about missing data

[changhe]

Hi,All
When I use the missing data, pebl return an error:

File "C:\Python25\lib\site-packages\pebl-1.01-py2.5-win32.egg\pebl
\learner\greedy.py", line 134, in _run_without_restarts
self.evaluator.restore_network()
AttributeError: 'MissingDataNetworkEvaluator' object has no attribute
'restore_network'

I check the class MissingDataNetworkEvaluator and its base class, and
cann't find attribute 'restore_network' .

How can I do with it. Thank you!

[Abhik Shah]

Hi,

I think this was a bug in an older version of pebl. The latest
download on google-code and the latest code in SVN both have the
restore_network method.

Thanks,
Abhik

[changhe]

Hi,Abhik
I can run it in the version1.0.2.
But the program runs so slowly when number of missing data is large.
what is a good way to solve the problem!

Thank you!
Changhe

> > How can I do with it. Thank you!- 隐藏被引用文字 -
>
> - 显示引用的文字 -

[Abhik Shah]

Hi Changhe,

You can use the gibbs.max_iterations
(http://ano.malo.us/pebl/docs/paramref.html#confparam-gibbs.max_iterations)
parameter to control how long the sampling takes.. but Gibbs sampling
is known to be slow and if you decrease the number of iterations too
much, the code will finish fast but the sampler will not have
converged.

If you have too many hidden variables, you might want to look at
variational methods. pebl doesn't include any but they are well
described in literature and shouldn't be too difficult to implement.

Thanks,
Abhik.

2010/11/5 changhe <fuch...@gmail.com>:

[changhe]

Hi, Abhik

Thank you very much

Best,
Changhe

On 11月5日, 下午2时32分, Abhik Shah <abhiks...@gmail.com> wrote:
> Hi Changhe,
>
> You can use the gibbs.max_iterations

> (http://ano.malo.us/pebl/docs/paramref.html#confparam-gibbs.max_iterat...)

> parameter to control how long the sampling takes.. but Gibbs sampling
> is known to be slow and if you decrease the number of iterations too
> much, the code will finish fast but the sampler will not have
> converged.
>
> If you have too many hidden variables, you might want to look at
> variational methods. pebl doesn't include any but they are well
> described in literature and shouldn't be too difficult to implement.
>
> Thanks,
> Abhik.
>

> 2010/11/5 changhe <fuchan...@gmail.com>:

>
>
>
> > Hi,Abhik
> > I can run it in the version1.0.2.
> > But the program runs so slowly when number of missing data is large.
> > what is a good way to solve the problem!
>
> > Thank you!
> > Changhe
>
> > On 11月4日, 下午5时26分, Abhik Shah <abhiks...@gmail.com> wrote:
> >> Hi,
>
> >> I think this was a bug in an older version of pebl. The latest
> >> download on google-code and the latest code in SVN both have the
> >> restore_network method.
>
> >> Thanks,
> >> Abhik
>
> >> On Thu, Nov 4, 2010 at 7:03 PM, changhe <fuchan...@gmail.com> wrote:
> >> > Hi,All
> >> > When I use the missing data, pebl return an error:
>
> >> > File "C:\Python25\lib\site-packages\pebl-1.01-py2.5-win32.egg\pebl
> >> > \learner\greedy.py", line 134, in _run_without_restarts
> >> > self.evaluator.restore_network()
> >> > AttributeError: 'MissingDataNetworkEvaluator' object has no attribute
> >> > 'restore_network'
>
> >> > I check the class MissingDataNetworkEvaluator and its base class, and
> >> > cann't find attribute 'restore_network' .
>
> >> > How can I do with it. Thank you!- 隐藏被引用文字 -
>

> >> - 显示引用的文字 -- 隐藏被引用文字 -
>
> - 显示引用的文字 -