Perceptron Passive Aggressive training

[Jörn Kottmann <ko...@gmail.com>]

Hi,

should we additionally implement passive aggressive training
for our perceptron and see how it then performs?
Benson mentioned that he is using it, so I guess it works great.

Any opinions?

If we think its a great idea it would be nice to create a jira for it.

Thanks,
Jörn

Re: Perceptron Passive Aggressive training

[Jason Baldridge <ja...@gmail.com>]

+1

This looks like a useful paper in this regard:

Shin Matsushima*, *Nobuyuki Shimizu*, *Kazuhiro Yoshida*, *Takashi Ninomiya,
Hiroshi Nakagawa: Exact Passive-Aggressive Algorithm for Multiclass
Classification Using Support Class.* the 2010 SIAM International Conference
on Data Mining* (SDM'2010).

http://www.google.com/url?sa=t&source=web&cd=5&ved=0CDsQFjAE&url=http%3A%2F%2Fwww.siam.org%2Fproceedings%2Fdatamining%2F2010%2Fdm10_027_matsushimas.pdf&rct=j&q=passive-aggressive%20algorithms&ei=PIfvTd_xB8HY0QGzn6j2DA&usg=AFQjCNGlkEbTh_Py4am0G49JCLgYgBpJCA&cad=rja

On Wed, Jun 8, 2011 at 9:06 AM, Jörn Kottmann <ko...@gmail.com> wrote:

> On 6/8/11 3:52 PM, Benson Margulies wrote:
>
>> As with all PA, it depends on your data. It helps more with less data,
>> as it were. But we love it.
>>
>> In the unlikely event that you have any problem reading the paper on
>> how to apply it to the sequence problem, let me know. I'm still not
>> organized to liberate our code, but I'm happy to extract explanations
>> from it.
>>
>
> I found one paper about it on google, which one do you mean?
>
> That is the paper I found, but still have to study it:
> Crammer, Koby. Dekel, Ofer. Keshet, Joseph. Shalev-Shwartz,
> Shai. Singer, Yoram. " Online Passive-Aggressive Algorithms ". Jerusalem,
> 2006
>
> Jörn
>



-- 
Jason Baldridge
Assistant Professor, Department of Linguistics
The University of Texas at Austin
http://www.jasonbaldridge.com
http://twitter.com/jasonbaldridge

Re: Perceptron Passive Aggressive training

[Benson Margulies <bi...@gmail.com>]

this is the paper. We had a couple of phone calls with Koby before we
sorted it out altogether.


On Wed, Jun 8, 2011 at 10:06 AM, Jörn Kottmann <ko...@gmail.com> wrote:
> On 6/8/11 3:52 PM, Benson Margulies wrote:
>>
>> As with all PA, it depends on your data. It helps more with less data,
>> as it were. But we love it.
>>
>> In the unlikely event that you have any problem reading the paper on
>> how to apply it to the sequence problem, let me know. I'm still not
>> organized to liberate our code, but I'm happy to extract explanations
>> from it.
>
> I found one paper about it on google, which one do you mean?
>
> That is the paper I found, but still have to study it:
> Crammer, Koby. Dekel, Ofer. Keshet, Joseph. Shalev-Shwartz,
> Shai. Singer, Yoram. " Online Passive-Aggressive Algorithms ". Jerusalem,
> 2006
>
> Jörn
>

Re: Perceptron Passive Aggressive training

[Jörn Kottmann <ko...@gmail.com>]

On 6/8/11 3:52 PM, Benson Margulies wrote:
> As with all PA, it depends on your data. It helps more with less data,
> as it were. But we love it.
>
> In the unlikely event that you have any problem reading the paper on
> how to apply it to the sequence problem, let me know. I'm still not
> organized to liberate our code, but I'm happy to extract explanations
> from it.

I found one paper about it on google, which one do you mean?

That is the paper I found, but still have to study it:
Crammer, Koby. Dekel, Ofer. Keshet, Joseph. Shalev-Shwartz,
Shai. Singer, Yoram. " Online Passive-Aggressive Algorithms ". 
Jerusalem, 2006

Jörn

Re: Perceptron Passive Aggressive training

[Benson Margulies <bi...@gmail.com>]

As with all PA, it depends on your data. It helps more with less data,
as it were. But we love it.

In the unlikely event that you have any problem reading the paper on
how to apply it to the sequence problem, let me know. I'm still not
organized to liberate our code, but I'm happy to extract explanations
from it.

On Wed, Jun 8, 2011 at 9:43 AM, Jörn Kottmann <ko...@gmail.com> wrote:
> Hi,
>
> should we additionally implement passive aggressive training
> for our perceptron and see how it then performs?
> Benson mentioned that he is using it, so I guess it works great.
>
> Any opinions?
>
> If we think its a great idea it would be nice to create a jira for it.
>
> Thanks,
> Jörn
>