Question about weights

[Cédrick Fairon]

Hello,

When using weights in graphs...

- is it necessary to give a specific weight to each path in the graph or does the system assume that path without weight have value "zero" ?

- is it possible to use weights in sub-graphs?
I kind of obtain what I want when I work with one graph, but as soon as I include the graph-with-weight in another graph, the weights do not discriminate the outputs anymore.


I have been trying so hard, that I have started to loose weight myself ;-)

C

[Artem Lukanin]

1) From my experience I can say, that paths have weight 0 by default.

2) I could not find an error in my graphs for several days, but it indeed appeared that weights are ignored in subgraphs. I began to use cascades and it fixed the problem, because now I call each subgraph, which has weights, in sequence. But I think, if it is not a bug, but a feature, it should be at least documented in the Weights section.

среда, 1 октября 2014 г., 23:50:10 UTC+6 пользователь Cédrick Fairon написал:

[Cristian Martinez]

Here are some considerations (as Revision #3705) about weighted paths that may be helpful:

- Weighted paths are still a experimental feature.
- The weight of a path is NOT the sum of the weights found in the path, but the last weight encountered on the path.
- Default weight is 'none' which means no weight is applied.
- Weight priority is none < 0 < i ; where i is a positive integer number. This notably means that a path of weight 0 is greater than another without explicit weight.
- Weights are applied only locally which means sub-graphs weights are never back-propagated.
- Weights have less priority than the match policy.For example, using a longest-matches policy, if we have a matched sequence m1 with a weight w1 and a longest one m2 with a weight w2, only m2 will be kept, even if w1 > w2.

Cheers.

[Denis Maurel]

Hi Christian,

I don't ubderstand? the Unitex manual said: "The weight of a path is the sum of the weights found in the path." and you "The weight of a path is NOT the sum of the weights found in the path, but the last weight encountered on the path."

If you are sure, please confirm us and change the manual...

Thank you very much

Best regards,

Denis Maurel


____________________________________
Professor Denis Maurel
Université François Rabelais Tours
LI (Computer Science Research Laboratory)
EPU-DI
64 avenue Jean-Portalis
37200 Tours
France
Phone: 33-2.47.36.14.35
Fax: 33-2.47.36.14.22
mailto:denis....@univ-tours.fr

http://www.univ-tours.fr/maurel

http://www.li.univ-tours.fr
http://tln.li.univ-tours.fr/

---

Here are some considerations (as Revision #3705) about weighted paths that may be helpful:

- Weighted paths are still a experimental feature.
- The weight of a path is NOT the sum of the weights found in the path, but the last weight encountered on the path.
- Default weight is 'none' which means no weight is applied.
- Weight priority is none < 0 < i ; where i is a positive integer number. This notably means that a path of weight 0 is greater than another without explicit weight.
- Weights are applied only locally which means sub-graphs weights are never back-propagated.
- Weights have less priority than the match policy.For example, using a longest-matches policy, if we have a matched sequence m1 with a weight w1 and a longest one m2 with a weight w2, only m2 will be kept, even if w1 > w2.

Cheers.

On Wednesday, October 1, 2014 7:50:10 PM UTC+2, Cédrick Fairon wrote:

--
You received this message because you are subscribed to the Google Groups "Unitex-GramLab" group.
To unsubscribe from this group and stop receiving emails from it, send an email to unitex-gramla...@googlegroups.com.
To post to this group, send email to unitex-...@googlegroups.com.
Visit this group at http://groups.google.com/group/unitex-gramlab.
To view this discussion on the web visit https://groups.google.com/d/msgid/unitex-gramlab/49fd5ed6-4969-4c60-8122-157875f54b99%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Cristian Martinez]

Hi Denis,

At least until the Revision #3705, the weight of a path is NOT the sum of the weights found in the path but the last weight encountered. Please see the attachment below illustrating this behavior.

Best Regards,

[Artem Lukanin]

Dear Cristian,

Thank you very much for the detailed explanation and the sample graph!

Artem

--
You received this message because you are subscribed to a topic in the Google Groups "Unitex-GramLab" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/unitex-gramlab/J03HMW--vAE/unsubscribe.
To unsubscribe from this group and all its topics, send an email to unitex-gramla...@googlegroups.com.

To post to this group, send email to unitex-...@googlegroups.com.
Visit this group at http://groups.google.com/group/unitex-gramlab.

To view this discussion on the web visit https://groups.google.com/d/msgid/unitex-gramlab/76087658-5f37-4c81-9f0c-6cac5379de90%40googlegroups.com.

[Denis Maurel]

Hi Cristian,

ok Thanks.

It is a big change. We use a lot weight in CasEN and we need to modify our graphs. Please, change the manual.

When we compile a graph with subgraphs, we have just one .fst2 Why it is not possible to use weight in subgraph? May be it is possible to change this?

Best regards,

Denis Maurel


____________________________________
Professor Denis Maurel
Université François Rabelais Tours
LI (Computer Science Research Laboratory)
EPU-DI
64 avenue Jean-Portalis
37200 Tours
France
Phone: 33-2.47.36.14.35
Fax: 33-2.47.36.14.22
mailto:denis....@univ-tours.fr

http://www.univ-tours.fr/maurel

http://www.li.univ-tours.fr
http://tln.li.univ-tours.fr/

---

Hi Denis,

At least until the Revision #3705, the weight of a path is NOT the sum of the weights found in the path but the last weight encountered. Please see the attachment below illustrating this behavior.

Best Regards,

On Wednesday, November 12, 2014 5:02:59 PM UTC+1, Denis MAUREL wrote:

[Cédrick Fairon]

I would also be interested by the possibility to use weights in the subgraphs!

C

--
You received this message because you are subscribed to the Google Groups "Unitex-GramLab" group.
To unsubscribe from this group and stop receiving emails from it, send an email to unitex-gramla...@googlegroups.com.
To post to this group, send email to unitex-...@googlegroups.com.
Visit this group at http://groups.google.com/group/unitex-gramlab.

To view this discussion on the web visit https://groups.google.com/d/msgid/unitex-gramlab/130292187.2936628.1415864573116.JavaMail.zimbra%40univ-tours.fr.

[Artem Lukanin]

I would like to use weights in subgraphs as well! It's a very convenient way to distinguish rules and exception

Artem

--
You received this message because you are subscribed to a topic in the Google Groups "Unitex-GramLab" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/unitex-gramlab/J03HMW--vAE/unsubscribe.
To unsubscribe from this group and all its topics, send an email to unitex-gramla...@googlegroups.com.

To post to this group, send email to unitex-...@googlegroups.com.
Visit this group at http://groups.google.com/group/unitex-gramlab.

To view this discussion on the web visit https://groups.google.com/d/msgid/unitex-gramlab/9471780C-4EFD-44DD-9E0D-51275F1D214B%40gmail.com.

[eric.laporte]

Dear Denis,

On Thursday, 13 November 2014 08:42:55 UTC+1, Denis MAUREL wrote:
<<

It is a big change. We use a lot weight in CasEN and we need to modify our graphs.

>>
Apparently, it is not a change, it has always been so. It was an error in the manual.

<<

Please, change the manual.

>>
I updated the manual. The new version is online. Apparently the behaviour of the graphs is not what users want, but at least the manual describes it faithfully now.

Best,
Eric

[Denis Maurel]

Hi Eric,

Thank you to update the manual. And to inform the Unitex community!

The very useful modification about weight would be the possibility to use them in subgraphs... If someone can implement that...

Best regards,

Denis Maurel


____________________________________
Professor Denis Maurel
Université François Rabelais Tours
LI (Computer Science Research Laboratory)
EPU-DI
64 avenue Jean-Portalis
37200 Tours
France
Phone: 33-2.47.36.14.35
Fax: 33-2.47.36.14.22
mailto:denis....@univ-tours.fr

http://www.univ-tours.fr/maurel

http://www.li.univ-tours.fr
http://tln.li.univ-tours.fr/

---

Dear Denis,