verne.freenode.net 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/
travis-ci has joined #mlpack
< travis-ci>
mlpack/mlpack#1138 (master - bdbc2fe : Ryan Curtin): The build was broken.
< Karl_>
zoq: yes, I played around a bit with it, but when I use the complete data set (50M x 6), I run out of memory. I retooled the naive method to mimic prcomp from R (although I'm not 100% I got it right). I didn't get the quic svd to work, but as I only have a few samples, the arma::svd function works ok.
< mentekid>
rcurtin: Everthing I've done on parallelizing ReturnIndicesFromTables is a waste...
< mentekid>
the result is most of the time slower than the sequential version
< mentekid>
do you think I should remove that and simply leave the parallel queries code?
< rcurtin>
mentekid: is the code you've written available in the PR you opened?
< rcurtin>
if so I will take a look at it and see if I have any comments to add
Mathnerd314 has joined #mlpack
< mentekid>
rcurtin: sorry, just saw this (stupid tray doesn't blink some times so I miss messages)
< mentekid>
let me see. I think I didn't push the per-query parallelizations
< mentekid>
the code for query parallelization is online though
< marcosirc>
zoq: Very well thanks! Having fun with the implementation of spill trees :)
< marcosirc>
Ok. I was wondering why it didn't output anything to the console.
< marcosirc>
after make run
< zoq>
Sounds like fun :) if you use LOG=True the output is disabled, I think it would be interesting to see some progress ... I don't know when the benchmark will be finished
< rcurtin>
zoq: are you running the full benchmark? in the past that has usually taken something like two weeks...
< zoq>
just the ALLKNN benchmarks
< zoq>
just checked: 'METHODBLOCK=ALLKNN run'
< rcurtin>
yeah, I might expect that to take 2 or 3 days maximum, but I am not sure exactly how many benchmarks and datasets are being run
< zoq>
I think for the next run we could decrease the number of iterations and remove some data sets that timed out.
< marcosirc>
zoq: but when I run in my computer I can see the "Info" log messages
< marcosirc>
like : "[INFO ] Method: ALLKNN"
< zoq>
hm, maybe when the complete benchmark block finished?
< marcosirc>
after each dataset is completed... Maybe Jenkins buffers the output until it finishes...
< zoq>
hm, okay, I thought LOG=True would suppress any output, maybe you're right and jenkins buffers the output.
< rcurtin>
my comments are not very informed here and it has been a long time, but I seem to remember monitoring the Jenkins jobs by watching the output in the console log
< rcurtin>
and I think LOG=True was set in that case, but like I said, I am not certain
< zoq>
maybe UPDATE=True does the trick, let's see
< zoq>
no, same result
< zoq>
just checked htop, it's still running and the database get's updated
< marcosirc>
zoq: Ok! I will wait. Thanks!
< zoq>
hopefully it will finish in the next hours
K4k has quit [Ping timeout: 252 seconds]
K4k has joined #mlpack
Mathnerd314 has quit [Quit: No Ping reply in 180 seconds.]