OpenCog pattern miner

[Hedra Seid]

Hello

I have been working with OpenCog pattern miner and it sometimes stack and stop mining at some point ... Have waited for Pattern miner result more than half an hour where the atomspace contains only 55 atoms and it stops at 40%. Any solutions?

[Ben Goertzel]

Shujing, perhaps you can help with this?

Hedra, it would be good if you could give precise instructions for replicating the problem.... that will make it easier for Shujing or Nil to help...

Thanks
Ben

On Nov 9, 2016 08:07, "Hedra Seid" <tank...@gmail.com> wrote:

Hello

I have been working with OpenCog pattern miner and it sometimes stack and stop mining at some point ... Have waited for Pattern miner result more than half an hour where the atomspace contains only 55 atoms and it stops at 40%. Any solutions?

--
You received this message because you are subscribed to the Google Groups "opencog" group.
To unsubscribe from this group and stop receiving emails from it, send an email to opencog+unsubscribe@googlegroups.com.
To post to this group, send email to ope...@googlegroups.com.
Visit this group at https://groups.google.com/group/opencog.
To view this discussion on the web visit https://groups.google.com/d/msgid/opencog/4ef5fae7-dd3f-410a-ae26-c9c4e8ee4f50%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Hedra Seid]

Okay, Ben

I checked the pattern miner with the provided example corpus file and it works fine. But the problem is it sometimes stop mining with unknown reason for me.

Here is what I tried

we are working on applying OpenCog pattern miner as a dialogue assistance to help the customer service in providing answers mining what has been said before (or mining from the knowledge base atomspace).

- We have dialogues parsed and loaded to the atomspace, and loading the opencog pattern miner module, the cogserver results the  following

Error: instance link contains variables:
(VariableNode "$wF0EfmscF9F25N7BAoBtBNmDChd5Ep0Cd9q0") ; [111651][111516]

- We write the Atomese representation of each dialogues (like the aiml to atomese representations) and load to the atomspace and start the opencog mining

here the pattern miner works and generates output files, but only the 2gram results are extracted.

- adding some knowledge tried the same data but the pattern miner stoped mining reaching 40% ... calculating the surprisingness

--

With Regards!

Hedra Seid

[Ben Goertzel]

Hedra,

I suggest you should upload to some repository the actual data that
you loaded into the Pattern Miner, and the exact script that you ran
to launch the pattern miner

That way Nil or Shujing can follow the exact steps that you did, and
see exactly how the bug was encountered.

Without a clear way to replicate the bug, it may be hard for them to debug it

I would like them to be able to exactly replicate what you did on
their machines...

thanks
ben

>>> email to opencog+u...@googlegroups.com.

>>> To post to this group, send email to ope...@googlegroups.com.
>>> Visit this group at https://groups.google.com/group/opencog.
>>> To view this discussion on the web visit
>>> https://groups.google.com/d/msgid/opencog/4ef5fae7-dd3f-410a-ae26-c9c4e8ee4f50%40googlegroups.com.
>>> For more options, visit https://groups.google.com/d/optout.
>
>
>
>
> --
> With Regards!
>
> Hedra Seid
>

--
Ben Goertzel, PhD
http://goertzel.org

“I tell my students, when you go to these meetings, see what direction
everyone is headed, so you can go in the opposite direction. Don’t
polish the brass on the bandwagon.” – V. S. Ramachandran

[Hedra Seid]

Hi Ben, Sorry for my late reply ...

Here attached are the scheme files I used to apply the OpenCog pattern miner

STEPS I followed

1. Start opencog server with the config file  -c ../lib/opencog_patternminer.conf and relex server

2. Telnet to opencog server on port 17001

3. Access the scheme shell

4. (use-modules (opencog) (opencog nlp) (opencog nlp chatbot) (opencog nlp relex2logic))

I import those modules to use (nlp-parse) and let relex parse the sentences in the dialogue but it creates a VariableNode instance which the Pattern miner considers as an error node type to have. To fix this I have to make the atomspace free.

5. (clear)

6. (load-scm-from-file "test-dialogue.scm")
    (load-scm-from-file "back-ground.scm")

here the atomspace contains 9900 atoms

7. Exit the scheme shell and load the test pettern miner agent module

   loadmodule /home/tanksha/opencog_ws/opencog/build/opencog/learning/PatternMiner/libTestPatternMinerAgent.so

8. From the opencog server I got this

Listening on port 17001
Debug: PatternMining start! Max gram = 4, mode = Depth_First
Start thread 0 from 0 to 7467
0% completed in Thread 0.

It stays like this for a long time ... the percent doesn’t increase.

>>> email to opencog+unsubscribe@googlegroups.com.

>>> To post to this group, send email to ope...@googlegroups.com.
>>> Visit this group at https://groups.google.com/group/opencog.
>>> To view this discussion on the web visit
>>> https://groups.google.com/d/msgid/opencog/4ef5fae7-dd3f-410a-ae26-c9c4e8ee4f50%40googlegroups.com.
>>> For more options, visit https://groups.google.com/d/optout.
>
>
>
>
> --
> With Regards!
>
> Hedra Seid
>

--
Ben Goertzel, PhD
http://goertzel.org

“I tell my students, when you go to these meetings, see what direction
everyone is headed, so you can go in the opposite direction. Don’t
polish the brass on the bandwagon.” – V. S. Ramachandran

[Ben Goertzel]

Thanks Hedra!

Nil -- I hate to distract you from the BC, but could you please take a
look at this sometime soon? It seems Shu Jing is too busy to help
out, and I know you have played a bit with the Pattern Miner a few
months ago...

thx
ben

>> >>> email to opencog+u...@googlegroups.com.

[Nil Geisweiller]

Hi,

the last time I reviewed the pattern miner I found that it was not ready
for general use, and estimated it would take me about 2 weeks to put it
in shape. I had started this branch for that
https://github.com/ngeiswei/opencog/commits/hack-pattern-miner

Anyway, I'll have a look at it within the next couple days, who knows,
it might not be much.

Hedra, could you create an issue on the opencog repo (basically
re-pasting your report and attaching your files there)? If you can
reduce the KB and dialogue size it's even better, maybe it could serve
as a unit test.

Thanks,
Nil

[Hedra Seid]

Hi

I created the issue

I also add another example which I was trying today without the use of relex server ... and it doesn’t work too. Find the issue #2477 from the OpenCog Github repository. Thanks.

email to opencog+unsubscribe@googlegroups.com.

To post to this group, send email to ope...@googlegroups.com.
Visit this group at https://groups.google.com/group/opencog.
To view this discussion on the web visit

https://groups.google.com/d/msgid/opencog/4ef5fae7-dd3f-410a-ae26-c9c4e8ee4f50%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

--
With Regards!

Hedra Seid

--
Ben Goertzel, PhD
http://goertzel.org

“I tell my students, when you go to these meetings, see what direction
everyone is headed, so you can go in the opposite direction. Don’t
polish the brass on the bandwagon.” – V. S. Ramachandran

--
With Regards!

Hedra Seid

[Ben Goertzel]

Nil,

Well, we do need to use the pattern miner for a couple projects pretty
soon... but that doesn't mean it has to be your job to fix it up...

It would be great though if you could write down a detailed list of
what you feel needs to be done to fix it up, so Hedra can give it a
shot (and maybe you can do some of the tricker bits?) ...

ben

[Nil Geisweiller]

On 11/14/2016 04:54 PM, Ben Goertzel wrote:
> Nil,
>
> Well, we do need to use the pattern miner for a couple projects pretty
> soon... but that doesn't mean it has to be your job to fix it up...
>
> It would be great though if you could write down a detailed list of
> what you feel needs to be done to fix it up, so Hedra can give it a
> shot (and maybe you can do some of the tricker bits?) ...

Sounds good. I'll look into it after tomorrow, merge my branch about it
so far and try to sum up what remains to be done.

Nil