00:06 < manish7294> And since it can also be G -> G -> G -> E routine, so the transformation matrix for gradient will be different than evaluate. So, I think we may be forced to recalculate impostors for Gradient() function as well.

00:09 < manish7294> zoq: Do you think except BBSGD there can be some other optimizer that uses this non-continous Evaluate(), Gradient() routine?

01:53 < rcurtin> manish7294: yes, I mean that we will not be able to avoid the impostor recalculation in Gradient()

01:53 < rcurtin> however, we can still use the same pruning syrategies

01:53 < rcurtin> strategies*

01:54 < rcurtin> we just cannot assume that Gradient() will be called after Evaluate()

02:03 < manish7294> rcurtin: So, then we will require two sets of cache matrices, right?

02:22 < rcurtin> no, we should be able to do it with only one

02:24 < manish7294> But gradient and evaluate will have to independently track their previous values

02:25 < manish7294> Or, a way could be to just have a flag to reset cache between different gradient and evaluate calls

02:27 < manish7294> Hmm, I think you'require right, a single one will suffice. As we will basically be doing same thing in both the functions.

02:28 < manish7294> Ah! Autocorrect changed you are to you'require :)

04:14 < rcurtin> right, and the bounds may get looser more quickly but it is okay

09:27 < travis-ci> manish7294/mlpack#64 (evalBounds - 7025a8b : Manish): The build has errored.

09:27 < travis-ci> Change view : https://github.com/manish7294/mlpack/compare/6df99cb70dda...7025a8bc3d34

09:27 < travis-ci> Build details : https://travis-ci.com/manish7294/mlpack/builds/79000296

14:52 < travis-ci> mlpack/mlpack#5292 (master - 884523d : Ryan Curtin): The build passed.

14:52 < travis-ci> Change view : https://github.com/mlpack/mlpack/compare/4c19146cb7b0...884523db6103

14:52 < travis-ci> Build details : https://travis-ci.org/mlpack/mlpack/builds/403905920

20:38 < rcurtin> ok, I believe that Symantec will not be willing to host the systems moving forward

20:39 < rcurtin> I think there was a misunderstanding I am trying to resolve through email, but my outlook at this point is "doubtful"

20:39 < rcurtin> so when I get home next week I will set up some more systems we can use

20:40 < zoq> rcurtin: okay, I'll get the two systems ready as well.

20:42 < rcurtin> I would imagine we have at least a week or more before the systems are shut down (if they don't change their mind) but they may go down as soon as Monday

20:42 < rcurtin> ShikharJ: manish7294: sorry about this, I know you are actively using these systems. we will figure out a solution :)

20:44 < ShikharJ> rcurtin: It's fine really. No need atleat for the coming few days :)

20:45 < zoq> rcurtin: If I remember right, they both use a benchmark system?

20:46 < ShikharJ> zoq: I use Savannah, but I'm not sure if anyone else uses that as well. I saw no other person's activity on that atleast for the time I was online.

20:47 < rcurtin> right, I think manish has been using slake

20:47 < rcurtin> I have two benchmarking systems from "the old days" at my house; one currently has Fedora (it was samedi.cc.gt.atl.ga.us long ago)

20:48 < rcurtin> and the other has Windows

20:48 < rcurtin> I will see if I can find a volunteer to turn the Fedora box on and get it online :)

20:48 < rcurtin> definitely less powerful than savannah or slake but still better than nothing

20:49 < zoq> If they go down on Monday, I can provide access to the first machine probably the next day (i7-3770 + 16GB).

20:50 < zoq> I'll have to setup the other one I guess it's an i7-6700 + 64GB.

20:52 < rcurtin> ha that is far nicer than those two systems I was talking about :)

20:53 < rcurtin> I think those are 2011-era i5s with 16GB RAM

20:53 < zoq> I think for benchmarking it would be nice to have two systems with the same specs.

20:54 < ShikharJ> rcurtin: I think I'll have more computing power in my laptop :P

20:55 < ShikharJ> But i7-6700 + 64GB is really great :)

20:56 < zoq> ShikharJ: Unfortunately it doesn't come with a GPU :(

20:59 < zoq> and not as powerfull as masterblaster with 256GB and two E5-2699 ...

21:04 < rcurtin> yeah, access to GPUs will be a little more difficult

21:04 < rcurtin> I'll see what I can dig up

21:04 < rcurtin> flight is going to take off shortly, back later...

21:04 < zoq> have fun :)

21:05 < rcurtin> as long as the air conditioning kicks in, I will :)

21:06 < zoq> I guess it's not going to be hot in Iceland.