ChanServ changed the topic of #mlpack to: "mlpack: a fast, flexible machine learning library :: We don't always respond instantly, but we will respond; please be patient :: Logs at
travis-ci has joined #mlpack
< travis-ci> saksham189/mlpack#19 (br - 5265741 : Saksham Bansal): The build failed.
< travis-ci> Change view :^...526574187506
travis-ci has left #mlpack []
< jenkins-mlpack2> Project docker mlpack nightly build build #172: STILL UNSTABLE in 3 hr 25 min:
ImQ009 has joined #mlpack
witness has joined #mlpack
ImQ009 has quit [Read error: Connection reset by peer]
Saksham_ has joined #mlpack
Saksham_ has quit [Ping timeout: 256 seconds]
mackie has joined #mlpack
< mackie> hello
< zoq> mackie: Hello there!
< mackie> i'm trying to understand the range selection and point selection strategy for dbscan
< mackie> I've been combing through the APIs but I can't seem to get a list of options to try
< mackie> I have a set of toy spatial 2D data points, and even though visually it looks like the answer should be obvious, it seems like the algorithms creates a different set of clusters
< mackie> I've checked here -, as well as the dbscan.hpp file
< mackie> It's not obvious to me what options are available
< zoq> mackie: Hmm, I just checked the src and to me it looks like the PointSelectionPolicy/pointSelector isn't used.
< zoq> I guess Ryan can probably provide some insight once he reads the message.
< mackie> i see thanks
< jenkins-mlpack2> Yippee, build fixed!
< jenkins-mlpack2> Project mlpack - git commit test build #72: FIXED in 34 min:
< jenkins-mlpack2> noreply: Merge pull request #1622 from rcurtin/test-fixes
mackie has quit [Ping timeout: 256 seconds]
< rcurtin> mackie: zoq: you're right, the PointSelectionPolicy isn't used, that is strange
< rcurtin> (admittedly there was only one PointSelectionPolicy implemented...)
< rcurtin> the meaning of PointSelectionPolicy was that instead of looping linearly over the points, it would choose the next point for DBSCAN randomly
travis-ci has joined #mlpack
< travis-ci> mlpack/mlpack#5696 (master - 6410e20 : Ryan Curtin): The build has errored.
travis-ci has left #mlpack []
< rcurtin> to change that back to how it should be, dbscan_impl.hpp at lines 193-198 should be changed
< rcurtin> basically, currently what's there is a for loop over all the points, which goes through them in order
< rcurtin> but actually it should be using something like pointSelector.Select() to get the next index
< rcurtin> let me open an issue about this...
< rcurtin> for RangeSearchType, that specifies the class that is used for the range search, but it will not affect the results
< rcurtin> even changing the PointSelectionPolicy (if it worked correctly) would not be too likely to have too huge an effect on the results
< rcurtin> basically epsilon and minPoints are the biggest parameters for DBSCAN
< rcurtin> you could also try MeanShift and see if that works...
< rcurtin> zoq: if you are happy with Dan's changes in ensmallen#60 and my changes in ensmallen#61 feel free to merge, then I can start the release process for ensmallen 1.12.0
< rcurtin> no hurry of course :)
< rcurtin> oops, I meant ensmallen#64 not ensmallen#5161
< rcurtin> strange typo...
< zoq> rcurtin: Everything looks good to me.
< rcurtin> perfect, 1.12.0 is released then :)
< zoq> I like the name
< rcurtin> :)
travis-ci has joined #mlpack
< travis-ci> mlpack/ensmallen#67 (ensmallen-1.12.0 - 8e236b2 : Ryan Curtin): The build passed.
< travis-ci> Change view :^...8e236b243451
travis-ci has left #mlpack []
ImQ009 has joined #mlpack
ImQ009 has quit [Quit: Leaving]