This is a tricky one. I see where the Path in question (the one
mentioned at the start of the exception) is being built: lines 142-144
in the VectorMatrixMultiplicationJob I wrote, at the tail end of the
main job method itself, is where the temporary Path is assigned for the
returned DRM.
But from there I'm still scratching my head, in particular as to how the
"data" folder is appended to the end. I follow it into DRM's times()
method, but I'm not entirely sure why it's crashing there, particularly
since that directory is DRM's temp directory.
My only lead is line 158 in TimesSquaredJob.java, where that full Path
from before appears to be prepended to an *input* vector path:
Path inputVectorPath = new Path(outputVectorPathBase, INPUT_VECTOR + '/'
+ now);
That doesn't make a whole lot of sense to me. But neither does it seem
to explain the behavior we're seeing.
I'll take another look at it again tomorrow. Sorry for the delay =/
On 5/26/11 11:06 AM, Jeff Eastman wrote:
Naturally :). At least you can run this in the debugger. My guess is there's
been some changes in DRM file handling that have not been handled by the
SpectralK driver. Likely a simple path fix.
-----Original Message-----
From: Shannon Quinn [mailto:[email protected]] On Behalf Of Shannon Quinn
Sent: Thursday, May 26, 2011 8:27 AM
To: [email protected]
Subject: [SPAM] Re: [VOTE] Release Mahout 0.5, take 2
Arg, of course this crops up when my family vacation begins. I will take a look
at it later today; it just may be pretty late in the evening.
On 5/26/11 10:21 AM, Jeff Eastman wrote:
Well, the bug is *surfaced* by an example calling the SpectralKMeansDriver,
which is in Mahout core, using arguments which have worked in the past. I'd
like to get Shannon's reaction before I decide if this is a show stopper or
not. I did mark it as fix in 0.6, and I doubt we have a lot of users yet with
the spectral clustering, so I'm on the fence.
Shannon?
-----Original Message-----
From: Grant Ingersoll [mailto:[email protected]]
Sent: Thursday, May 26, 2011 8:05 AM
To: [email protected]
Subject: Re: [VOTE] Release Mahout 0.5, take 2
I'm hoping to take a look tonight, but agree the bug isn't a show stopper. +0
as of now, hopefully a +1 by the end of the day.
On May 26, 2011, at 7:27 AM, Benson Margulies wrote:
Me too. +1
On Thu, May 26, 2011 at 7:18 AM, Sean Owen<[email protected]> wrote:
(FWIW I vote +1 for the release in spite of the bug in an example
found by
Jeff.)
On Mon, May 23, 2011 at 8:03 PM, Sean Owen<[email protected]> wrote:
https://repository.apache.org/content/repositories/orgapachemahout-
001/org/apache/mahout/
Enjoy!
