[graph] Why the Vertex and Edge interfaces?

[James Carman]

Sorry if this was double-posted.  I think I accidentally sent it from
my "personal" email account too (which isn't subscribed).

Anyway, shouldn't we just let anything (perhaps restrict it to
Serializables) be a vertex or an edge?

-
To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
For additional commands, e-mail: dev-h...@commons.apache.org

Re: [graph] Why the Vertex and Edge interfaces?

[Simone Tripodi]

Hi James!

the email was posted twice because looks like you posted to old
jakarta address :P

The main reason because these interfaces are there, is that when I
started resurrecting it I just opened my Graph book, I started
defining element according to definitions :P

Some already implemented algorithms require specific informations
about edges - specifically WeightedEdges - so while for Vertex a
complete generalization can still be applied, Edge requires some
assumptions at a certain point.

Claudio is aware also about algorithms where weights are associated to
Vertex - he's preparing his PhD research on graphes - maybe he can
show us a more long-vision roadmap and evaluate benefits on
simplifying the design.

Do you have already a proposal how to modify the actual design?
TIA, all the best,
-Simo

http://people.apache.org/~simonetripodi/
http://simonetripodi.livejournal.com/
http://twitter.com/simonetripodi
http://www.99soft.org/



On Fri, Mar 2, 2012 at 2:45 AM, James Carman  wrote:
> Sorry if this was double-posted.  I think I accidentally sent it from
> my "personal" email account too (which isn't subscribed).
>
> Anyway, shouldn't we just let anything (perhaps restrict it to
> Serializables) be a vertex or an edge?
>
> -
> To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
> For additional commands, e-mail: dev-h...@commons.apache.org
>

-
To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
For additional commands, e-mail: dev-h...@commons.apache.org

Re: [graph] Why the Vertex and Edge interfaces?

[James Carman]

Yeah, I would say something like this:

public interface Graph
{
...
}

this is assuming you'd want to enforce the serialization stuff.
Otherwise it'd just be Graph

On Fri, Mar 2, 2012 at 2:47 AM, Simone Tripodi  wrote:
> Hi James!
>
> the email was posted twice because looks like you posted to old
> jakarta address :P
>
> The main reason because these interfaces are there, is that when I
> started resurrecting it I just opened my Graph book, I started
> defining element according to definitions :P
>
> Some already implemented algorithms require specific informations
> about edges - specifically WeightedEdges - so while for Vertex a
> complete generalization can still be applied, Edge requires some
> assumptions at a certain point.
>
> Claudio is aware also about algorithms where weights are associated to
> Vertex - he's preparing his PhD research on graphes - maybe he can
> show us a more long-vision roadmap and evaluate benefits on
> simplifying the design.
>
> Do you have already a proposal how to modify the actual design?
> TIA, all the best,
> -Simo
>
> http://people.apache.org/~simonetripodi/
> http://simonetripodi.livejournal.com/
> http://twitter.com/simonetripodi
> http://www.99soft.org/
>
>
>
> On Fri, Mar 2, 2012 at 2:45 AM, James Carman  
> wrote:
>> Sorry if this was double-posted.  I think I accidentally sent it from
>> my "personal" email account too (which isn't subscribed).
>>
>> Anyway, shouldn't we just let anything (perhaps restrict it to
>> Serializables) be a vertex or an edge?
>>
>> -
>> To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
>> For additional commands, e-mail: dev-h...@commons.apache.org
>>
>
> -
> To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
> For additional commands, e-mail: dev-h...@commons.apache.org
>

-
To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
For additional commands, e-mail: dev-h...@commons.apache.org

Re: [graph] Why the Vertex and Edge interfaces?

[Simone Tripodi]

Hi James!

thanks for the suggestions! I think your idea is more than valid, I
just need to find a slot to experiment, but feel free to provide your
contribution!

All the best,
-Simo

http://people.apache.org/~simonetripodi/
http://simonetripodi.livejournal.com/
http://twitter.com/simonetripodi
http://www.99soft.org/



On Fri, Mar 2, 2012 at 8:50 AM, James Carman  wrote:
> Yeah, I would say something like this:
>
> public interface Graph
> {
> ...
> }
>
> this is assuming you'd want to enforce the serialization stuff.
> Otherwise it'd just be Graph
>
> On Fri, Mar 2, 2012 at 2:47 AM, Simone Tripodi  
> wrote:
>> Hi James!
>>
>> the email was posted twice because looks like you posted to old
>> jakarta address :P
>>
>> The main reason because these interfaces are there, is that when I
>> started resurrecting it I just opened my Graph book, I started
>> defining element according to definitions :P
>>
>> Some already implemented algorithms require specific informations
>> about edges - specifically WeightedEdges - so while for Vertex a
>> complete generalization can still be applied, Edge requires some
>> assumptions at a certain point.
>>
>> Claudio is aware also about algorithms where weights are associated to
>> Vertex - he's preparing his PhD research on graphes - maybe he can
>> show us a more long-vision roadmap and evaluate benefits on
>> simplifying the design.
>>
>> Do you have already a proposal how to modify the actual design?
>> TIA, all the best,
>> -Simo
>>
>> http://people.apache.org/~simonetripodi/
>> http://simonetripodi.livejournal.com/
>> http://twitter.com/simonetripodi
>> http://www.99soft.org/
>>
>>
>>
>> On Fri, Mar 2, 2012 at 2:45 AM, James Carman  
>> wrote:
>>> Sorry if this was double-posted.  I think I accidentally sent it from
>>> my "personal" email account too (which isn't subscribed).
>>>
>>> Anyway, shouldn't we just let anything (perhaps restrict it to
>>> Serializables) be a vertex or an edge?
>>>
>>> -
>>> To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
>>> For additional commands, e-mail: dev-h...@commons.apache.org
>>>
>>
>> -
>> To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
>> For additional commands, e-mail: dev-h...@commons.apache.org
>>
>
> -
> To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
> For additional commands, e-mail: dev-h...@commons.apache.org
>

-
To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
For additional commands, e-mail: dev-h...@commons.apache.org

Re: [graph] Why the Vertex and Edge interfaces?

[Claudio Squarcella]

Hi all,


Claudio is aware also about algorithms where weights are associated to
Vertex - he's preparing his PhD research on graphes - maybe he can
show us a more long-vision roadmap and evaluate benefits on
simplifying the design.


yes there are algorithms with weights on vertices. Of course those with 
weighted edges (like the ones already implemented) are much more 
widespread and frequently used, but still we cannot forget about that. 
Also, although on a secondary level, labels on vertices/edges are kind 
of important in many situations (including testing, debugging) where I 
think it is good to keep them distinct from the standard "toString" 
method (you might want to represent only a subset of info in the label, 
etc).


Matthew Pocock suggested an alternative approach back in the days of 
weight abstraction:


 * the graph itself is extremely simple and naked: no weights/labels on
   vertices/edges;
 * all properties are stored in some external structure, which I
   imagine composed of associative maps (Map, etc etc).

He motivated the idea with a "personal use case": often graphs are used 
and reused with the same structure but different weights (and/or labels, 
etc). Now if James' question becomes a second use case, maybe it's the 
right time to exhume that idea ;)


Ciao,
Claudio

--
Claudio Squarcella
PhD student at Roma Tre University
http://www.dia.uniroma3.it/~squarcel
http://squarcella.com/


-
To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
For additional commands, e-mail: dev-h...@commons.apache.org

Re: [graph] Why the Vertex and Edge interfaces?

[James Ring]

Java Collections classes typically specify that they are serializable if
their type parameter(s) are serializable. It shouldn't be required.
On Mar 1, 2012 9:51 PM, "James Carman"  wrote:

Re: [graph] Why the Vertex and Edge interfaces?

[Ted Dunning]

Having weights on vertices is quite common.  Consider any probability
transition network.  The weight on each node is the probability of being in
that state and the weights on the edges are conditional probabilties.

Page rank is a related example of having weights on nodes.

On Fri, Mar 2, 2012 at 12:40 AM, Claudio Squarcella <
squar...@dia.uniroma3.it> wrote:

> Hi all,
>
>  Claudio is aware also about algorithms where weights are associated to
>> Vertex - he's preparing his PhD research on graphes - maybe he can
>> show us a more long-vision roadmap and evaluate benefits on
>> simplifying the design.
>>
>
> yes there are algorithms with weights on vertices. Of course those with
> weighted edges (like the ones already implemented) are much more widespread
> and frequently used, but still we cannot forget about that. Also, although
> on a secondary level, labels on vertices/edges are kind of important in
> many situations (including testing, debugging) where I think it is good to
> keep them distinct from the standard "toString" method (you might want to
> represent only a subset of info in the label, etc).
>
> Matthew Pocock suggested an alternative approach back in the days of
> weight abstraction:
>
>  * the graph itself is extremely simple and naked: no weights/labels on
>   vertices/edges;
>  * all properties are stored in some external structure, which I
>   imagine composed of associative maps (Map, etc etc).
>
> He motivated the idea with a "personal use case": often graphs are used
> and reused with the same structure but different weights (and/or labels,
> etc). Now if James' question becomes a second use case, maybe it's the
> right time to exhume that idea ;)
>
> Ciao,
> Claudio
>
> --
> Claudio Squarcella
> PhD student at Roma Tre University
> http://www.dia.uniroma3.it/~**squarcel
> http://squarcella.com/
>
>
> --**--**-
> To unsubscribe, e-mail: 
> dev-unsubscribe@commons.**apache.org
> For additional commands, e-mail: dev-h...@commons.apache.org
>
>

Re: [graph] Why the Vertex and Edge interfaces?

[Simone Tripodi]

Hi Ted,

thanks for your valuable feedback! Please take in consideration that
contributions are more than welcome, so if you intend sending
contributions about this algorithms, they would bu much more than
appreciated!

best,
-Simo

http://people.apache.org/~simonetripodi/
http://simonetripodi.livejournal.com/
http://twitter.com/simonetripodi
http://www.99soft.org/



On Fri, Mar 2, 2012 at 9:07 PM, Ted Dunning  wrote:
> Having weights on vertices is quite common.  Consider any probability
> transition network.  The weight on each node is the probability of being in
> that state and the weights on the edges are conditional probabilties.
>
> Page rank is a related example of having weights on nodes.
>
> On Fri, Mar 2, 2012 at 12:40 AM, Claudio Squarcella <
> squar...@dia.uniroma3.it> wrote:
>
>> Hi all,
>>
>>  Claudio is aware also about algorithms where weights are associated to
>>> Vertex - he's preparing his PhD research on graphes - maybe he can
>>> show us a more long-vision roadmap and evaluate benefits on
>>> simplifying the design.
>>>
>>
>> yes there are algorithms with weights on vertices. Of course those with
>> weighted edges (like the ones already implemented) are much more widespread
>> and frequently used, but still we cannot forget about that. Also, although
>> on a secondary level, labels on vertices/edges are kind of important in
>> many situations (including testing, debugging) where I think it is good to
>> keep them distinct from the standard "toString" method (you might want to
>> represent only a subset of info in the label, etc).
>>
>> Matthew Pocock suggested an alternative approach back in the days of
>> weight abstraction:
>>
>>  * the graph itself is extremely simple and naked: no weights/labels on
>>   vertices/edges;
>>  * all properties are stored in some external structure, which I
>>   imagine composed of associative maps (Map, etc etc).
>>
>> He motivated the idea with a "personal use case": often graphs are used
>> and reused with the same structure but different weights (and/or labels,
>> etc). Now if James' question becomes a second use case, maybe it's the
>> right time to exhume that idea ;)
>>
>> Ciao,
>> Claudio
>>
>> --
>> Claudio Squarcella
>> PhD student at Roma Tre University
>> http://www.dia.uniroma3.it/~**squarcel
>> http://squarcella.com/
>>
>>
>> --**--**-
>> To unsubscribe, e-mail: 
>> dev-unsubscribe@commons.**apache.org
>> For additional commands, e-mail: dev-h...@commons.apache.org
>>
>>

-
To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
For additional commands, e-mail: dev-h...@commons.apache.org

Re: [graph] Why the Vertex and Edge interfaces?

[Ted Dunning]

No way that I have time to make real contributions, but if I see a quick
something to drop into the discussion, I will.

On Fri, Mar 2, 2012 at 12:25 PM, Simone Tripodi wrote:

> Hi Ted,
>
> thanks for your valuable feedback! Please take in consideration that
> contributions are more than welcome, so if you intend sending
> contributions about this algorithms, they would bu much more than
> appreciated!
>
> best,
> -Simo
>
> http://people.apache.org/~simonetripodi/
> http://simonetripodi.livejournal.com/
> http://twitter.com/simonetripodi
> http://www.99soft.org/
>
>
>
> On Fri, Mar 2, 2012 at 9:07 PM, Ted Dunning  wrote:
> > Having weights on vertices is quite common.  Consider any probability
> > transition network.  The weight on each node is the probability of being
> in
> > that state and the weights on the edges are conditional probabilties.
> >
> > Page rank is a related example of having weights on nodes.
> >
> > On Fri, Mar 2, 2012 at 12:40 AM, Claudio Squarcella <
> > squar...@dia.uniroma3.it> wrote:
> >
> >> Hi all,
> >>
> >>  Claudio is aware also about algorithms where weights are associated to
> >>> Vertex - he's preparing his PhD research on graphes - maybe he can
> >>> show us a more long-vision roadmap and evaluate benefits on
> >>> simplifying the design.
> >>>
> >>
> >> yes there are algorithms with weights on vertices. Of course those with
> >> weighted edges (like the ones already implemented) are much more
> widespread
> >> and frequently used, but still we cannot forget about that. Also,
> although
> >> on a secondary level, labels on vertices/edges are kind of important in
> >> many situations (including testing, debugging) where I think it is good
> to
> >> keep them distinct from the standard "toString" method (you might want
> to
> >> represent only a subset of info in the label, etc).
> >>
> >> Matthew Pocock suggested an alternative approach back in the days of
> >> weight abstraction:
> >>
> >>  * the graph itself is extremely simple and naked: no weights/labels on
> >>   vertices/edges;
> >>  * all properties are stored in some external structure, which I
> >>   imagine composed of associative maps (Map, etc etc).
> >>
> >> He motivated the idea with a "personal use case": often graphs are used
> >> and reused with the same structure but different weights (and/or labels,
> >> etc). Now if James' question becomes a second use case, maybe it's the
> >> right time to exhume that idea ;)
> >>
> >> Ciao,
> >> Claudio
> >>
> >> --
> >> Claudio Squarcella
> >> PhD student at Roma Tre University
> >> http://www.dia.uniroma3.it/~**squarcel<
> http://www.dia.uniroma3.it/~squarcel>
> >> http://squarcella.com/
> >>
> >>
> >>
> --**--**-
> >> To unsubscribe, e-mail: dev-unsubscribe@commons.**apache.org<
> dev-unsubscr...@commons.apache.org>
> >> For additional commands, e-mail: dev-h...@commons.apache.org
> >>
> >>
>
> -
> To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
> For additional commands, e-mail: dev-h...@commons.apache.org
>
>

Re: [graph] Why the Vertex and Edge interfaces?

[Simone Tripodi]

That would be appreciated as well, thanks!
-Simo

http://people.apache.org/~simonetripodi/
http://simonetripodi.livejournal.com/
http://twitter.com/simonetripodi
http://www.99soft.org/



On Fri, Mar 2, 2012 at 9:30 PM, Ted Dunning  wrote:
> No way that I have time to make real contributions, but if I see a quick
> something to drop into the discussion, I will.
>
> On Fri, Mar 2, 2012 at 12:25 PM, Simone Tripodi 
> wrote:
>
>> Hi Ted,
>>
>> thanks for your valuable feedback! Please take in consideration that
>> contributions are more than welcome, so if you intend sending
>> contributions about this algorithms, they would bu much more than
>> appreciated!
>>
>> best,
>> -Simo
>>
>> http://people.apache.org/~simonetripodi/
>> http://simonetripodi.livejournal.com/
>> http://twitter.com/simonetripodi
>> http://www.99soft.org/
>>
>>
>>
>> On Fri, Mar 2, 2012 at 9:07 PM, Ted Dunning  wrote:
>> > Having weights on vertices is quite common.  Consider any probability
>> > transition network.  The weight on each node is the probability of being
>> in
>> > that state and the weights on the edges are conditional probabilties.
>> >
>> > Page rank is a related example of having weights on nodes.
>> >
>> > On Fri, Mar 2, 2012 at 12:40 AM, Claudio Squarcella <
>> > squar...@dia.uniroma3.it> wrote:
>> >
>> >> Hi all,
>> >>
>> >>  Claudio is aware also about algorithms where weights are associated to
>> >>> Vertex - he's preparing his PhD research on graphes - maybe he can
>> >>> show us a more long-vision roadmap and evaluate benefits on
>> >>> simplifying the design.
>> >>>
>> >>
>> >> yes there are algorithms with weights on vertices. Of course those with
>> >> weighted edges (like the ones already implemented) are much more
>> widespread
>> >> and frequently used, but still we cannot forget about that. Also,
>> although
>> >> on a secondary level, labels on vertices/edges are kind of important in
>> >> many situations (including testing, debugging) where I think it is good
>> to
>> >> keep them distinct from the standard "toString" method (you might want
>> to
>> >> represent only a subset of info in the label, etc).
>> >>
>> >> Matthew Pocock suggested an alternative approach back in the days of
>> >> weight abstraction:
>> >>
>> >>  * the graph itself is extremely simple and naked: no weights/labels on
>> >>   vertices/edges;
>> >>  * all properties are stored in some external structure, which I
>> >>   imagine composed of associative maps (Map, etc etc).
>> >>
>> >> He motivated the idea with a "personal use case": often graphs are used
>> >> and reused with the same structure but different weights (and/or labels,
>> >> etc). Now if James' question becomes a second use case, maybe it's the
>> >> right time to exhume that idea ;)
>> >>
>> >> Ciao,
>> >> Claudio
>> >>
>> >> --
>> >> Claudio Squarcella
>> >> PhD student at Roma Tre University
>> >> http://www.dia.uniroma3.it/~**squarcel<
>> http://www.dia.uniroma3.it/~squarcel>
>> >> http://squarcella.com/
>> >>
>> >>
>> >>
>> --**--**-
>> >> To unsubscribe, e-mail: dev-unsubscribe@commons.**apache.org<
>> dev-unsubscr...@commons.apache.org>
>> >> For additional commands, e-mail: dev-h...@commons.apache.org
>> >>
>> >>
>>
>> -
>> To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
>> For additional commands, e-mail: dev-h...@commons.apache.org
>>
>>

-
To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
For additional commands, e-mail: dev-h...@commons.apache.org

Re: [graph] Why the Vertex and Edge interfaces?

[James Carman]

My point isn't that we would not need weights ever but we're putting a
needless restriction on the users (implementing a marker interface).  The
Edge and Vertex interfaces add no value.

The weights can be external, too.  It's only a function from edge to
weight.  Your algorithm can take a function for its weights.  The files
library does it similar to this.
On Mar 2, 2012 3:08 PM, "Ted Dunning"  wrote:

> Having weights on vertices is quite common.  Consider any probability
> transition network.  The weight on each node is the probability of being in
> that state and the weights on the edges are conditional probabilties.
>
> Page rank is a related example of having weights on nodes.
>
> On Fri, Mar 2, 2012 at 12:40 AM, Claudio Squarcella <
> squar...@dia.uniroma3.it> wrote:
>
> > Hi all,
> >
> >  Claudio is aware also about algorithms where weights are associated to
> >> Vertex - he's preparing his PhD research on graphes - maybe he can
> >> show us a more long-vision roadmap and evaluate benefits on
> >> simplifying the design.
> >>
> >
> > yes there are algorithms with weights on vertices. Of course those with
> > weighted edges (like the ones already implemented) are much more
> widespread
> > and frequently used, but still we cannot forget about that. Also,
> although
> > on a secondary level, labels on vertices/edges are kind of important in
> > many situations (including testing, debugging) where I think it is good
> to
> > keep them distinct from the standard "toString" method (you might want to
> > represent only a subset of info in the label, etc).
> >
> > Matthew Pocock suggested an alternative approach back in the days of
> > weight abstraction:
> >
> >  * the graph itself is extremely simple and naked: no weights/labels on
> >   vertices/edges;
> >  * all properties are stored in some external structure, which I
> >   imagine composed of associative maps (Map, etc etc).
> >
> > He motivated the idea with a "personal use case": often graphs are used
> > and reused with the same structure but different weights (and/or labels,
> > etc). Now if James' question becomes a second use case, maybe it's the
> > right time to exhume that idea ;)
> >
> > Ciao,
> > Claudio
> >
> > --
> > Claudio Squarcella
> > PhD student at Roma Tre University
> > http://www.dia.uniroma3.it/~**squarcel<
> http://www.dia.uniroma3.it/~squarcel>
> > http://squarcella.com/
> >
> >
> > --**--**-
> > To unsubscribe, e-mail: dev-unsubscribe@commons.**apache.org<
> dev-unsubscr...@commons.apache.org>
> > For additional commands, e-mail: dev-h...@commons.apache.org
> >
> >
>

Re: [graph] Why the Vertex and Edge interfaces?

[Simone Tripodi]

Hi James,

looking at the the code I think we can safety keep the Weighted*
version AND use external functions you described. It is the same
Comparable/Comparator pattern, WDYT?

best,
-Simo

http://people.apache.org/~simonetripodi/
http://simonetripodi.livejournal.com/
http://twitter.com/simonetripodi
http://www.99soft.org/



On Fri, Mar 2, 2012 at 10:09 PM, James Carman
 wrote:
> My point isn't that we would not need weights ever but we're putting a
> needless restriction on the users (implementing a marker interface).  The
> Edge and Vertex interfaces add no value.
>
> The weights can be external, too.  It's only a function from edge to
> weight.  Your algorithm can take a function for its weights.  The files
> library does it similar to this.
> On Mar 2, 2012 3:08 PM, "Ted Dunning"  wrote:
>
>> Having weights on vertices is quite common.  Consider any probability
>> transition network.  The weight on each node is the probability of being in
>> that state and the weights on the edges are conditional probabilties.
>>
>> Page rank is a related example of having weights on nodes.
>>
>> On Fri, Mar 2, 2012 at 12:40 AM, Claudio Squarcella <
>> squar...@dia.uniroma3.it> wrote:
>>
>> > Hi all,
>> >
>> >  Claudio is aware also about algorithms where weights are associated to
>> >> Vertex - he's preparing his PhD research on graphes - maybe he can
>> >> show us a more long-vision roadmap and evaluate benefits on
>> >> simplifying the design.
>> >>
>> >
>> > yes there are algorithms with weights on vertices. Of course those with
>> > weighted edges (like the ones already implemented) are much more
>> widespread
>> > and frequently used, but still we cannot forget about that. Also,
>> although
>> > on a secondary level, labels on vertices/edges are kind of important in
>> > many situations (including testing, debugging) where I think it is good
>> to
>> > keep them distinct from the standard "toString" method (you might want to
>> > represent only a subset of info in the label, etc).
>> >
>> > Matthew Pocock suggested an alternative approach back in the days of
>> > weight abstraction:
>> >
>> >  * the graph itself is extremely simple and naked: no weights/labels on
>> >   vertices/edges;
>> >  * all properties are stored in some external structure, which I
>> >   imagine composed of associative maps (Map, etc etc).
>> >
>> > He motivated the idea with a "personal use case": often graphs are used
>> > and reused with the same structure but different weights (and/or labels,
>> > etc). Now if James' question becomes a second use case, maybe it's the
>> > right time to exhume that idea ;)
>> >
>> > Ciao,
>> > Claudio
>> >
>> > --
>> > Claudio Squarcella
>> > PhD student at Roma Tre University
>> > http://www.dia.uniroma3.it/~**squarcel<
>> http://www.dia.uniroma3.it/~squarcel>
>> > http://squarcella.com/
>> >
>> >
>> > --**--**-
>> > To unsubscribe, e-mail: dev-unsubscribe@commons.**apache.org<
>> dev-unsubscr...@commons.apache.org>
>> > For additional commands, e-mail: dev-h...@commons.apache.org
>> >
>> >
>>

-
To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
For additional commands, e-mail: dev-h...@commons.apache.org

Re: [graph] Why the Vertex and Edge interfaces?

[Simone Tripodi]

of course, in that situation Vertex/Weight are just useless marker :P

-Simo

http://people.apache.org/~simonetripodi/
http://simonetripodi.livejournal.com/
http://twitter.com/simonetripodi
http://www.99soft.org/



On Fri, Mar 2, 2012 at 10:14 PM, Simone Tripodi
 wrote:
> Hi James,
>
> looking at the the code I think we can safety keep the Weighted*
> version AND use external functions you described. It is the same
> Comparable/Comparator pattern, WDYT?
>
> best,
> -Simo
>
> http://people.apache.org/~simonetripodi/
> http://simonetripodi.livejournal.com/
> http://twitter.com/simonetripodi
> http://www.99soft.org/
>
>
>
> On Fri, Mar 2, 2012 at 10:09 PM, James Carman
>  wrote:
>> My point isn't that we would not need weights ever but we're putting a
>> needless restriction on the users (implementing a marker interface).  The
>> Edge and Vertex interfaces add no value.
>>
>> The weights can be external, too.  It's only a function from edge to
>> weight.  Your algorithm can take a function for its weights.  The files
>> library does it similar to this.
>> On Mar 2, 2012 3:08 PM, "Ted Dunning"  wrote:
>>
>>> Having weights on vertices is quite common.  Consider any probability
>>> transition network.  The weight on each node is the probability of being in
>>> that state and the weights on the edges are conditional probabilties.
>>>
>>> Page rank is a related example of having weights on nodes.
>>>
>>> On Fri, Mar 2, 2012 at 12:40 AM, Claudio Squarcella <
>>> squar...@dia.uniroma3.it> wrote:
>>>
>>> > Hi all,
>>> >
>>> >  Claudio is aware also about algorithms where weights are associated to
>>> >> Vertex - he's preparing his PhD research on graphes - maybe he can
>>> >> show us a more long-vision roadmap and evaluate benefits on
>>> >> simplifying the design.
>>> >>
>>> >
>>> > yes there are algorithms with weights on vertices. Of course those with
>>> > weighted edges (like the ones already implemented) are much more
>>> widespread
>>> > and frequently used, but still we cannot forget about that. Also,
>>> although
>>> > on a secondary level, labels on vertices/edges are kind of important in
>>> > many situations (including testing, debugging) where I think it is good
>>> to
>>> > keep them distinct from the standard "toString" method (you might want to
>>> > represent only a subset of info in the label, etc).
>>> >
>>> > Matthew Pocock suggested an alternative approach back in the days of
>>> > weight abstraction:
>>> >
>>> >  * the graph itself is extremely simple and naked: no weights/labels on
>>> >   vertices/edges;
>>> >  * all properties are stored in some external structure, which I
>>> >   imagine composed of associative maps (Map, etc etc).
>>> >
>>> > He motivated the idea with a "personal use case": often graphs are used
>>> > and reused with the same structure but different weights (and/or labels,
>>> > etc). Now if James' question becomes a second use case, maybe it's the
>>> > right time to exhume that idea ;)
>>> >
>>> > Ciao,
>>> > Claudio
>>> >
>>> > --
>>> > Claudio Squarcella
>>> > PhD student at Roma Tre University
>>> > http://www.dia.uniroma3.it/~**squarcel<
>>> http://www.dia.uniroma3.it/~squarcel>
>>> > http://squarcella.com/
>>> >
>>> >
>>> > --**--**-
>>> > To unsubscribe, e-mail: dev-unsubscribe@commons.**apache.org<
>>> dev-unsubscr...@commons.apache.org>
>>> > For additional commands, e-mail: dev-h...@commons.apache.org
>>> >
>>> >
>>>

-
To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
For additional commands, e-mail: dev-h...@commons.apache.org

Re: [graph] Why the Vertex and Edge interfaces?

[Claudio Squarcella]

Hi,


The weights can be external, too.  It's only a function from edge to
weight.  Your algorithm can take a function for its weights.  The files
library does it similar to this.


what if that mapping function becomes a responsibility of WeightedGraph 
itself? And more generally, what if any property of vertices and/or 
edges is moved to the containing graph?


We could externalize all different graph properties to appropriate 
interfaces (HasWeightsOnEdges, HasLabelsOnVertices, etc) and then each 
algorithm specifies the needed input graph including the subset of 
interfaces it needs to implement. We do something like that with weight 
operations already.


Claudio



On Mar 2, 2012 3:08 PM, "Ted Dunning"  wrote:


Having weights on vertices is quite common.  Consider any probability
transition network.  The weight on each node is the probability of being in
that state and the weights on the edges are conditional probabilties.

Page rank is a related example of having weights on nodes.

On Fri, Mar 2, 2012 at 12:40 AM, Claudio Squarcella<
squar...@dia.uniroma3.it>  wrote:


Hi all,

  Claudio is aware also about algorithms where weights are associated to

Vertex - he's preparing his PhD research on graphes - maybe he can
show us a more long-vision roadmap and evaluate benefits on
simplifying the design.


yes there are algorithms with weights on vertices. Of course those with
weighted edges (like the ones already implemented) are much more

widespread

and frequently used, but still we cannot forget about that. Also,

although

on a secondary level, labels on vertices/edges are kind of important in
many situations (including testing, debugging) where I think it is good

to

keep them distinct from the standard "toString" method (you might want to
represent only a subset of info in the label, etc).

Matthew Pocock suggested an alternative approach back in the days of
weight abstraction:

  * the graph itself is extremely simple and naked: no weights/labels on
   vertices/edges;
  * all properties are stored in some external structure, which I
   imagine composed of associative maps (Map, etc etc).

He motivated the idea with a "personal use case": often graphs are used
and reused with the same structure but different weights (and/or labels,
etc). Now if James' question becomes a second use case, maybe it's the
right time to exhume that idea ;)

Ciao,
Claudio

--
Claudio Squarcella
PhD student at Roma Tre University
http://www.dia.uniroma3.it/~**squarcel<

http://www.dia.uniroma3.it/~squarcel>

http://squarcella.com/


--**--**-
To unsubscribe, e-mail: dev-unsubscribe@commons.**apache.org<

dev-unsubscr...@commons.apache.org>

For additional commands, e-mail: dev-h...@commons.apache.org




--
Claudio Squarcella
PhD student at Roma Tre University
http://www.dia.uniroma3.it/~squarcel
http://squarcella.com/


-
To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
For additional commands, e-mail: dev-h...@commons.apache.org

Re: [graph] Why the Vertex and Edge interfaces?

[Simone Tripodi]

Hola,

> what if that mapping function becomes a responsibility of WeightedGraph
> itself?
> And more generally, what if any property of vertices and/or edges is
> moved to the containing graph?
>

that would imply that Graph implementations have to implement vertices
and/or edges metadata indexing, that would be anyway less performant
than accessing directly on metadata contained in current node/arc -
just count the number of time you should have to touch the adapted
data structures, of course will be at least one more than the actual.

> We could externalize all different graph properties to appropriate
> interfaces (HasWeightsOnEdges, HasLabelsOnVertices, etc) and then each
> algorithm specifies the needed input graph including the subset of
> interfaces it needs to implement. We do something like that with weight
> operations already.

I am worried that with that approach the number of interfaces would
proliferate like pollen during Spring, users - and devs - would get
easily lost

weights are something already complicated for being a simple concept,
please apologize for the little offtopic:

Zero, name of an element, contains `zero` method to get the zero (it
is still confusing to me), Monoid  extends Zero and Semigroup - given
the use inside graph math, Zero#zero and Semigroup#append can be moved
directly to Monoid and rename it as WeightOperation - does it remind
you something? :P

best,
-Simo

http://people.apache.org/~simonetripodi/
http://simonetripodi.livejournal.com/
http://twitter.com/simonetripodi
http://www.99soft.org/



On Fri, Mar 2, 2012 at 10:22 PM, Claudio Squarcella
 wrote:
> Hi,
>
>
>> The weights can be external, too.  It's only a function from edge to
>> weight.  Your algorithm can take a function for its weights.  The files
>> library does it similar to this.
>
>
> what if that mapping function becomes a responsibility of WeightedGraph
> itself? And more generally, what if any property of vertices and/or edges is
> moved to the containing graph?
>
> We could externalize all different graph properties to appropriate
> interfaces (HasWeightsOnEdges, HasLabelsOnVertices, etc) and then each
> algorithm specifies the needed input graph including the subset of
> interfaces it needs to implement. We do something like that with weight
> operations already.
>
> Claudio
>
>
>
>> On Mar 2, 2012 3:08 PM, "Ted Dunning"  wrote:
>>
>>> Having weights on vertices is quite common.  Consider any probability
>>> transition network.  The weight on each node is the probability of being
>>> in
>>> that state and the weights on the edges are conditional probabilties.
>>>
>>> Page rank is a related example of having weights on nodes.
>>>
>>> On Fri, Mar 2, 2012 at 12:40 AM, Claudio Squarcella<
>>> squar...@dia.uniroma3.it>  wrote:
>>>
 Hi all,

  Claudio is aware also about algorithms where weights are associated to
>
> Vertex - he's preparing his PhD research on graphes - maybe he can
> show us a more long-vision roadmap and evaluate benefits on
> simplifying the design.
>
 yes there are algorithms with weights on vertices. Of course those with
 weighted edges (like the ones already implemented) are much more
>>>
>>> widespread

 and frequently used, but still we cannot forget about that. Also,
>>>
>>> although

 on a secondary level, labels on vertices/edges are kind of important in
 many situations (including testing, debugging) where I think it is good
>>>
>>> to

 keep them distinct from the standard "toString" method (you might want
 to
 represent only a subset of info in the label, etc).

 Matthew Pocock suggested an alternative approach back in the days of
 weight abstraction:

  * the graph itself is extremely simple and naked: no weights/labels on
   vertices/edges;
  * all properties are stored in some external structure, which I
   imagine composed of associative maps (Map, etc etc).

 He motivated the idea with a "personal use case": often graphs are used
 and reused with the same structure but different weights (and/or labels,
 etc). Now if James' question becomes a second use case, maybe it's the
 right time to exhume that idea ;)

 Ciao,
 Claudio

 --
 Claudio Squarcella
 PhD student at Roma Tre University
 http://www.dia.uniroma3.it/~**squarcel<
>>>
>>> http://www.dia.uniroma3.it/~squarcel>

 http://squarcella.com/



 --**--**-
 To unsubscribe, e-mail: dev-unsubscribe@commons.**apache.org<
>>>
>>> dev-unsubscr...@commons.apache.org>

 For additional commands, e-mail: dev-h...@commons.apache.org


>
> --
> Claudio Squarcella
> PhD student at Roma Tre University
> http://www.dia.uniroma3.it/~squarcel
> http://squarcella.com/
>
>
> -
> To unsubscribe, e-mail: 

Re: [graph] Why the Vertex and Edge interfaces?

[Simone Tripodi]

Anyway, nothing has to prevent you to make a prototype and propose a patch ;)
-Simo

http://people.apache.org/~simonetripodi/
http://simonetripodi.livejournal.com/
http://twitter.com/simonetripodi
http://www.99soft.org/



On Fri, Mar 2, 2012 at 11:02 PM, Simone Tripodi
 wrote:
> Hola,
>
>> what if that mapping function becomes a responsibility of WeightedGraph
>> itself?
>> And more generally, what if any property of vertices and/or edges is
>> moved to the containing graph?
>>
>
> that would imply that Graph implementations have to implement vertices
> and/or edges metadata indexing, that would be anyway less performant
> than accessing directly on metadata contained in current node/arc -
> just count the number of time you should have to touch the adapted
> data structures, of course will be at least one more than the actual.
>
>> We could externalize all different graph properties to appropriate
>> interfaces (HasWeightsOnEdges, HasLabelsOnVertices, etc) and then each
>> algorithm specifies the needed input graph including the subset of
>> interfaces it needs to implement. We do something like that with weight
>> operations already.
>
> I am worried that with that approach the number of interfaces would
> proliferate like pollen during Spring, users - and devs - would get
> easily lost
>
> weights are something already complicated for being a simple concept,
> please apologize for the little offtopic:
>
> Zero, name of an element, contains `zero` method to get the zero (it
> is still confusing to me), Monoid  extends Zero and Semigroup - given
> the use inside graph math, Zero#zero and Semigroup#append can be moved
> directly to Monoid and rename it as WeightOperation - does it remind
> you something? :P
>
> best,
> -Simo
>
> http://people.apache.org/~simonetripodi/
> http://simonetripodi.livejournal.com/
> http://twitter.com/simonetripodi
> http://www.99soft.org/
>
>
>
> On Fri, Mar 2, 2012 at 10:22 PM, Claudio Squarcella
>  wrote:
>> Hi,
>>
>>
>>> The weights can be external, too.  It's only a function from edge to
>>> weight.  Your algorithm can take a function for its weights.  The files
>>> library does it similar to this.
>>
>>
>> what if that mapping function becomes a responsibility of WeightedGraph
>> itself? And more generally, what if any property of vertices and/or edges is
>> moved to the containing graph?
>>
>> We could externalize all different graph properties to appropriate
>> interfaces (HasWeightsOnEdges, HasLabelsOnVertices, etc) and then each
>> algorithm specifies the needed input graph including the subset of
>> interfaces it needs to implement. We do something like that with weight
>> operations already.
>>
>> Claudio
>>
>>
>>
>>> On Mar 2, 2012 3:08 PM, "Ted Dunning"  wrote:
>>>
 Having weights on vertices is quite common.  Consider any probability
 transition network.  The weight on each node is the probability of being
 in
 that state and the weights on the edges are conditional probabilties.

 Page rank is a related example of having weights on nodes.

 On Fri, Mar 2, 2012 at 12:40 AM, Claudio Squarcella<
 squar...@dia.uniroma3.it>  wrote:

> Hi all,
>
>  Claudio is aware also about algorithms where weights are associated to
>>
>> Vertex - he's preparing his PhD research on graphes - maybe he can
>> show us a more long-vision roadmap and evaluate benefits on
>> simplifying the design.
>>
> yes there are algorithms with weights on vertices. Of course those with
> weighted edges (like the ones already implemented) are much more

 widespread
>
> and frequently used, but still we cannot forget about that. Also,

 although
>
> on a secondary level, labels on vertices/edges are kind of important in
> many situations (including testing, debugging) where I think it is good

 to
>
> keep them distinct from the standard "toString" method (you might want
> to
> represent only a subset of info in the label, etc).
>
> Matthew Pocock suggested an alternative approach back in the days of
> weight abstraction:
>
>  * the graph itself is extremely simple and naked: no weights/labels on
>   vertices/edges;
>  * all properties are stored in some external structure, which I
>   imagine composed of associative maps (Map, etc etc).
>
> He motivated the idea with a "personal use case": often graphs are used
> and reused with the same structure but different weights (and/or labels,
> etc). Now if James' question becomes a second use case, maybe it's the
> right time to exhume that idea ;)
>
> Ciao,
> Claudio
>
> --
> Claudio Squarcella
> PhD student at Roma Tre University
> http://www.dia.uniroma3.it/~**squarcel<

 http://www.dia.uniroma3.it/~squarcel>
>
> http://squarcella.com/
>
>
>
> --**-

Re: [graph] Why the Vertex and Edge interfaces?

[James Carman]

This is what the asf is all about, man.  I like it when we have projects
like this getting a lot of traffic.  It's great to see so many folks
getting involved.  Thanks, guys!  This is what keeps me coming back!

Sent from tablet device.  Please excuse typos and brevity.
On Mar 2, 2012 5:50 PM, "Simone Tripodi"  wrote:

> Anyway, nothing has to prevent you to make a prototype and propose a patch
> ;)
> -Simo
>
> http://people.apache.org/~simonetripodi/
> http://simonetripodi.livejournal.com/
> http://twitter.com/simonetripodi
> http://www.99soft.org/
>
>
>
> On Fri, Mar 2, 2012 at 11:02 PM, Simone Tripodi
>  wrote:
> > Hola,
> >
> >> what if that mapping function becomes a responsibility of WeightedGraph
> >> itself?
> >> And more generally, what if any property of vertices and/or edges is
> >> moved to the containing graph?
> >>
> >
> > that would imply that Graph implementations have to implement vertices
> > and/or edges metadata indexing, that would be anyway less performant
> > than accessing directly on metadata contained in current node/arc -
> > just count the number of time you should have to touch the adapted
> > data structures, of course will be at least one more than the actual.
> >
> >> We could externalize all different graph properties to appropriate
> >> interfaces (HasWeightsOnEdges, HasLabelsOnVertices, etc) and then each
> >> algorithm specifies the needed input graph including the subset of
> >> interfaces it needs to implement. We do something like that with weight
> >> operations already.
> >
> > I am worried that with that approach the number of interfaces would
> > proliferate like pollen during Spring, users - and devs - would get
> > easily lost
> >
> > weights are something already complicated for being a simple concept,
> > please apologize for the little offtopic:
> >
> > Zero, name of an element, contains `zero` method to get the zero (it
> > is still confusing to me), Monoid  extends Zero and Semigroup - given
> > the use inside graph math, Zero#zero and Semigroup#append can be moved
> > directly to Monoid and rename it as WeightOperation - does it remind
> > you something? :P
> >
> > best,
> > -Simo
> >
> > http://people.apache.org/~simonetripodi/
> > http://simonetripodi.livejournal.com/
> > http://twitter.com/simonetripodi
> > http://www.99soft.org/
> >
> >
> >
> > On Fri, Mar 2, 2012 at 10:22 PM, Claudio Squarcella
> >  wrote:
> >> Hi,
> >>
> >>
> >>> The weights can be external, too.  It's only a function from edge to
> >>> weight.  Your algorithm can take a function for its weights.  The files
> >>> library does it similar to this.
> >>
> >>
> >> what if that mapping function becomes a responsibility of WeightedGraph
> >> itself? And more generally, what if any property of vertices and/or
> edges is
> >> moved to the containing graph?
> >>
> >> We could externalize all different graph properties to appropriate
> >> interfaces (HasWeightsOnEdges, HasLabelsOnVertices, etc) and then each
> >> algorithm specifies the needed input graph including the subset of
> >> interfaces it needs to implement. We do something like that with weight
> >> operations already.
> >>
> >> Claudio
> >>
> >>
> >>
> >>> On Mar 2, 2012 3:08 PM, "Ted Dunning"  wrote:
> >>>
>  Having weights on vertices is quite common.  Consider any probability
>  transition network.  The weight on each node is the probability of
> being
>  in
>  that state and the weights on the edges are conditional probabilties.
> 
>  Page rank is a related example of having weights on nodes.
> 
>  On Fri, Mar 2, 2012 at 12:40 AM, Claudio Squarcella<
>  squar...@dia.uniroma3.it>  wrote:
> 
> > Hi all,
> >
> >  Claudio is aware also about algorithms where weights are associated
> to
> >>
> >> Vertex - he's preparing his PhD research on graphes - maybe he can
> >> show us a more long-vision roadmap and evaluate benefits on
> >> simplifying the design.
> >>
> > yes there are algorithms with weights on vertices. Of course those
> with
> > weighted edges (like the ones already implemented) are much more
> 
>  widespread
> >
> > and frequently used, but still we cannot forget about that. Also,
> 
>  although
> >
> > on a secondary level, labels on vertices/edges are kind of important
> in
> > many situations (including testing, debugging) where I think it is
> good
> 
>  to
> >
> > keep them distinct from the standard "toString" method (you might
> want
> > to
> > represent only a subset of info in the label, etc).
> >
> > Matthew Pocock suggested an alternative approach back in the days of
> > weight abstraction:
> >
> >  * the graph itself is extremely simple and naked: no weights/labels
> on
> >   vertices/edges;
> >  * all properties are stored in some external structure, which I
> >   imagine composed of associative maps (Map, etc etc).
> >
> > He motiv

Re: [graph] Why the Vertex and Edge interfaces?

[Simone Tripodi]

and I am glad you feel comfortable in [graph] James! we have only to
learn from an experienced guy like you!!
best,
-Simo

http://people.apache.org/~simonetripodi/
http://simonetripodi.livejournal.com/
http://twitter.com/simonetripodi
http://www.99soft.org/



On Sat, Mar 3, 2012 at 1:12 AM, James Carman
 wrote:
> This is what the asf is all about, man.  I like it when we have projects
> like this getting a lot of traffic.  It's great to see so many folks
> getting involved.  Thanks, guys!  This is what keeps me coming back!
>
> Sent from tablet device.  Please excuse typos and brevity.
> On Mar 2, 2012 5:50 PM, "Simone Tripodi"  wrote:
>
>> Anyway, nothing has to prevent you to make a prototype and propose a patch
>> ;)
>> -Simo
>>
>> http://people.apache.org/~simonetripodi/
>> http://simonetripodi.livejournal.com/
>> http://twitter.com/simonetripodi
>> http://www.99soft.org/
>>
>>
>>
>> On Fri, Mar 2, 2012 at 11:02 PM, Simone Tripodi
>>  wrote:
>> > Hola,
>> >
>> >> what if that mapping function becomes a responsibility of WeightedGraph
>> >> itself?
>> >> And more generally, what if any property of vertices and/or edges is
>> >> moved to the containing graph?
>> >>
>> >
>> > that would imply that Graph implementations have to implement vertices
>> > and/or edges metadata indexing, that would be anyway less performant
>> > than accessing directly on metadata contained in current node/arc -
>> > just count the number of time you should have to touch the adapted
>> > data structures, of course will be at least one more than the actual.
>> >
>> >> We could externalize all different graph properties to appropriate
>> >> interfaces (HasWeightsOnEdges, HasLabelsOnVertices, etc) and then each
>> >> algorithm specifies the needed input graph including the subset of
>> >> interfaces it needs to implement. We do something like that with weight
>> >> operations already.
>> >
>> > I am worried that with that approach the number of interfaces would
>> > proliferate like pollen during Spring, users - and devs - would get
>> > easily lost
>> >
>> > weights are something already complicated for being a simple concept,
>> > please apologize for the little offtopic:
>> >
>> > Zero, name of an element, contains `zero` method to get the zero (it
>> > is still confusing to me), Monoid  extends Zero and Semigroup - given
>> > the use inside graph math, Zero#zero and Semigroup#append can be moved
>> > directly to Monoid and rename it as WeightOperation - does it remind
>> > you something? :P
>> >
>> > best,
>> > -Simo
>> >
>> > http://people.apache.org/~simonetripodi/
>> > http://simonetripodi.livejournal.com/
>> > http://twitter.com/simonetripodi
>> > http://www.99soft.org/
>> >
>> >
>> >
>> > On Fri, Mar 2, 2012 at 10:22 PM, Claudio Squarcella
>> >  wrote:
>> >> Hi,
>> >>
>> >>
>> >>> The weights can be external, too.  It's only a function from edge to
>> >>> weight.  Your algorithm can take a function for its weights.  The files
>> >>> library does it similar to this.
>> >>
>> >>
>> >> what if that mapping function becomes a responsibility of WeightedGraph
>> >> itself? And more generally, what if any property of vertices and/or
>> edges is
>> >> moved to the containing graph?
>> >>
>> >> We could externalize all different graph properties to appropriate
>> >> interfaces (HasWeightsOnEdges, HasLabelsOnVertices, etc) and then each
>> >> algorithm specifies the needed input graph including the subset of
>> >> interfaces it needs to implement. We do something like that with weight
>> >> operations already.
>> >>
>> >> Claudio
>> >>
>> >>
>> >>
>> >>> On Mar 2, 2012 3:08 PM, "Ted Dunning"  wrote:
>> >>>
>>  Having weights on vertices is quite common.  Consider any probability
>>  transition network.  The weight on each node is the probability of
>> being
>>  in
>>  that state and the weights on the edges are conditional probabilties.
>> 
>>  Page rank is a related example of having weights on nodes.
>> 
>>  On Fri, Mar 2, 2012 at 12:40 AM, Claudio Squarcella<
>>  squar...@dia.uniroma3.it>  wrote:
>> 
>> > Hi all,
>> >
>> >  Claudio is aware also about algorithms where weights are associated
>> to
>> >>
>> >> Vertex - he's preparing his PhD research on graphes - maybe he can
>> >> show us a more long-vision roadmap and evaluate benefits on
>> >> simplifying the design.
>> >>
>> > yes there are algorithms with weights on vertices. Of course those
>> with
>> > weighted edges (like the ones already implemented) are much more
>> 
>>  widespread
>> >
>> > and frequently used, but still we cannot forget about that. Also,
>> 
>>  although
>> >
>> > on a secondary level, labels on vertices/edges are kind of important
>> in
>> > many situations (including testing, debugging) where I think it is
>> good
>> 
>>  to
>> >
>> > keep them distinct from the standard "toString" method (you might
>> want
>> > to

Re: [graph] Why the Vertex and Edge interfaces?

[Claudio Squarcella]

Hi,

first of all: yes, I will play with this stuff as soon as I find the time :)


what if that mapping function becomes a responsibility of WeightedGraph
itself?
And more generally, what if any property of vertices and/or edges is
moved to the containing graph?

that would imply that Graph implementations have to implement vertices
and/or edges metadata indexing, that would be anyway less performant
than accessing directly on metadata contained in current node/arc -
just count the number of time you should have to touch the adapted
data structures, of course will be at least one more than the actual.


that is absolutely right. Not asymptotically if the implementation is 
good (hashmaps are already good enough with their read time which is 
basically constant), but still there is one more step of indirection. So 
we would need to test and compare performances, hopefully with 
acceptable results.



We could externalize all different graph properties to appropriate
interfaces (HasWeightsOnEdges, HasLabelsOnVertices, etc) and then each
algorithm specifies the needed input graph including the subset of
interfaces it needs to implement. We do something like that with weight
operations already.

I am worried that with that approach the number of interfaces would
proliferate like pollen during Spring, users - and devs - would get
easily lost


but that would happen anyway as soon as we implement an algorithm with 
weights on vertices, right? Here are the options I see:


 * with the current approach: one interface for edge-weighted graphs
   (EdgeWeightedGraph, renaming the current WeightedGraph?), one for
   vertex-weighted graphs (VertexWeightedGraph) and maybe even one for
   weights on both edges and vertices (EdgeAndVertexWeightedGraph?) --
   not to talk about their counterparts with labels, etc;
 * with the proposed approach: a Graph would implement
   HasWeightsOnEdges and/or HasWeightsOnVertices -- and maybe also
   HasLabelsOnEdges if needed.



weights are something already complicated for being a simple concept,
please apologize for the little offtopic:

Zero, name of an element, contains `zero` method to get the zero (it
is still confusing to me), Monoid  extends Zero and Semigroup - given
the use inside graph math, Zero#zero and Semigroup#append can be moved
directly to Monoid and rename it as WeightOperation - does it remind
you something? :P


I can agree with most of what you say but I would still call the result 
Monoid, or maybe even better Addition -- because that is what it is, a 
Monoid representing the sum operation. "WeightOperation" sounds more 
like a general "container" which can include Addition, Comparator, and 
maybe in the near future Multiplication or who knows what -- which again 
is pretty much what happens now with the wrappers for Double, Integer, etc.


I know that this kind of "mismatch" is not your favourite ;) but would 
you really call "Operations" something which is just an "Addition" -- or 
viceversa "DoubleWeightAddition" something that might later be expanded 
with other operations?


Ciao and thanks for your feedback!
Claudio



best,
-Simo

http://people.apache.org/~simonetripodi/
http://simonetripodi.livejournal.com/
http://twitter.com/simonetripodi
http://www.99soft.org/



On Fri, Mar 2, 2012 at 10:22 PM, Claudio Squarcella
  wrote:

Hi,



The weights can be external, too.  It's only a function from edge to
weight.  Your algorithm can take a function for its weights.  The files
library does it similar to this.


what if that mapping function becomes a responsibility of WeightedGraph
itself? And more generally, what if any property of vertices and/or edges is
moved to the containing graph?

We could externalize all different graph properties to appropriate
interfaces (HasWeightsOnEdges, HasLabelsOnVertices, etc) and then each
algorithm specifies the needed input graph including the subset of
interfaces it needs to implement. We do something like that with weight
operations already.

Claudio




On Mar 2, 2012 3:08 PM, "Ted Dunning"wrote:


Having weights on vertices is quite common.  Consider any probability
transition network.  The weight on each node is the probability of being
in
that state and the weights on the edges are conditional probabilties.

Page rank is a related example of having weights on nodes.

On Fri, Mar 2, 2012 at 12:40 AM, Claudio Squarcella<
squar...@dia.uniroma3.it>wrote:


Hi all,

  Claudio is aware also about algorithms where weights are associated to

Vertex - he's preparing his PhD research on graphes - maybe he can
show us a more long-vision roadmap and evaluate benefits on
simplifying the design.


yes there are algorithms with weights on vertices. Of course those with
weighted edges (like the ones already implemented) are much more

widespread

and frequently used, but still we cannot forget about that. Also,

although

on a secondary level, labels on vertices/edges are kind of important in
many situations (including test

Re: [graph] Why the Vertex and Edge interfaces?

[Simone Tripodi]

> first of all: yes, I will play with this stuff as soon as I find the time :)

this is scaring... go Orb.io, go! :)

> constant), but still there is one more step of indirection. So we would need
> to test and compare performances, hopefully with acceptable results.

sounds it would be better if you implement that stuff in a branch to
keep both implementations, IMHO

>  * with the current approach: one interface for edge-weighted graphs
>   (EdgeWeightedGraph, renaming the current WeightedGraph?), one for
>   vertex-weighted graphs (VertexWeightedGraph) and maybe even one for
>   weights on both edges and vertices (EdgeAndVertexWeightedGraph?) --
>   not to talk about their counterparts with labels, etc;
>  * with the proposed approach: a Graph would implement
>   HasWeightsOnEdges and/or HasWeightsOnVertices -- and maybe also
>   HasLabelsOnEdges if needed.

do you remember that we reintroduced the WeightedGraph just for the
type inference issue on fluent APIs in Eclipse, do you? ;) It would
have been worked otherwise as well dropping the WeightedGraph and
expressing types only on Vertex/Edges

> I can agree with most of what you say but I would still call the result
> Monoid, or maybe even better Addition -- because that is what it is, a
> Monoid representing the sum operation. "WeightOperation" sounds more like a
> general "container" which can include Addition, Comparator, and maybe in the
> near future Multiplication or who knows what -- which again is pretty much
> what happens now with the wrappers for Double, Integer, etc.
>

you cheater always know how to confuse me :P

> I know that this kind of "mismatch" is not your favourite ;) but would you
> really call "Operations" something which is just an "Addition" -- or
> viceversa "DoubleWeightAddition" something that might later be expanded with
> other operations?

I am confused now: this is what you did in the concrete implementation!

time to sleep, cannot reply anymore, read tomorrow

-Simo

http://people.apache.org/~simonetripodi/
http://simonetripodi.livejournal.com/
http://twitter.com/simonetripodi
http://www.99soft.org/



On Sat, Mar 3, 2012 at 1:37 AM, Claudio Squarcella
 wrote:
> Hi,
>

>
>
>>> what if that mapping function becomes a responsibility of WeightedGraph
>>> itself?
>>> And more generally, what if any property of vertices and/or edges is
>>> moved to the containing graph?
>>
>> that would imply that Graph implementations have to implement vertices
>> and/or edges metadata indexing, that would be anyway less performant
>> than accessing directly on metadata contained in current node/arc -
>> just count the number of time you should have to touch the adapted
>> data structures, of course will be at least one more than the actual.
>
>
> that is absolutely right. Not asymptotically if the implementation is good
> (hashmaps are already good enough with their read time which is basically
> constant), but still there is one more step of indirection. So we would need
> to test and compare performances, hopefully with acceptable results.
>
>
>>> We could externalize all different graph properties to appropriate
>>> interfaces (HasWeightsOnEdges, HasLabelsOnVertices, etc) and then each
>>> algorithm specifies the needed input graph including the subset of
>>> interfaces it needs to implement. We do something like that with weight
>>> operations already.
>>
>> I am worried that with that approach the number of interfaces would
>> proliferate like pollen during Spring, users - and devs - would get
>> easily lost
>
>
> but that would happen anyway as soon as we implement an algorithm with
> weights on vertices, right? Here are the options I see:
>
>  * with the current approach: one interface for edge-weighted graphs
>   (EdgeWeightedGraph, renaming the current WeightedGraph?), one for
>   vertex-weighted graphs (VertexWeightedGraph) and maybe even one for
>   weights on both edges and vertices (EdgeAndVertexWeightedGraph?) --
>   not to talk about their counterparts with labels, etc;
>  * with the proposed approach: a Graph would implement
>   HasWeightsOnEdges and/or HasWeightsOnVertices -- and maybe also
>   HasLabelsOnEdges if needed.
>
>
>
>> weights are something already complicated for being a simple concept,
>> please apologize for the little offtopic:
>>
>> Zero, name of an element, contains `zero` method to get the zero (it
>> is still confusing to me), Monoid  extends Zero and Semigroup - given
>> the use inside graph math, Zero#zero and Semigroup#append can be moved
>> directly to Monoid and rename it as WeightOperation - does it remind
>> you something? :P
>
>
> I can agree with most of what you say but I would still call the result
> Monoid, or maybe even better Addition -- because that is what it is, a
> Monoid representing the sum operation. "WeightOperation" sounds more like a
> general "container" which can include Addition, Comparator, and maybe in the
> near future Multiplication or who knows what -- which again is prett

Re: [graph] Why the Vertex and Edge interfaces?

[Claudio Squarcella]

Hi,

On 03/03/2012 02:21, Simone Tripodi wrote:

first of all: yes, I will play with this stuff as soon as I find the time :)

this is scaring... go Orb.io, go! :)


constant), but still there is one more step of indirection. So we would need
to test and compare performances, hopefully with acceptable results.

sounds it would be better if you implement that stuff in a branch to
keep both implementations, IMHO


sure :)


  * with the current approach: one interface for edge-weighted graphs
   (EdgeWeightedGraph, renaming the current WeightedGraph?), one for
   vertex-weighted graphs (VertexWeightedGraph) and maybe even one for
   weights on both edges and vertices (EdgeAndVertexWeightedGraph?) --
   not to talk about their counterparts with labels, etc;
  * with the proposed approach: a Graph would implement
   HasWeightsOnEdges and/or HasWeightsOnVertices -- and maybe also
   HasLabelsOnEdges if needed.

do you remember that we reintroduced the WeightedGraph just for the
type inference issue on fluent APIs in Eclipse, do you? ;) It would
have been worked otherwise as well dropping the WeightedGraph and
expressing types only on Vertex/Edges


that is right. On the other hand with "HasWeightedEdges" we could drop 
"WeightedEdge" and so on: one interface in, one out.


Or we could even save both approaches as alternative implementations. 
That is:


 * a set of interfaces: e.g. HasWeightedEdges#getWeight(edge),
   HasWeightedVertices#getWeight(vertex), etc.
 * #1 implementation with external properties: the graph keeps the
   mapping between edge/vertex and weight, so that any type can be used
   for edges/vertices
 * #2 classical implementation: we keep markers like WeightedEdge and
   WeightedVertex but only the graph knows about them. algorithms keep
   asking the graph for weights of edges/vertices, and the graph in
   turn asks the edge/vertex itself (passed as parameter).

WDYT?


I know that this kind of "mismatch" is not your favourite ;) but would you
really call "Operations" something which is just an "Addition" -- or
viceversa "DoubleWeightAddition" something that might later be expanded with
other operations?

I am confused now: this is what you did in the concrete implementation!


no, trying to be clearer: you propose to rename Monoid into 
WeightOperation, which is like renaming Addition into Operation. Or 
alternatively to call the current *WeightBaseOperations something like 
*WeightMonoid. In both cases I disagree because I would prefer to keep a 
clear distinction between single well-defined properties/operations 
(like Addition or Comparator) and the comprehensive implementation (e.g. 
DoubleWeightBaseOperations) that implements all the operations it can 
implement with Doubles.


Hoping we'll converge somewhere, maybe asymptotically ;)
Claudio



time to sleep, cannot reply anymore, read tomorrow

-Simo

http://people.apache.org/~simonetripodi/
http://simonetripodi.livejournal.com/
http://twitter.com/simonetripodi
http://www.99soft.org/



On Sat, Mar 3, 2012 at 1:37 AM, Claudio Squarcella
  wrote:

Hi,



what if that mapping function becomes a responsibility of WeightedGraph
itself?
And more generally, what if any property of vertices and/or edges is
moved to the containing graph?

that would imply that Graph implementations have to implement vertices
and/or edges metadata indexing, that would be anyway less performant
than accessing directly on metadata contained in current node/arc -
just count the number of time you should have to touch the adapted
data structures, of course will be at least one more than the actual.


that is absolutely right. Not asymptotically if the implementation is good
(hashmaps are already good enough with their read time which is basically
constant), but still there is one more step of indirection. So we would need
to test and compare performances, hopefully with acceptable results.



We could externalize all different graph properties to appropriate
interfaces (HasWeightsOnEdges, HasLabelsOnVertices, etc) and then each
algorithm specifies the needed input graph including the subset of
interfaces it needs to implement. We do something like that with weight
operations already.

I am worried that with that approach the number of interfaces would
proliferate like pollen during Spring, users - and devs - would get
easily lost


but that would happen anyway as soon as we implement an algorithm with
weights on vertices, right? Here are the options I see:

  * with the current approach: one interface for edge-weighted graphs
   (EdgeWeightedGraph, renaming the current WeightedGraph?), one for
   vertex-weighted graphs (VertexWeightedGraph) and maybe even one for
   weights on both edges and vertices (EdgeAndVertexWeightedGraph?) --
   not to talk about their counterparts with labels, etc;
  * with the proposed approach: a Graph would implement
   HasWeightsOnEdges and/or HasWeightsOnVertices -- and maybe also
   HasLabelsOnEdges if needed.




weights are somet

Re: [graph] Why the Vertex and Edge interfaces?

[Simone Tripodi]

Hola Claud.io,

>  * a set of interfaces: e.g. HasWeightedEdges#getWeight(edge),
>   HasWeightedVertices#getWeight(vertex), etc.
>  * #1 implementation with external properties: the graph keeps the
>   mapping between edge/vertex and weight, so that any type can be used
>   for edges/vertices
>  * #2 classical implementation: we keep markers like WeightedEdge and
>   WeightedVertex but only the graph knows about them. algorithms keep
>   asking the graph for weights of edges/vertices, and the graph in
>   turn asks the edge/vertex itself (passed as parameter).
>
> WDYT?

Apologize but I still don't understand what the benefit is on storing
nodes/arcs data in the Graph data structure - sounds to me like a
Collection where, to know the int value of its elements, have
to query the data structure, i.e.

Collection integerCollection = ...;
for ( Integer ptr : integerCollection )
{
int value = integerCollection.getInt( ptr );
}

It is weird to me even reading it.

When I started modeling Graph, I had collections in mind - above all
to simplify its adoption - I would be more than pleased to drop
Weighted* version of graphs and come back to the previous situation,
but with the annoying  type inference issue fixed.

> no, trying to be clearer: you propose to rename Monoid into WeightOperation,
> which is like renaming Addition into Operation. Or alternatively to call the
> current *WeightBaseOperations something like *WeightMonoid. In both cases I
> disagree because I would prefer to keep a clear distinction between single
> well-defined properties/operations (like Addition or Comparator) and the
> comprehensive implementation (e.g. DoubleWeightBaseOperations) that
> implements all the operations it can implement with Doubles.

OK, concept is clear, I still don't agree on the nomenclature, anyway.
Actually *WeightBaseOperations describe
*WeightAdditionalOrderedMonoid, so *BaseOperations doesn't sound self
explaining.

Moreover, The Zero interface is the *additional* monoid identity
element, if someone has to implement the Multiplication Monoid I
wouldn't expect he implements the One interface wich declares O one()
method.
This is why IMHO in the current algebra model, Zero has to be dropped
and has to be modified in:

/**
 * semigroup is an algebraic structure consisting of a set together
with an associative binary operation.
 */
interface Semigroup
{

E append( E s1, E s2 );

}

/**
 * A {@link Monoid} is a {@link Semigroup} with an identity value.
 */
public interface Monoid
extends Semigroup
{

   E identity();

   E inverse( E element );

}

that would continue working for every Monoid specialization. Or not? thoughts?
Thanks and best,
-Simo

http://people.apache.org/~simonetripodi/
http://simonetripodi.livejournal.com/
http://twitter.com/simonetripodi
http://www.99soft.org/



On Sat, Mar 3, 2012 at 1:43 PM, Claudio Squarcella
 wrote:
> Hi,
>
>
> On 03/03/2012 02:21, Simone Tripodi wrote:
>>>
>>> first of all: yes, I will play with this stuff as soon as I find the time
>>> :)
>>
>> this is scaring... go Orb.io, go! :)
>>
>>> constant), but still there is one more step of indirection. So we would
>>> need
>>> to test and compare performances, hopefully with acceptable results.
>>
>> sounds it would be better if you implement that stuff in a branch to
>> keep both implementations, IMHO
>
>
> sure :)
>
>
>>>  * with the current approach: one interface for edge-weighted graphs
>>>   (EdgeWeightedGraph, renaming the current WeightedGraph?), one for
>>>   vertex-weighted graphs (VertexWeightedGraph) and maybe even one for
>>>   weights on both edges and vertices (EdgeAndVertexWeightedGraph?) --
>>>   not to talk about their counterparts with labels, etc;
>>>  * with the proposed approach: a Graph would implement
>>>   HasWeightsOnEdges and/or HasWeightsOnVertices -- and maybe also
>>>   HasLabelsOnEdges if needed.
>>
>> do you remember that we reintroduced the WeightedGraph just for the
>> type inference issue on fluent APIs in Eclipse, do you? ;) It would
>> have been worked otherwise as well dropping the WeightedGraph and
>> expressing types only on Vertex/Edges
>
>
> that is right. On the other hand with "HasWeightedEdges" we could drop
> "WeightedEdge" and so on: one interface in, one out.
>
> Or we could even save both approaches as alternative implementations. That
> is:
>
>  * a set of interfaces: e.g. HasWeightedEdges#getWeight(edge),
>   HasWeightedVertices#getWeight(vertex), etc.
>  * #1 implementation with external properties: the graph keeps the
>   mapping between edge/vertex and weight, so that any type can be used
>   for edges/vertices
>  * #2 classical implementation: we keep markers like WeightedEdge and
>   WeightedVertex but only the graph knows about them. algorithms keep
>   asking the graph for weights of edges/vertices, and the graph in
>   turn asks the edge/vertex itself (passed as parameter).
>
> WDYT?
>
>
>>> I know that this kind of "mismatch" is not your favourite ;) but wou

Re: [graph] Why the Vertex and Edge interfaces?

[Claudio Squarcella]

Hi,


Apologize but I still don't understand what the benefit is on storing
nodes/arcs data in the Graph data structure


back to James' first email: any type could be immediately used as 
edge/vertex without wrappers, while specific data related to the domain 
of graphs (weights, labels) would be handled separately. I actually 
think this is useful: back to my days of "DIY Java graphs" I implemented 
something similar to what we have now, just to think every time: "why 
should I wrap my objects with these markers? I already know my Router is 
a Vertex in the graph..."



- sounds to me like a
Collection  where, to know the int value of its elements, have
to query the data structure, i.e.

Collection  integerCollection = ...;
for ( Integer ptr : integerCollection )
{
 int value = integerCollection.getInt( ptr );
}

It is weird to me even reading it.

When I started modeling Graph, I had collections in mind - above all
to simplify its adoption - I would be more than pleased to drop
Weighted* version of graphs and come back to the previous situation,
but with the annoying  type inference issue fixed.


Here's the way I see it. A Graph implementing HasWeightedEdges 
would have something like this inside:


Map edgeWeights = new HashMap();

[... populate map, other methods, etc ...]

// implementing HasWeightedEdges#getEdgeWeight
public W getEdgeWeight(E edge)
{
[... check null etc...]
return edgeWeights.get(edge);
}


no, trying to be clearer: you propose to rename Monoid into WeightOperation,
which is like renaming Addition into Operation. Or alternatively to call the
current *WeightBaseOperations something like *WeightMonoid. In both cases I
disagree because I would prefer to keep a clear distinction between single
well-defined properties/operations (like Addition or Comparator) and the
comprehensive implementation (e.g. DoubleWeightBaseOperations) that
implements all the operations it can implement with Doubles.

OK, concept is clear, I still don't agree on the nomenclature, anyway.
Actually *WeightBaseOperations describe
*WeightAdditionalOrderedMonoid, so *BaseOperations doesn't sound self
explaining.


then let's find a better name, but why *OrderedMonoid? Assume we replace 
the whole set of current interfaces (see my comment to the next 
paragraph below) with just Addition and Comparable (the latter already 
exists of course). There is no need to create another interface to merge 
the two (ComparableAddition? OrderedAddition?). Then we have:


public class DoubleWeightOperations
implements Addition, Comparator
{
[...]
}

I would not rename the class to DoubleWeightAddition or even 
DoubleWeightComparableAddition. What if later we need to e.g. add a 
function that normalizes weights by some factor, or returns the 
reciprocal of a weight, etc? With the above code it would suffice to add 
new interfaces:


public class DoubleWeightOperations
implements Addition, Comparator, Normalization, Reciprocal
{
[...]
}




Moreover, The Zero interface is the *additional* monoid identity
element, if someone has to implement the Multiplication Monoid I
wouldn't expect he implements the One interface wich declares O one()
method.
This is why IMHO in the current algebra model, Zero has to be dropped
and has to be modified in:


That is fine for me. I simply followed pure theory while implementing 
that and used all possible granularity. Now looking at our 
implementations I think it is save enough to even merge Zero, Semigroup 
and Monoid (so that's one step further than your example below) and call 
the result Addition so that its role is clear to everybody. Does that 
sound good? :)


Claudio



/**
  * semigroup is an algebraic structure consisting of a set together
with an associative binary operation.
  */
interface Semigroup
{

 E append( E s1, E s2 );

}

/**
  * A {@link Monoid} is a {@link Semigroup} with an identity value.
  */
public interface Monoid
 extends Semigroup
{

E identity();

E inverse( E element );

}

that would continue working for every Monoid specialization. Or not? thoughts?
Thanks and best,
-Simo

http://people.apache.org/~simonetripodi/
http://simonetripodi.livejournal.com/
http://twitter.com/simonetripodi
http://www.99soft.org/



On Sat, Mar 3, 2012 at 1:43 PM, Claudio Squarcella
  wrote:

Hi,


On 03/03/2012 02:21, Simone Tripodi wrote:

first of all: yes, I will play with this stuff as soon as I find the time
:)

this is scaring... go Orb.io, go! :)


constant), but still there is one more step of indirection. So we would
need
to test and compare performances, hopefully with acceptable results.

sounds it would be better if you implement that stuff in a branch to
keep both implementations, IMHO


sure :)



  * with the current approach: one interface for edge-weighted graphs
   (EdgeWeightedGraph, renaming the current WeightedGraph?), one for
   vertex-weighted graphs (VertexWeightedGraph) and maybe even one for
   weights on both edges and 

Re: [graph] Why the Vertex and Edge interfaces?

[Simone Tripodi]

Cloud.io,

> back to James' first email: any type could be immediately used as
> edge/vertex without wrappers, while specific data related to the domain of
> graphs (weights, labels) would be handled separately. I actually think this
> is useful: back to my days of "DIY Java graphs" I implemented something
> similar to what we have now, just to think every time: "why should I wrap my
> objects with these markers? I already know my Router is a Vertex in the
> graph..."

I already showed be open on dropping elements, do I have to copy my
first reply as well so we start again? :)
OK, I started collecting various thoughts and trying to converge them
to a common path: Vertex is something we can safety drop because we
know its nature at priori, markers are unnecessary.This is fine.

>
> Here's the way I see it. A Graph implementing HasWeightedEdges would
> have something like this inside:
>
> Map edgeWeights = new HashMap();
>
> [... populate map, other methods, etc ...]
>
> // implementing HasWeightedEdges#getEdgeWeight
> public W getEdgeWeight(E edge)
> {
>    [... check null etc...]
>    return edgeWeights.get(edge);
>
> }
>

what is the sense, at that point, on keeping the Edge?!! It would be
more than enough to know which is the Head and which is the Tail in
the Edge to get the W!

> then let's find a better name, but why *OrderedMonoid?

maybe because they implement OrderedMonoid? :)

> Assume we replace the
> whole set of current interfaces (see my comment to the next paragraph below)
> with just Addition and Comparable (the latter already exists of course).
> There is no need to create another interface to merge the two
> (ComparableAddition? OrderedAddition?). Then we have:
>

how much would Addition and Multiplication interface differ each other?

> public class DoubleWeightOperations
>    implements Addition, Comparator
> {
>    [...]
> }
>
> I would not rename the class to DoubleWeightAddition or even
> DoubleWeightComparableAddition. What if later we need to e.g. add a function
> that normalizes weights by some factor, or returns the reciprocal of a
> weight, etc? With the above code it would suffice to add new interfaces:
>
> public class DoubleWeightOperations
>    implements Addition, Comparator, Normalization, Reciprocal
> {
>    [...]
>
> }
>
>

that would be fine, what is not clear is that Monoids expose the same
interface, so *Operations class implementation canot declare same
method multiple times

>
> That is fine for me. I simply followed pure theory while implementing that
> and used all possible granularity.

questionable, that is why we are still speaking about it

> Now looking at our implementations I
> think it is save enough to even merge Zero, Semigroup and Monoid (so that's
> one step further than your example below) and call the result Addition so
> that its role is clear to everybody. Does that sound good? :)

Sounds more than good, it is what I already proposed messages ago:

> Zero, name of an element, contains `zero` method to get the zero (it
> is still confusing to me), Monoid  extends Zero and Semigroup - given
> the use inside graph math, Zero#zero and Semigroup#append can be moved
> directly to Monoid and rename it as WeightOperation

despite the rename, I still like the Monoid :P

enough talk IMHO, time to code and make concrete proposals

best,
-Simo

http://people.apache.org/~simonetripodi/
http://simonetripodi.livejournal.com/
http://twitter.com/simonetripodi
http://www.99soft.org/



On Sat, Mar 3, 2012 at 2:58 PM, Claudio Squarcella
 wrote:
> Hi,
>
>
>> Apologize but I still don't understand what the benefit is on storing
>> nodes/arcs data in the Graph data structure
>
>
> back to James' first email: any type could be immediately used as
> edge/vertex without wrappers, while specific data related to the domain of
> graphs (weights, labels) would be handled separately. I actually think this
> is useful: back to my days of "DIY Java graphs" I implemented something
> similar to what we have now, just to think every time: "why should I wrap my
> objects with these markers? I already know my Router is a Vertex in the
> graph..."
>
>
>> - sounds to me like a
>> Collection  where, to know the int value of its elements, have
>> to query the data structure, i.e.
>>
>> Collection  integerCollection = ...;
>> for ( Integer ptr : integerCollection )
>> {
>>     int value = integerCollection.getInt( ptr );
>> }
>>
>> It is weird to me even reading it.
>>
>> When I started modeling Graph, I had collections in mind - above all
>> to simplify its adoption - I would be more than pleased to drop
>> Weighted* version of graphs and come back to the previous situation,
>> but with the annoying  type inference issue fixed.
>
>
> Here's the way I see it. A Graph implementing HasWeightedEdges would
> have something like this inside:
>
> Map edgeWeights = new HashMap();
>
> [... populate map, other methods, etc ...]
>
> // implementing HasWeightedEdges#getEdgeWeight
> public W getEdgeWeig

Re: [graph] Why the Vertex and Edge interfaces?

[James Carman]

Yes code will speak volumes.  Perhaps a couple proposal branches if we have
incompatible ideas?
On Mar 3, 2012 10:47 AM, "Simone Tripodi"  wrote:

> Cloud.io,
>
> > back to James' first email: any type could be immediately used as
> > edge/vertex without wrappers, while specific data related to the domain
> of
> > graphs (weights, labels) would be handled separately. I actually think
> this
> > is useful: back to my days of "DIY Java graphs" I implemented something
> > similar to what we have now, just to think every time: "why should I
> wrap my
> > objects with these markers? I already know my Router is a Vertex in the
> > graph..."
>
> I already showed be open on dropping elements, do I have to copy my
> first reply as well so we start again? :)
> OK, I started collecting various thoughts and trying to converge them
> to a common path: Vertex is something we can safety drop because we
> know its nature at priori, markers are unnecessary.This is fine.
>
> >
> > Here's the way I see it. A Graph implementing HasWeightedEdges would
> > have something like this inside:
> >
> > Map edgeWeights = new HashMap();
> >
> > [... populate map, other methods, etc ...]
> >
> > // implementing HasWeightedEdges#getEdgeWeight
> > public W getEdgeWeight(E edge)
> > {
> >[... check null etc...]
> >return edgeWeights.get(edge);
> >
> > }
> >
>
> what is the sense, at that point, on keeping the Edge?!! It would be
> more than enough to know which is the Head and which is the Tail in
> the Edge to get the W!
>
> > then let's find a better name, but why *OrderedMonoid?
>
> maybe because they implement OrderedMonoid? :)
>
> > Assume we replace the
> > whole set of current interfaces (see my comment to the next paragraph
> below)
> > with just Addition and Comparable (the latter already exists of course).
> > There is no need to create another interface to merge the two
> > (ComparableAddition? OrderedAddition?). Then we have:
> >
>
> how much would Addition and Multiplication interface differ each other?
>
> > public class DoubleWeightOperations
> >implements Addition, Comparator
> > {
> >[...]
> > }
> >
> > I would not rename the class to DoubleWeightAddition or even
> > DoubleWeightComparableAddition. What if later we need to e.g. add a
> function
> > that normalizes weights by some factor, or returns the reciprocal of a
> > weight, etc? With the above code it would suffice to add new interfaces:
> >
> > public class DoubleWeightOperations
> >implements Addition, Comparator, Normalization, Reciprocal
> > {
> >[...]
> >
> > }
> >
> >
>
> that would be fine, what is not clear is that Monoids expose the same
> interface, so *Operations class implementation canot declare same
> method multiple times
>
> >
> > That is fine for me. I simply followed pure theory while implementing
> that
> > and used all possible granularity.
>
> questionable, that is why we are still speaking about it
>
> > Now looking at our implementations I
> > think it is save enough to even merge Zero, Semigroup and Monoid (so
> that's
> > one step further than your example below) and call the result Addition so
> > that its role is clear to everybody. Does that sound good? :)
>
> Sounds more than good, it is what I already proposed messages ago:
>
> > Zero, name of an element, contains `zero` method to get the zero (it
> > is still confusing to me), Monoid  extends Zero and Semigroup - given
> > the use inside graph math, Zero#zero and Semigroup#append can be moved
> > directly to Monoid and rename it as WeightOperation
>
> despite the rename, I still like the Monoid :P
>
> enough talk IMHO, time to code and make concrete proposals
>
> best,
> -Simo
>
> http://people.apache.org/~simonetripodi/
> http://simonetripodi.livejournal.com/
> http://twitter.com/simonetripodi
> http://www.99soft.org/
>
>
>
> On Sat, Mar 3, 2012 at 2:58 PM, Claudio Squarcella
>  wrote:
> > Hi,
> >
> >
> >> Apologize but I still don't understand what the benefit is on storing
> >> nodes/arcs data in the Graph data structure
> >
> >
> > back to James' first email: any type could be immediately used as
> > edge/vertex without wrappers, while specific data related to the domain
> of
> > graphs (weights, labels) would be handled separately. I actually think
> this
> > is useful: back to my days of "DIY Java graphs" I implemented something
> > similar to what we have now, just to think every time: "why should I
> wrap my
> > objects with these markers? I already know my Router is a Vertex in the
> > graph..."
> >
> >
> >> - sounds to me like a
> >> Collection  where, to know the int value of its elements, have
> >> to query the data structure, i.e.
> >>
> >> Collection  integerCollection = ...;
> >> for ( Integer ptr : integerCollection )
> >> {
> >> int value = integerCollection.getInt( ptr );
> >> }
> >>
> >> It is weird to me even reading it.
> >>
> >> When I started modeling Graph, I had collections in mind - above all
> >> to simplify its adoption - I would b

Re: [graph] Why the Vertex and Edge interfaces?

[Simone Tripodi]

+1 for branches!
-Simo

http://people.apache.org/~simonetripodi/
http://simonetripodi.livejournal.com/
http://twitter.com/simonetripodi
http://www.99soft.org/



On Sat, Mar 3, 2012 at 5:01 PM, James Carman
 wrote:
> Yes code will speak volumes.  Perhaps a couple proposal branches if we have
> incompatible ideas?
> On Mar 3, 2012 10:47 AM, "Simone Tripodi"  wrote:
>
>> Cloud.io,
>>
>> > back to James' first email: any type could be immediately used as
>> > edge/vertex without wrappers, while specific data related to the domain
>> of
>> > graphs (weights, labels) would be handled separately. I actually think
>> this
>> > is useful: back to my days of "DIY Java graphs" I implemented something
>> > similar to what we have now, just to think every time: "why should I
>> wrap my
>> > objects with these markers? I already know my Router is a Vertex in the
>> > graph..."
>>
>> I already showed be open on dropping elements, do I have to copy my
>> first reply as well so we start again? :)
>> OK, I started collecting various thoughts and trying to converge them
>> to a common path: Vertex is something we can safety drop because we
>> know its nature at priori, markers are unnecessary.This is fine.
>>
>> >
>> > Here's the way I see it. A Graph implementing HasWeightedEdges would
>> > have something like this inside:
>> >
>> > Map edgeWeights = new HashMap();
>> >
>> > [... populate map, other methods, etc ...]
>> >
>> > // implementing HasWeightedEdges#getEdgeWeight
>> > public W getEdgeWeight(E edge)
>> > {
>> >    [... check null etc...]
>> >    return edgeWeights.get(edge);
>> >
>> > }
>> >
>>
>> what is the sense, at that point, on keeping the Edge?!! It would be
>> more than enough to know which is the Head and which is the Tail in
>> the Edge to get the W!
>>
>> > then let's find a better name, but why *OrderedMonoid?
>>
>> maybe because they implement OrderedMonoid? :)
>>
>> > Assume we replace the
>> > whole set of current interfaces (see my comment to the next paragraph
>> below)
>> > with just Addition and Comparable (the latter already exists of course).
>> > There is no need to create another interface to merge the two
>> > (ComparableAddition? OrderedAddition?). Then we have:
>> >
>>
>> how much would Addition and Multiplication interface differ each other?
>>
>> > public class DoubleWeightOperations
>> >    implements Addition, Comparator
>> > {
>> >    [...]
>> > }
>> >
>> > I would not rename the class to DoubleWeightAddition or even
>> > DoubleWeightComparableAddition. What if later we need to e.g. add a
>> function
>> > that normalizes weights by some factor, or returns the reciprocal of a
>> > weight, etc? With the above code it would suffice to add new interfaces:
>> >
>> > public class DoubleWeightOperations
>> >    implements Addition, Comparator, Normalization, Reciprocal
>> > {
>> >    [...]
>> >
>> > }
>> >
>> >
>>
>> that would be fine, what is not clear is that Monoids expose the same
>> interface, so *Operations class implementation canot declare same
>> method multiple times
>>
>> >
>> > That is fine for me. I simply followed pure theory while implementing
>> that
>> > and used all possible granularity.
>>
>> questionable, that is why we are still speaking about it
>>
>> > Now looking at our implementations I
>> > think it is save enough to even merge Zero, Semigroup and Monoid (so
>> that's
>> > one step further than your example below) and call the result Addition so
>> > that its role is clear to everybody. Does that sound good? :)
>>
>> Sounds more than good, it is what I already proposed messages ago:
>>
>> > Zero, name of an element, contains `zero` method to get the zero (it
>> > is still confusing to me), Monoid  extends Zero and Semigroup - given
>> > the use inside graph math, Zero#zero and Semigroup#append can be moved
>> > directly to Monoid and rename it as WeightOperation
>>
>> despite the rename, I still like the Monoid :P
>>
>> enough talk IMHO, time to code and make concrete proposals
>>
>> best,
>> -Simo
>>
>> http://people.apache.org/~simonetripodi/
>> http://simonetripodi.livejournal.com/
>> http://twitter.com/simonetripodi
>> http://www.99soft.org/
>>
>>
>>
>> On Sat, Mar 3, 2012 at 2:58 PM, Claudio Squarcella
>>  wrote:
>> > Hi,
>> >
>> >
>> >> Apologize but I still don't understand what the benefit is on storing
>> >> nodes/arcs data in the Graph data structure
>> >
>> >
>> > back to James' first email: any type could be immediately used as
>> > edge/vertex without wrappers, while specific data related to the domain
>> of
>> > graphs (weights, labels) would be handled separately. I actually think
>> this
>> > is useful: back to my days of "DIY Java graphs" I implemented something
>> > similar to what we have now, just to think every time: "why should I
>> wrap my
>> > objects with these markers? I already know my Router is a Vertex in the
>> > graph..."
>> >
>> >
>> >> - sounds to me like a
>> >> Collection  where, to know the int value of its elements, have
>> >> 

Re: [graph] Why the Vertex and Edge interfaces?

[Oliver Heger]

Hi, if I am allowed to throw in some (unfinished) thoughts into this 
discussion:


My preference would also be to keep the actual graph data structure lean 
and simple. As James pointed out, if algorithms need additional 
information, a function from the vertex/edge type to the target type is 
required. This could indeed be implemented in a very generic way using a 
functor interface (maybe with an underlying map as a default 
implementation).


The advantage over a pack of interfaces as I see it is that such 
functors could be added dynamically to a graph. Thus it could be adapted 
for the execution of different algorithms at runtime.


Just my 2 cents.
Oliver

Am 03.03.2012 16:47, schrieb Simone Tripodi:

Cloud.io,


back to James' first email: any type could be immediately used as
edge/vertex without wrappers, while specific data related to the domain of
graphs (weights, labels) would be handled separately. I actually think this
is useful: back to my days of "DIY Java graphs" I implemented something
similar to what we have now, just to think every time: "why should I wrap my
objects with these markers? I already know my Router is a Vertex in the
graph..."


I already showed be open on dropping elements, do I have to copy my
first reply as well so we start again? :)
OK, I started collecting various thoughts and trying to converge them
to a common path: Vertex is something we can safety drop because we
know its nature at priori, markers are unnecessary.This is fine.



Here's the way I see it. A Graph  implementing HasWeightedEdges would
have something like this inside:

Map  edgeWeights = new HashMap();

[... populate map, other methods, etc ...]

// implementing HasWeightedEdges#getEdgeWeight
public W getEdgeWeight(E edge)
{
[... check null etc...]
return edgeWeights.get(edge);

}



what is the sense, at that point, on keeping the Edge?!! It would be
more than enough to know which is the Head and which is the Tail in
the Edge to get the W!


then let's find a better name, but why *OrderedMonoid?


maybe because they implement OrderedMonoid? :)


Assume we replace the
whole set of current interfaces (see my comment to the next paragraph below)
with just Addition and Comparable (the latter already exists of course).
There is no need to create another interface to merge the two
(ComparableAddition? OrderedAddition?). Then we have:



how much would Addition and Multiplication interface differ each other?


public class DoubleWeightOperations
implements Addition, Comparator
{
[...]
}

I would not rename the class to DoubleWeightAddition or even
DoubleWeightComparableAddition. What if later we need to e.g. add a function
that normalizes weights by some factor, or returns the reciprocal of a
weight, etc? With the above code it would suffice to add new interfaces:

public class DoubleWeightOperations
implements Addition, Comparator, Normalization, Reciprocal
{
[...]

}




that would be fine, what is not clear is that Monoids expose the same
interface, so *Operations class implementation canot declare same
method multiple times



That is fine for me. I simply followed pure theory while implementing that
and used all possible granularity.


questionable, that is why we are still speaking about it


Now looking at our implementations I
think it is save enough to even merge Zero, Semigroup and Monoid (so that's
one step further than your example below) and call the result Addition so
that its role is clear to everybody. Does that sound good? :)


Sounds more than good, it is what I already proposed messages ago:


Zero, name of an element, contains `zero` method to get the zero (it
is still confusing to me), Monoid  extends Zero and Semigroup - given
the use inside graph math, Zero#zero and Semigroup#append can be moved
directly to Monoid and rename it as WeightOperation


despite the rename, I still like the Monoid :P

enough talk IMHO, time to code and make concrete proposals

best,
-Simo

http://people.apache.org/~simonetripodi/
http://simonetripodi.livejournal.com/
http://twitter.com/simonetripodi
http://www.99soft.org/



On Sat, Mar 3, 2012 at 2:58 PM, Claudio Squarcella
  wrote:

Hi,



Apologize but I still don't understand what the benefit is on storing
nodes/arcs data in the Graph data structure



back to James' first email: any type could be immediately used as
edge/vertex without wrappers, while specific data related to the domain of
graphs (weights, labels) would be handled separately. I actually think this
is useful: back to my days of "DIY Java graphs" I implemented something
similar to what we have now, just to think every time: "why should I wrap my
objects with these markers? I already know my Router is a Vertex in the
graph..."



- sounds to me like a
Collectionwhere, to know the int value of its elements, have
to query the data structure, i.e.

CollectionintegerCollection = ...;
for ( Integer ptr : integerCollection )
{
 int value = integerCollection.ge

Re: [graph] Why the Vertex and Edge interfaces?

[Simone Tripodi]

Hi Oliver,

of course that you can, everybody here is invited on giving his
contribution! And every contribution is always much more than
appreciated :)

best,
-Simo

http://people.apache.org/~simonetripodi/
http://simonetripodi.livejournal.com/
http://twitter.com/simonetripodi
http://www.99soft.org/



On Sat, Mar 3, 2012 at 5:03 PM, Oliver Heger
 wrote:
> Hi, if I am allowed to throw in some (unfinished) thoughts into this
> discussion:
>
> My preference would also be to keep the actual graph data structure lean and
> simple. As James pointed out, if algorithms need additional information, a
> function from the vertex/edge type to the target type is required. This
> could indeed be implemented in a very generic way using a functor interface
> (maybe with an underlying map as a default implementation).
>
> The advantage over a pack of interfaces as I see it is that such functors
> could be added dynamically to a graph. Thus it could be adapted for the
> execution of different algorithms at runtime.
>
> Just my 2 cents.
> Oliver
>
> Am 03.03.2012 16:47, schrieb Simone Tripodi:
>
>> Cloud.io,
>>
>>> back to James' first email: any type could be immediately used as
>>> edge/vertex without wrappers, while specific data related to the domain
>>> of
>>> graphs (weights, labels) would be handled separately. I actually think
>>> this
>>> is useful: back to my days of "DIY Java graphs" I implemented something
>>> similar to what we have now, just to think every time: "why should I wrap
>>> my
>>> objects with these markers? I already know my Router is a Vertex in the
>>> graph..."
>>
>>
>> I already showed be open on dropping elements, do I have to copy my
>> first reply as well so we start again? :)
>> OK, I started collecting various thoughts and trying to converge them
>> to a common path: Vertex is something we can safety drop because we
>> know its nature at priori, markers are unnecessary.This is fine.
>>
>>>
>>> Here's the way I see it. A Graph  implementing HasWeightedEdges
>>> would
>>> have something like this inside:
>>>
>>> Map  edgeWeights = new HashMap();
>>>
>>> [... populate map, other methods, etc ...]
>>>
>>> // implementing HasWeightedEdges#getEdgeWeight
>>> public W getEdgeWeight(E edge)
>>> {
>>>    [... check null etc...]
>>>    return edgeWeights.get(edge);
>>>
>>> }
>>>
>>
>> what is the sense, at that point, on keeping the Edge?!! It would be
>> more than enough to know which is the Head and which is the Tail in
>> the Edge to get the W!
>>
>>> then let's find a better name, but why *OrderedMonoid?
>>
>>
>> maybe because they implement OrderedMonoid? :)
>>
>>> Assume we replace the
>>> whole set of current interfaces (see my comment to the next paragraph
>>> below)
>>> with just Addition and Comparable (the latter already exists of course).
>>> There is no need to create another interface to merge the two
>>> (ComparableAddition? OrderedAddition?). Then we have:
>>>
>>
>> how much would Addition and Multiplication interface differ each other?
>>
>>> public class DoubleWeightOperations
>>>    implements Addition, Comparator
>>> {
>>>    [...]
>>> }
>>>
>>> I would not rename the class to DoubleWeightAddition or even
>>> DoubleWeightComparableAddition. What if later we need to e.g. add a
>>> function
>>> that normalizes weights by some factor, or returns the reciprocal of a
>>> weight, etc? With the above code it would suffice to add new interfaces:
>>>
>>> public class DoubleWeightOperations
>>>    implements Addition, Comparator, Normalization, Reciprocal
>>> {
>>>    [...]
>>>
>>> }
>>>
>>>
>>
>> that would be fine, what is not clear is that Monoids expose the same
>> interface, so *Operations class implementation canot declare same
>> method multiple times
>>
>>>
>>> That is fine for me. I simply followed pure theory while implementing
>>> that
>>> and used all possible granularity.
>>
>>
>> questionable, that is why we are still speaking about it
>>
>>> Now looking at our implementations I
>>> think it is save enough to even merge Zero, Semigroup and Monoid (so
>>> that's
>>> one step further than your example below) and call the result Addition so
>>> that its role is clear to everybody. Does that sound good? :)
>>
>>
>> Sounds more than good, it is what I already proposed messages ago:
>>
>>> Zero, name of an element, contains `zero` method to get the zero (it
>>> is still confusing to me), Monoid  extends Zero and Semigroup - given
>>> the use inside graph math, Zero#zero and Semigroup#append can be moved
>>> directly to Monoid and rename it as WeightOperation
>>
>>
>> despite the rename, I still like the Monoid :P
>>
>> enough talk IMHO, time to code and make concrete proposals
>>
>> best,
>> -Simo
>>
>> http://people.apache.org/~simonetripodi/
>> http://simonetripodi.livejournal.com/
>> http://twitter.com/simonetripodi
>> http://www.99soft.org/
>>
>>
>>
>> On Sat, Mar 3, 2012 at 2:58 PM, Claudio Squarcella
>>   wrote:
>>>
>>> Hi,
>>>
>>>
 Apologize but I still don't understand 

Re: [graph] Why the Vertex and Edge interfaces?

[Claudio Squarcella]

Hi Simone,

answering briefly below:


Vertex is something we can safety drop because we
know its nature at priori, markers are unnecessary.This is fine.


+1.


what is the sense, at that point, on keeping the Edge?!! It would be
more than enough to know which is the Head and which is the Tail in
the Edge to get the W!


good observation. My 2 cents: it might still make sense for users to map 
their existing domain (including "edges") to the graph (e.g. Routers to 
Vertices and Cables to Edges) and "get it back" as soon as they are done 
with graph operations (e.g. once they find the shortest path, they 
automatically have the sequence of Cables they need).



maybe because they implement OrderedMonoid? :)
[...]
how much would Addition and Multiplication interface differ each other?
[...]
that would be fine, what is not clear is that Monoids expose the same
interface, so *Operations class implementation canot declare same
method multiple times


answering all the above: that is another reason why I would like our 
current Monoid to be called Addition (and Addition#sum instead of 
Monoid#append, etc), so that it is semantically clearer and later we can 
introduce Multiplication as a completely independent interface.



enough talk IMHO, time to code and make concrete proposals


Sure!
I'll play with weights first, because I already know what I want to do.
As for Vertex/Edge markers I still see valuable feedback coming in, so 
I'll wait a bit.


Branching is ok -- especially for the second part which sounds like a 
real earthquake ;)


Ciao,
Claudio

--
Claudio Squarcella
PhD student at Roma Tre University
http://www.dia.uniroma3.it/~squarcel
http://squarcella.com/


-
To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
For additional commands, e-mail: dev-h...@commons.apache.org

Re: [graph] Why the Vertex and Edge interfaces?

[Simone Tripodi]

please follow up code modifications on SANDBOX-404

http://people.apache.org/~simonetripodi/
http://simonetripodi.livejournal.com/
http://twitter.com/simonetripodi
http://www.99soft.org/



On Sat, Mar 3, 2012 at 5:11 PM, Claudio Squarcella
 wrote:
> Hi Simone,
>
> answering briefly below:
>
>
>> Vertex is something we can safety drop because we
>> know its nature at priori, markers are unnecessary.This is fine.
>
>
> +1.
>
>
>> what is the sense, at that point, on keeping the Edge?!! It would be
>> more than enough to know which is the Head and which is the Tail in
>> the Edge to get the W!
>
>
> good observation. My 2 cents: it might still make sense for users to map
> their existing domain (including "edges") to the graph (e.g. Routers to
> Vertices and Cables to Edges) and "get it back" as soon as they are done
> with graph operations (e.g. once they find the shortest path, they
> automatically have the sequence of Cables they need).
>
>> maybe because they implement OrderedMonoid? :)
>> [...]
>>
>> how much would Addition and Multiplication interface differ each other?
>> [...]
>> that would be fine, what is not clear is that Monoids expose the same
>> interface, so *Operations class implementation canot declare same
>> method multiple times
>
>
> answering all the above: that is another reason why I would like our current
> Monoid to be called Addition (and Addition#sum instead of Monoid#append,
> etc), so that it is semantically clearer and later we can introduce
> Multiplication as a completely independent interface.
>
>
>> enough talk IMHO, time to code and make concrete proposals
>
>
> Sure!
> I'll play with weights first, because I already know what I want to do.
> As for Vertex/Edge markers I still see valuable feedback coming in, so I'll
> wait a bit.
>
> Branching is ok -- especially for the second part which sounds like a real
> earthquake ;)
>
>
> Ciao,
> Claudio
>
> --
> Claudio Squarcella
> PhD student at Roma Tre University
> http://www.dia.uniroma3.it/~squarcel
> http://squarcella.com/
>
>
> -
> To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
> For additional commands, e-mail: dev-h...@commons.apache.org
>

-
To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
For additional commands, e-mail: dev-h...@commons.apache.org

Re: [graph] Why the Vertex and Edge interfaces?

[Simone Tripodi]

hola again,

> good observation. My 2 cents: it might still make sense for users to map
> their existing domain (including "edges") to the graph (e.g. Routers to
> Vertices and Cables to Edges) and "get it back" as soon as they are done
> with graph operations (e.g. once they find the shortest path, they
> automatically have the sequence of Cables they need).

so, at least WeightedEdge is something that still has to exist - what
is the reason to query the Network to know the Cable capacity?!? Isn't
it enough getting it directly from the Cable itself?

-Simo

http://people.apache.org/~simonetripodi/
http://simonetripodi.livejournal.com/
http://twitter.com/simonetripodi
http://www.99soft.org/

-
To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
For additional commands, e-mail: dev-h...@commons.apache.org

Re: [graph] Why the Vertex and Edge interfaces?

[James Carman]

We also need to consider k-partite graphs.  Not all nodes will be of the
same type.

Sent from tablet device.  Please excuse typos and brevity.
On Mar 3, 2012 11:19 AM, "Simone Tripodi"  wrote:

> hola again,
>
> > good observation. My 2 cents: it might still make sense for users to map
> > their existing domain (including "edges") to the graph (e.g. Routers to
> > Vertices and Cables to Edges) and "get it back" as soon as they are done
> > with graph operations (e.g. once they find the shortest path, they
> > automatically have the sequence of Cables they need).
>
> so, at least WeightedEdge is something that still has to exist - what
> is the reason to query the Network to know the Cable capacity?!? Isn't
> it enough getting it directly from the Cable itself?
>
> -Simo
>
> http://people.apache.org/~simonetripodi/
> http://simonetripodi.livejournal.com/
> http://twitter.com/simonetripodi
> http://www.99soft.org/
>
> -
> To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
> For additional commands, e-mail: dev-h...@commons.apache.org
>
>

Re: [graph] Why the Vertex and Edge interfaces?

[Simone Tripodi]

Hi James,
would a Graph, which Vertices are of type Object, help on this?
TIA,
-Simo

http://people.apache.org/~simonetripodi/
http://simonetripodi.livejournal.com/
http://twitter.com/simonetripodi
http://www.99soft.org/



On Sun, Mar 4, 2012 at 2:10 AM, James Carman
 wrote:
> We also need to consider k-partite graphs.  Not all nodes will be of the
> same type.
>
> Sent from tablet device.  Please excuse typos and brevity.
> On Mar 3, 2012 11:19 AM, "Simone Tripodi"  wrote:
>
>> hola again,
>>
>> > good observation. My 2 cents: it might still make sense for users to map
>> > their existing domain (including "edges") to the graph (e.g. Routers to
>> > Vertices and Cables to Edges) and "get it back" as soon as they are done
>> > with graph operations (e.g. once they find the shortest path, they
>> > automatically have the sequence of Cables they need).
>>
>> so, at least WeightedEdge is something that still has to exist - what
>> is the reason to query the Network to know the Cable capacity?!? Isn't
>> it enough getting it directly from the Cable itself?
>>
>> -Simo
>>
>> http://people.apache.org/~simonetripodi/
>> http://simonetripodi.livejournal.com/
>> http://twitter.com/simonetripodi
>> http://www.99soft.org/
>>
>> -
>> To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
>> For additional commands, e-mail: dev-h...@commons.apache.org
>>
>>

-
To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
For additional commands, e-mail: dev-h...@commons.apache.org

Re: [graph] Why the Vertex and Edge interfaces?

[James Carman]

It would, but there is another property of these graphs that it wouldn't
enforce.  The vertices aren't supposed to be connected to other vertices in
the same set.
On Mar 4, 2012 8:13 AM, "Simone Tripodi"  wrote:

> Hi James,
> would a Graph, which Vertices are of type Object, help on this?
> TIA,
> -Simo
>
> http://people.apache.org/~simonetripodi/
> http://simonetripodi.livejournal.com/
> http://twitter.com/simonetripodi
> http://www.99soft.org/
>
>
>
> On Sun, Mar 4, 2012 at 2:10 AM, James Carman
>  wrote:
> > We also need to consider k-partite graphs.  Not all nodes will be of the
> > same type.
> >
> > Sent from tablet device.  Please excuse typos and brevity.
> > On Mar 3, 2012 11:19 AM, "Simone Tripodi" 
> wrote:
> >
> >> hola again,
> >>
> >> > good observation. My 2 cents: it might still make sense for users to
> map
> >> > their existing domain (including "edges") to the graph (e.g. Routers
> to
> >> > Vertices and Cables to Edges) and "get it back" as soon as they are
> done
> >> > with graph operations (e.g. once they find the shortest path, they
> >> > automatically have the sequence of Cables they need).
> >>
> >> so, at least WeightedEdge is something that still has to exist - what
> >> is the reason to query the Network to know the Cable capacity?!? Isn't
> >> it enough getting it directly from the Cable itself?
> >>
> >> -Simo
> >>
> >> http://people.apache.org/~simonetripodi/
> >> http://simonetripodi.livejournal.com/
> >> http://twitter.com/simonetripodi
> >> http://www.99soft.org/
> >>
> >> -
> >> To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
> >> For additional commands, e-mail: dev-h...@commons.apache.org
> >>
> >>
>
> -
> To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
> For additional commands, e-mail: dev-h...@commons.apache.org
>
>

Followup - [graph] Why the Vertex and Edge interfaces?

[Simone Tripodi]

Hi all guys,

this message just to invite you on testing the experimental branch[1]
where I got rid completely of marcher interfaces, such as
Vertex/Edge/WeightedGraph and related stuff and various combinations
of them.
I took advantage to keep only needed generics in builder chains, that
are used to interfere types in the next builder, reducing the
verbosity and improving the readability.

Code is now IMHO much more versatile on hosting users' data, they are
now free to define Graph to measure the page
relationship, for example :P

Please have a look and share your thoughts - if there are no
objections, I'd propose to merge it to trunk.

TIA, all the best and have a nice WE,
-Simo

PS: I muted exporter ATM since, given the current design, we should
think a different strategy to define exporting properties for
vertices/edges. There was an open issue[2] about it, it is maybe time
to resurrect it :P

[1] 
https://svn.apache.org/repos/asf/commons/sandbox/graph/branches/drop-marker-interfaces-feature
[2] https://issues.apache.org/jira/browse/SANDBOX-339

http://people.apache.org/~simonetripodi/
http://simonetripodi.livejournal.com/
http://twitter.com/simonetripodi
http://www.99soft.org/

-
To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
For additional commands, e-mail: dev-h...@commons.apache.org

Re: Followup - [graph] Why the Vertex and Edge interfaces?

[squarcel]

Hi Simone,

good job, thanks for experimenting!

Looks generally ok to me. I will take some more time before giving thumbs
up to merge it -- and not other fingers :P

In the meantime it might be worth to share my naive idea for graph
exporters here (I might implement it in the next days):
we could explicitly define a family of Mappers (e.g. WeightMapper,
LabelMapper, maybe later CoordinateMapper, etc), so that

 * the whole thing gets a bit more human-readable, and
 * when exporting a graph the user can specify as many Mappers as
   wanted (e.g. passing an iterator or simply an arbitrary number of
   args), but then the exporter is responsible for only selecting those
   that match the export format and (silently? with an exception?) drop
   the others.

Claudio


> Hi all guys,
>
> this message just to invite you on testing the experimental branch[1]
> where I got rid completely of marcher interfaces, such as
> Vertex/Edge/WeightedGraph and related stuff and various combinations
> of them.
> I took advantage to keep only needed generics in builder chains, that
> are used to interfere types in the next builder, reducing the
> verbosity and improving the readability.
>
> Code is now IMHO much more versatile on hosting users' data, they are
> now free to define Graph to measure the page
> relationship, for example :P
>
> Please have a look and share your thoughts - if there are no
> objections, I'd propose to merge it to trunk.
>
> TIA, all the best and have a nice WE,
> -Simo
>
> PS: I muted exporter ATM since, given the current design, we should
> think a different strategy to define exporting properties for
> vertices/edges. There was an open issue[2] about it, it is maybe time
> to resurrect it :P
>
> [1]
> https://svn.apache.org/repos/asf/commons/sandbox/graph/branches/drop-marker-interfaces-feature
> [2] https://issues.apache.org/jira/browse/SANDBOX-339
>
> http://people.apache.org/~simonetripodi/
> http://simonetripodi.livejournal.com/
> http://twitter.com/simonetripodi
> http://www.99soft.org/
>
> -
> To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
> For additional commands, e-mail: dev-h...@commons.apache.org
>
>


-
This email was sent using SquirrelMail.
https://email.dia.uniroma3.it
Web Site: http://www.squirrelmail.org


-
To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
For additional commands, e-mail: dev-h...@commons.apache.org