Implemented the Moran process on graphs #799

marcharper · 2017-01-01T05:26:41Z

drvinceknight

Awesome! Looking forward to having this in.

Some questions (that I don't doubt you have the answer to but just want to make sure we have considered things) and subsequently a request for tests.

drvinceknight · 2017-01-01T11:05:06Z

axelrod/__init__.py

@@ -5,6 +5,7 @@

 # The order of imports matters!
 from .version import __version__
+from . import graph


I would have thought this line wasn't needed? Am I wrong?

It depends on how you access the graph class in code, axelrod.graph.Graph will fail without this line.

We could use from .graph import Graph, complete_graph, ...

It depends on how you access the graph class in code, axelrod.graph.Graph will fail

I thought it wouldn't fail but happy as it is :)

drvinceknight · 2017-01-01T11:10:44Z

axelrod/graph.py

+Weighted undirected sparse graphs.
+
+Original source:
+https://github.com/marcharper/stationary/blob/master/stationary/utils/graph.py


Not against this. A couple of initial thoughts:

As you have built this for your stationary library you no doubt have a reason but why we wouldn't just use networkx (which would have the quick lookups we want here)? (1 negative side is adding another dependency to the library)

Can we add some tests please?

The spatial tournament does not necessarily need a graph object (it just loops through the edges once via the match generator) but if we're going to have an axelrod.graph class we should also use that there (for consistency). I see two approaches:

We allow the spatial tournament to take both a list of edges of one of these graph objects (ensures backwards compatibility);

We let the Moran process take a list of edges and obfuscate the building of this underlying graph object there. I think this would be my preferred approach.

To be clear: If we go with my suggestion of 3.i then that would be work for a different PR.

I figured this would come up which is why I didn't add tests yet. We can switch this out for another library, I just used what I had laying around. It's a bit overkill since we don't need the weights (yet?), but it optimized in the right ways -- sparse implementation, precomputed neighbors. So we could also just simplify it to parts needed (relatively easy).

I'm not opposed to networkX but I'm not familiar with its API, what internal representation it uses (adjacency matrix? sparse matrix?) etc. Thoughts?

I agree re: consistent API with spatial tournaments -- we could also use isinstance to be flexible on the input, and also have the graph's __repr__ method just output the list of edges. There are cases however that a list of edges doesn't cover, like some types of random graphs.

There are cases however that a list of edges doesn't cover, like some types of random graphs.

That's the killer point right there! I guess that also includes the possibility of graphs that change over time etc which also rules out networkX (which really is a data representation library coupled with graph theoretic algorithms as opposed to the flexibility we might need here).

I suggest going for flexibility on the input (so we can open an issue for my suggestion of 3.i).

In order to get the __repr__ method to output the list of edges I expect we could use an edges attribute or otherwise? That could be a useful attribute to have anyway?

Yeah, we can save the input edges, or output a sorted list.

There are cases in the literature where the graph changes, it's called "active linking".

Turns out we need the graph to be weighted for some applications in the literature. If the reproduction graph is weighted then it changes the death probabilities in the birth-death case.

drvinceknight · 2017-01-01T11:12:31Z

axelrod/graph.py

+    return graph
+
+
+def complete_graph(length, directed=False):


Do we need these extra functions? Networkx can be used to easily get a variety of graphs. (Just a thought)

I needed them for testing...

Cool, we can also use these in some of the tests for Spatial Tournaments (not in this PR).

I started to add more, an N x M grid could be useful... we can just add them as we go.

drvinceknight · 2017-01-01T11:14:32Z

axelrod/graph.py

+        """Returns a dictionary of the outgoing edges of source with weights."""
+        return self.out_mapping[source]
+
+    def out_vertices(self, source):


If undirected, why not just have a adjacent_vertices method? (To replace both the in and out ones?).

We may want to allow the graphs to be directed, see e.g. here.

drvinceknight · 2017-01-01T11:16:53Z

docs/tutorials/getting_started/moran.rst

+Moran Process on Graphs
+-----------------------
+
+The library also provides a graph-based Moran process with `MoranProcessGraph`.


Would you be able to include a reference text/paper in the bibliography?

Yes I can add a reference.

drvinceknight · 2017-01-02T08:09:44Z

There's a py3.6 failure (on the merge with master).

drvinceknight · 2017-01-02T08:11:17Z

docs/reference/bibliography.rst

@@ -26,6 +26,7 @@ documentation.
 .. [PRISON1998] LIFL (1998) PRISON. Available at: http://www.lifl.fr/IPD/ipd.frame.html (Accessed: 19 September 2016).
 .. [Robson1989] Robson, Arthur, (1989), EFFICIENCY IN EVOLUTIONARY GAMES: DARWIN, NASH AND SECRET HANDSHAKE, Working Papers, Michigan - Center for Research on Economic & Social Theory, http://EconPapers.repec.org/RePEc:fth:michet:89-22.
 .. [Singer-Clark2014] Singer-Clark, T. (2014). Morality Metrics On Iterated Prisoner’s Dilemma Players.
+.. [Shakarian2013] Shakarian, P., Roos, P. & Moores, G. A Novel Analytical Method for Evolutionary Graph Theory Problems.


Thanks for this: could you cite it in the moran.rst file?

drvinceknight · 2017-01-02T15:23:37Z

There's a py3.6 failure (on the merge with master).

@marcharper I took a look at the tests and pushed a fix. I pushed directly to this PR branch but feel free to change/modify/delete if you'd rather not.

Thanks for this: could you cite it in the moran.rst file?

I also took the opportunity to pop the citation where I thought it should go. Only trying to be helpful so if you're not a fan please throw it out :)

marcharper · 2017-01-02T17:39:18Z

Thanks for the test fix!

meatballs · 2017-01-03T11:08:33Z

axelrod/graph.py

+        """Returns a dictionary of the outgoing edges of source with weights."""
+        return self.out_mapping[source]
+
+    def out_vertices(self, source):


Might some of these methods be better as properties?

Getters would make sense for e.g vertices not sure about the others. Maybe edges makes sense with a setter, essentially rewriting the graph. I'm open to suggestions.

Let's leave it for now. If we fancy refactoring the graph, we can do so later.

marcharper added 4 commits December 31, 2016 17:20

Add Moran process on graphs

8b9459a

Docs for Moran Process on Graphs

96c73e0

Death-birth Moran process variant

af69177

Update Moran doctests

88989f7

drvinceknight requested changes Jan 1, 2017

View reviewed changes

marcharper force-pushed the graph_moran branch from dc9ce08 to 5332637 Compare January 2, 2017 01:48

Graph and Moran tests

0985230

marcharper force-pushed the graph_moran branch from 5332637 to 0985230 Compare January 2, 2017 02:51

drvinceknight requested changes Jan 2, 2017

View reviewed changes

drvinceknight added 2 commits January 2, 2017 15:18

Fix seed for py3.5 and py3.6 compatibility.

d4bb462

Cite reference in text for moran docs.

496fe41

drvinceknight approved these changes Jan 2, 2017

View reviewed changes

drvinceknight added the ready-to-merge label Jan 2, 2017

meatballs reviewed Jan 3, 2017

View reviewed changes

meatballs merged commit 67856f9 into master Jan 5, 2017

meatballs deleted the graph_moran branch January 5, 2017 10:18

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Implemented the Moran process on graphs #799

Implemented the Moran process on graphs #799

marcharper commented Jan 1, 2017 •

edited

drvinceknight left a comment

drvinceknight Jan 1, 2017

marcharper Jan 1, 2017 •

edited

drvinceknight Jan 1, 2017

drvinceknight Jan 1, 2017

drvinceknight Jan 1, 2017

marcharper Jan 1, 2017 •

edited

drvinceknight Jan 1, 2017

drvinceknight Jan 1, 2017

marcharper Jan 1, 2017

marcharper Jan 1, 2017

marcharper Jan 1, 2017

drvinceknight Jan 1, 2017

marcharper Jan 1, 2017

drvinceknight Jan 1, 2017

marcharper Jan 1, 2017

drvinceknight Jan 1, 2017

marcharper Jan 1, 2017

drvinceknight Jan 1, 2017

drvinceknight Jan 1, 2017

marcharper Jan 1, 2017

drvinceknight commented Jan 2, 2017

drvinceknight Jan 2, 2017

drvinceknight commented Jan 2, 2017

marcharper commented Jan 2, 2017

meatballs Jan 3, 2017

marcharper Jan 4, 2017

meatballs Jan 5, 2017

Implemented the Moran process on graphs #799

Implemented the Moran process on graphs #799

Conversation

marcharper commented Jan 1, 2017 • edited

drvinceknight left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

marcharper Jan 1, 2017 • edited

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

marcharper Jan 1, 2017 • edited

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

drvinceknight commented Jan 2, 2017

Choose a reason for hiding this comment

drvinceknight commented Jan 2, 2017

marcharper commented Jan 2, 2017

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

marcharper commented Jan 1, 2017 •

edited

marcharper Jan 1, 2017 •

edited

marcharper Jan 1, 2017 •

edited