02:33 < systematic> I'm new to mlpack. I want to make a program that thinks about any topic. Here is my general algorithm: 10--pick or accept topic. 20--Make statements relevant to topic. 30--Ask why about ethical questions. 40--statements-->questions. 50--questions-->open-ended questions. 60--Brainstorm several answers to open-ended questions. 70--prove/disprove answers. 80--estimate truth values for those answers neither proven nor disproven. 90--

02:34 < systematic> Would machine learning be able to learn those skills?

03:37 < rcurtin> systematic: I think that is a bit optimistic :)

03:37 < rcurtin> I am not sure current machine learning algorithms can handle the problem in the way that you want

03:54 < systematic> How do data scientists figure out how to make machine learning figure out whether they are looking at a pen. I guess it would be similar to that sort of machine learning. I want to be able to tell the program: This is a good argument. This is a bad argument. Tell me whether this other argument is a good argument.

03:57 < systematic> How do I get the ml to recognize a pattern?

04:07 < systematic> What kind of patterns has machine learning been able to find, and what formats does it accept? If I have to make tables, I'll make tables.

04:48 < systematic> Let's say I wanted to compare how far something was on the y axis with several other dimensions, which tool would I use to find those correlations?

13:21 < rcurtin> systematic: sorry for the slow response, I had to get some sleep :)

13:21 < rcurtin> typically machine learning algorithms will recognize patterns by the use of large sets of labeled training data

13:22 < rcurtin> so your typical data scientist might solve a problem by casting their problem as a classification problem for instance

13:22 < rcurtin> to take an example say you want to classify spam vs. clean emails

13:23 < rcurtin> obe way to do this will be to gather a dataset of millions of spam emails and millions of clean emails

13:23 < rcurtin> *one way

13:25 < rcurtin> then you would train some machine learning model using that data

13:26 < rcurtin> so part of the problem with your idea is that you would have to obtain a very large set of true and false statements, for the part where you want to infer truth

13:26 < rcurtin> and unfortunately that is a very hard thing to do

14:31 < Trion> systematic: seems like you want create a chatbot. I think this will get you an idea, when the person says something e.g. "Hello how are you?", tokenize using a nlp package, create a token vector of 1s and 0s and use a classification algorithm to map the [ tokenvector ] to [probability of replies]. Send back the reply with most probability

14:35 < rcurtin> yeah, I think this could be an approach that gets you close, but the part about ethics and truth will probably unfortunately not be feasible with the data that is available today

14:45 < Trion> :P Humans love bots that make mistakes! Must be because our brains be like "Haha! you can't achieve it in a day of training what I achieved in millions of years of evolution"

14:53 < rcurtin> hah

15:08 < rcurtin> zoq: you beat me to the krylov fix, I was just looking into it when I saw the commit emails :)

15:08 < rcurtin> unfortunately there's not time to re-run the matrix build; masterblaster will go down in four hours for the move

15:13 < zoq> rcurtin: oh, okay ... hopefully it is back by the end of the week?

15:16 < rcurtin> yeah, it should be online by friday... hopefully

15:16 < rcurtin> I will post updates as I hear them

15:16 < rcurtin> the guy in LA I am working with says he'll box it up immediately after I shut it down and send it off

15:16 < rcurtin> hopefully he'll give me a tracking number so I can hit 'refresh' in my browser over and over again :)

15:17 < zoq> sounds promising :)

16:02 < Trion> zoq: can I assume for the lines that come after a for loop with openMP pragma that all the loops have completed before it?

16:03 < rcurtin> Trion: yeah, the for loop itself will run in parallel, but it will all be synchronized after the loop ends

16:04 < Trion> Nice :-) makes code so easy!

18:56 < rcurtin> ok, goodbye masterblaster...

18:56 < rcurtin> let's hope the downtime isn't too long

20:49 < sumedhghaisas> zoq, rcurtin: Hey... Do we have baysian regression? just wanted to run couple of experiments... I did not see that as a module though could part of some module

20:53 < rcurtin> sumedhghaisas: no, no bayesian regression... but feel free to implement as always :)

20:53 < rcurtin> masterblaster UPS tracking number: 1Z6R981Y0191859151

21:16 < zoq> rcurtin: 25.90 lbs, I thought it's heavier.

21:17 < rcurtin> yeah that has to be an error, that thing is a 2U rackmounted system

21:17 < rcurtin> 25.9kg maybe I could believe but even that seems light