learning world for some number of years---reproducibility in experiments, availability of datasets, sharing of code, etc.

16:39 < rcurtin> unfortunately it seems most of the talks on this type of topic present the problem above and then say something like "we can solve it by using this system that we have built or this initiative we have created"

16:39 < rcurtin> but... it is hard to actually get any sort of coherent consensus on which system or initiative to go with, so I think a lot of these efforts end up not capturing a large part of the community

16:45 < zoq> I see it's difficult to find something everyone agrees on, I guess not even the big player are able to agree on something.

16:46 < rcurtin> yeah, I wonder if there will be a solution eventually, but at the very least in the machine learning community the situation for reproducibility is a lot better---people usually put their code on github

16:49 < zoq> hm, but there are still a lot of people writing something really neat without providing the code or the dataset.

16:50 < rcurtin> that's true, there is definitely still a ways to go :)

18:01 < dfne> hi

18:03 < dfne> i am trying to use the gmm training on a dataset, however I keep getting ' chol(): decomposition failed ' error

18:03 < dfne> I would be happy if anybody can help me to solve it

18:04 < dfne> I tried to add some small noise to the diagonal, but it did not help

18:23 < dfne> The data I use is in this link https://drive.google.com/open?id=0B6MxzUFY5BNHbWZvSWlOWTV2M2c

18:23 < dfne> columns are features

18:24 < dfne> rows represent each data point

20:32 < zoq> sumedhghaisas_: Hello, would be great if you could write the weekly report for the last week, it's always interesting to read what everyone has done over the week :)

22:57 < sumedhghaisas_> zoq: Hey Marcus... sorry for that. Yeah I will do that and this week's report tonight itself. :)