naywhayare changed the topic of #mlpack to: http://www.mlpack.org/ -- We don't respond instantly... but we will respond. Give it a few minutes. Or hours. -- Channel logs: http://www.mlpack.org/irc/
< jenkins-mlpack>
* michaelfox99: Hierarchical model support
< jenkins-mlpack>
* michaelfox99: Changes to work with new, hierarchical GMMsHierarchical GMMs store params in GaussianDistributions. Makes code clearer and simplifies Save/Load.
< jenkins-mlpack>
* michaelfox99: Hierarchical GMMs store params in GaussianDistributions. Makes code clearer and simplifies Save/Load.
< sumedhghaisas>
naywhayare: hey you there??
< naywhayare>
sumedhghaisas: yeah, I am here
< naywhayare>
marcus_zoq: I'd say just push the changes; Udit will probably say something if there's an issue
< sumedhghaisas>
naywhayare: I want to create a sparse matrix with diagmat(vec)... how to do that??
< naywhayare>
I'm not sure it's possible
< naywhayare>
actually, hang on...
< naywhayare>
I was going to say that diagmat(vec) returns a mat and we can't use operator overloading
< naywhayare>
but actually diagmat(vec) returns a Proxy object
< naywhayare>
(oops, I said operator overloading and I meant return type overloading... sorry)
< marcus_zoq>
naywhayare: Okay, I guess I should wait until michael fixed the build.
< naywhayare>
marcus_zoq: hah, yeah. lots of commits...
< naywhayare>
sumedhghaisas: but I'm not sure exactly what changes would need to be made to Armadillo to do that
< naywhayare>
so I'd just manually fill the diagonal of the sparse matrix...
sumedhghaisas has quit [Ping timeout: 264 seconds]
< marcus_zoq>
naywhayare: The benchmark job failed, looks like a jenkins problem; the script is still running, so maybe everything is fine, I need to check the results ...
< naywhayare>
FATAL: Unable to delete script file /tmp/hudson4701634670281430404.sh
< naywhayare>
but /tmp isn't full, and it isn't mounted via NFS or anything
< jenkins-mlpack>
Starting build #2072 for job mlpack - svn checkin test (previous build: STILL FAILING -- last SUCCESS #2069 14 hr ago)
< jenkins-mlpack>
* michaelfox99: Hierarchical model support
< jenkins-mlpack>
* michaelfox99: Hierarchical model support
< jenkins-mlpack>
* michaelfox99: Hierarchical model support
< jenkins-mlpack>
* michaelfox99: Hierarchical model support
< naywhayare>
oh! I bet I know what happened
< naywhayare>
I bet the kerberos ticket expired (after about a week) and so the jenkins user lost write permission since they were no longer authenticated
< naywhayare>
so what I'll need to do is configure the jenkins job on shoeshine to load a keytab (which won't expire)
< naywhayare>
I'm headed to lab now but I'll do it when I get there
< jenkins-mlpack>
Starting build #2073 for job mlpack - svn checkin test (previous build: STILL FAILING -- last SUCCESS #2069 14 hr ago)
< marcus_zoq>
naywhayare: Okay, do you think, jenkins isn't able to write the benchmark.db database file? If this is the case we can stop the script and restart before the failure point.
< naywhayare>
marcus_zoq: yeah, at the point at which the build failed, jenkins became no longer able to write the benchmark.db file
< naywhayare>
so that particular file probably has good results from the first week
< marcus_zoq>
naywhayare: Okay, I will look into the benchmark.db file and restart the job as soon as you fixed the authentication timeout?
< naywhayare>
marcus_zoq: ok, I modified all the benchmark jobs so they manually authenticate with kerberos with a non-expiring ticket (...I think), so whenever you are ready to restart it, it should be ready
< naywhayare>
yeah, that sounds good. it should be ready to go when you are ready to restart it
< marcus_zoq>
I will restart the job tonight.
< naywhayare>
ok, great; thanks!
jbc__ has joined #mlpack
< jbc__>
hey what’s the test_file format for the classifiers? Just a csv? tabbed delimited?
< jenkins-mlpack>
Starting build #2074 for job mlpack - svn checkin test (previous build: STILL FAILING -- last SUCCESS #2069 15 hr ago)
< naywhayare>
jbc__: yeah, csv will work
< naywhayare>
you can use multiple formats... it will autodetect them based on their extension
< naywhayare>
and I now realize I didn't write a little tutorial for that
< jbc__>
it will it cast for the least memory? So if I have a large set, but I could use short instead of int?
< jbc__>
If you’d like I could write up my experiences and then throw it your way?
< naywhayare>
andrewmw94: while you're in Minnesota, if you somehow have a chance to get to here, http://goo.gl/maps/z9mqu, it's possible to find a waterfall called Devil's Den
< naywhayare>
it qualifies as the coolest place I have ever been
< naywhayare>
(since you mentioned Minnesota I've been digging for exactly what the place is called and where it is)
< andrewmw94>
ah cool
< andrewmw94>
let me take a look at where it is
< naywhayare>
just south of Mankato
< andrewmw94>
drives in MN can be pretty long
< naywhayare>
yeah... it was about an hour from Chaska
< naywhayare>
I don't think the pictures in that blog do it justice, but you can't just drive to it either... you need a canoe or kayak or something
< andrewmw94>
yeah. That looks awesome but I think it would be at least 4 hours each way.
< andrewmw94>
Maybe on some future trip. My Mom's family lives in MN so we travel there fairly often.
< naywhayare>
:)
< naywhayare>
I would like to go back to that waterfall at some point... but it's a bit more than a four hour drive for me :)
< andrewmw94>
I went to a fall in NY that claimed to be the tallest. It was pretty lame. A bunch of short drops which, added together, were the tallest, but each individual was only a few meters. Still, it was more secluded than Niagra, so it was nice.
< andrewmw94>
I'm glad to see that most sites don't count them as the tallest unless they are more or less straight drops
< jenkins-mlpack>
Ryan Curtin: Minor formatting changes for row_col_iterators, mostly to adhere to the
< jenkins-mlpack>
Whitesmiths style.
< jenkins-mlpack>
Starting build #2076 for job mlpack - svn checkin test (previous build: SUCCESS)
sumedhghaisas has joined #mlpack
sumedhghaisas has quit [Client Quit]
sumedhghaisas has joined #mlpack
< sumedhghaisas>
naywhayare: hey sorry... I didn't get your msges ... net was down... did you send any??
< naywhayare>
13:32 < naywhayare> sumedhghaisas: but I'm not sure exactly what changes would need to be made to Armadillo to do that
< naywhayare>
13:33 < naywhayare> so I'd just manually fill the diagonal of the sparse matrix...
< sumedhghaisas>
ohh okay... I can try if its not too difficult... :)
< sumedhghaisas>
naywhayare: we need to make a copy assignment overload for arma::Op<arma::Col<double>, arma::op_diagmat>...
< sumedhghaisas>
any idea how to do that??
< naywhayare>
the problem is that SpMat does not unwrap Op classes; it unwraps SpOp classes
< naywhayare>
I think it would be easiest here to just write a loop that fills each element in the diagonal
< sumedhghaisas>
yeah... I was thinking about some method to unwrap Op classes...
< sumedhghaisas>
I will just do the loop thing for now..
< naywhayare>
personally I kind of think it would be better if Op worked with both sparse and dense matrices
< naywhayare>
and then it could unwrap into whatever the user specified... i.e.
< naywhayare>
arma::sp_mat x = a + b + c + d (for a,b,c,d of type sp_mat) would unwrap into type sp_mat, but arma::mat x = a + b + c + d (again with a,b,c,d of type sp_mat) would unwrap into type mat
< naywhayare>
but I'm not yet completely clear on how this can be done (I haven't put much time into it), and the sparse matrices still need some of the factorizations and other functionality before that becomes a real possibility
< jenkins-mlpack>
* andrewmw94: Point deletion bug fix.
< jenkins-mlpack>
* Ryan Curtin: Test sparse iterators, when we can.
sumedhghaisas has quit [Ping timeout: 245 seconds]
sumedhghaisas has joined #mlpack
< sumedhghaisas>
naywhayare: ohh I should now implement algorithms with row_col_iterator right??
< sumedhghaisas>
did you send the patch to conrad??
< naywhayare>
yeah, I CC'ed you on the email
< sumedhghaisas>
yes... got it...
< sumedhghaisas>
creating sp_mat from location and values will be faster than creating the matrix and then adding single points right??
< naywhayare>
yes
randomguy has joined #mlpack
< sumedhghaisas>
naywhayare: is it possible to convert sp_mat::const_row_iterator to sp_mat::const_row_col_iterator(a.k.a sp_mat::const_iterator)...
< sumedhghaisas>
its not implemented right now... but this will speed things up... its possible for mat iterators...
randomguy has quit [Quit: Page closed]
< naywhayare>
sumedhghaisas: I suppose we could add that support, but what situation are you encountering this problem in?
< sumedhghaisas>
I want to traverse a row... but I have realized that I have made a stupid mistake... I could just do it with const_row_iterator... sorry :)
< sumedhghaisas>
naywhayare: and what to be done with plain_svd tests??
< sumedhghaisas>
and this line... arma::sign(arma::trans(V.row(i))) ... this is valid right??
< sumedhghaisas>
I think because V is sp_mat its creating problems...
< sumedhghaisas>
sign function does not exist for sp_mat :(
sumedhghaisas has quit [Ping timeout: 264 seconds]