vowpal_wabbit at Yahoo! Groups

vowpal wabbit question

Jin Huang — Wed, 22 May 2013 07:13:14 GMT

Now, i test VW using two method First Method :just start VW to train a model ,after that ,using VW continue to train the second model, and the CPU usage is

????????? ?????????????? ???? ???? ?

balyyieva181 — Wed, 22 May 2013 03:24:37 GMT

?? ????? ????? ??????? ???-???? ?? ????? ??? ???????????? http://goo.gl/d4IX4?/ISIeWR ?????

Re: need your help!

John Langford — Tue, 21 May 2013 13:16:41 GMT

You could be suffering from an io bottleneck. ... You could be suffering from an io bottleneck. On Tue, May 21, 2013 at 5:01 AM, Jin Huang < hj51yc@... >

need your help!

Jin Huang — Tue, 21 May 2013 09:01:01 GMT

hi, recently i use the VW for training my dataset , i want to train about one hundred models,then i use multi-thread to call the vw.exe ,and i just start 20

Re: Cross validation

corelgallery — Tue, 21 May 2013 00:56:50 GMT

John, Thanks for your quick reply. A few follow-ups below. A) I am struggling with how to best accomplish this. The following do not seem to work: vw -d

Re: Cross validation

John Langford — Mon, 20 May 2013 23:46:25 GMT

A) Yes, that should be no problem. B) Effectively yes at present. We have discussed a hack (creating a command example that allows you to control the learning

Re: Reusing cache with different outcomes

John Langford — Mon, 20 May 2013 23:41:33 GMT

There is no method at present to reuse the cache in this case. -John

Reusing cache with different outcomes

corelgallery — Mon, 20 May 2013 21:27:38 GMT

I would often like to reuse the same feature vectors with different outcome data. Is there some way to keep these separate? Or must I regenerate the entire

Cross validation

corelgallery — Mon, 20 May 2013 21:25:06 GMT

I would like to do some basic automated cross-validation of my results. My current approach seems to constrained by a couple of aspects that make me think I

????????? ????????????

???? — Mon, 20 May 2013 20:19:23 GMT

????????????????? ??????? http://goo.gl/JSKKR?/sBbvYRev y?

Re: weights_per_problem

John Langford — Mon, 20 May 2013 14:30:25 GMT

I'm almost sure this is a combination of line 261: https://github.com/JohnLangford/vowpal_wabbit/blame/master/vowpalwabbit/oaa.cc and line 126 here:

Re: weights_per_problem

Paul Mineiro — Mon, 20 May 2013 13:50:30 GMT

Convincing but perplexing, the only change is the alteration of weights_per_problem in parse_args.cc (

????? ???????

?????? — Mon, 20 May 2013 08:25:57 GMT

????? ? ?????????: http://goo.gl/Sps2G?/kEuCWfF . ???????????? ??????? ?????? ??????? ????????

Re: weights_per_problem

John Langford — Mon, 20 May 2013 06:11:12 GMT

That's quite convincing. -John

Re: weights_per_problem

Paul Mineiro — Sun, 19 May 2013 23:08:34 GMT

The demo results I am discussing are pure OAA, there is no nn involved. Ergo I think a bug has been introduced. p Skickat från min iPhone