13:24:18 From Sofus Kjærsgaard Stray : I don't know if it's dirt on the glas or something inside it 13:24:29 From Ann-Sofie Priergaard Zinck : Is it supposed to be a video? 13:24:46 From Runi : Nothing is moving 13:24:53 From Viki Lavro : I saw something moving 13:25:02 From Runi : Ah yes 13:25:02 From Viki Lavro : on the top right secion 13:25:03 From Runi : I see it 13:25:13 From Ann-Sofie Priergaard Zinck : Ok I see it now 13:25:23 From Yane García : yeah tiny spots on the left 13:27:06 From Marta Mrozowska : increase contrast? 13:27:19 From Bagne : difference image 13:27:37 From Bagne : yes 13:27:45 From Mikkel Rasmus Schmidt : Subtract the first frame 13:31:57 From Yane García : how do you differenciate bubbles from particles of interests 13:35:06 From Haider Fadil Ali Hussein Al-Saadi : they have different behaviours and movement? 13:35:34 From Andy : Different size of the inputs 13:36:35 From Haider Fadil Ali Hussein Al-Saadi : use radius 13:36:36 From Haider Fadil Ali Hussein Al-Saadi : ? 13:36:43 From Emil Thyge Skaaning Kjaer : PCA? 13:37:29 From Yane García : size distribution? 13:37:34 From Andy : Smart 13:40:53 From Yane García : what is the diff between the 2 rows 13:41:03 From Sofus Kjærsgaard Stray : 0 and 10 days I think 13:41:32 From Andy : What if you suddenly get data from a larger bottle (33/50 cl)? Can you then compare bottles? 13:44:17 From Rasmus Salmon : Partially integrate the curves in some way? 13:46:12 From Yane García : to get sort of % of … particles 13:46:13 From Yane García : ? 13:46:15 From Yane García : oh 13:49:16 From Haider Fadil Ali Hussein Al-Saadi : we get an image and perform regression on the image? 13:49:44 From Haider Fadil Ali Hussein Al-Saadi : but how do you make an histogram 13:49:47 From Haider Fadil Ali Hussein Al-Saadi : without performing 13:49:50 From Julius Terp : But do you then only look at size, and discard the idea of a difference between particles and strings? I guess some would argue that either strings or particles is worse to have in your beer? 13:49:52 From Haider Fadil Ali Hussein Al-Saadi : classification 13:50:06 From Emil Martiny : I am confused on why this is a machine learning problem at all, isn't this histogram the result we want, and then we just want to have cutoff on the histogram 13:51:11 From Emil Martiny : I just wanna take a weighted mean of the histogram 13:56:22 From Yane García : for me the first picture makes sence because of the distribution 13:56:31 From Yane García : s 13:59:52 From Andy : How would these plots look like, if you used a weighted mean of the histograms? Would they be the same if the error is due to the labels?? 14:05:12 From Emil Martiny : it was only like 3 minutes ago, i figured that PIA was a person and not a corny name for an algorithm 14:05:21 From Emil Martiny : /machine 14:05:41 From Sofus Kjærsgaard Stray : Yeah I thought pia was an algortihm as well 14:06:04 From Emil Martiny : I got it when you said beer-taster 14:10:34 From Yane García : cool 14:11:46 From Emil Thyge Skaaning Kjaer : "tasted" 14:12:04 From Yane García : is just one person tasting? 14:12:24 From Andy : I would be willing to help her :) 14:12:33 From Haider Fadil Ali Hussein Al-Saadi : https://www.youtube.com/watch?v=SLP9mbCuhJc 14:12:38 From Yane García : haha 14:44:30 From Rasmus Ørsøe : I have a technical question about pytorch specific to the project 14:46:34 From Troels Christian Petersen : Yes? 14:47:01 From Rasmus Ørsøe : Let's say I have three nodes with n features each in some configuration 14:47:13 From Troels Christian Petersen : yes... 14:47:32 From Rasmus Ørsøe : I'd like that algorithm to take this as an input and spit out 5 numbers (as close to the targets as possible) 14:48:12 From Troels Christian Petersen : Yes…. 14:48:13 From Rasmus Ørsøe : If I pass the graph through a convolutional layer in torch, the output is on the form [n_nodes, n_features] 14:48:42 From Rasmus Ørsøe : To reduce the output to the form [1, features] I've used pooling in pytorch 14:48:45 From Rasmus Ørsøe : Is this the 14:48:48 From Rasmus Ørsøe : 'right' way? 14:49:12 From Rasmus Ørsøe : Does the question make sense? 14:50:36 From Rasmus Ørsøe : Yeah this is the issue 14:50:41 From Rasmus Ørsøe : Sorry 14:50:41 From Troels Christian Petersen : OK… 14:51:02 From Rasmus Ørsøe : If we boil it down to the essence 14:51:30 From Rasmus Ørsøe : I guess my question is just what layers are used to boil an input down to match the correct values? 14:51:42 From Rasmus Ørsøe : is it pooling layers? 14:51:48 From Rasmus Ørsøe : Or am I misusing the library? 14:52:56 From Rasmus Ørsøe : Yes. 14:53:29 From Aske R. : shouldn't you just tell it to make an output of 1 value, atleast for a regression problem? 14:53:33 From Rasmus Ørsøe : Yes. 14:55:17 From Mikkel Rasmus Schmidt : Can you say that again? 14:56:10 From Rasmus Ørsøe : That's pooling? 14:56:26 From Aske R. : we can recognize that from the example code that we are trying to run 14:56:39 From Rasmus Ørsøe : Thanks. 14:58:56 From Rasmus Ørsøe : It does. Thank you. 14:59:00 From Troels Christian Petersen : Cool... 15:00:53 From Emil Martiny : facebook video messages 15:01:00 From Emil Martiny : google hangouts