635 Implementing spatial tournaments #654

Nikoleta-v3 · 2016-07-13T13:38:26Z

Closes #635.

Implemented a new tournament class and a new match generator class for spatial tournaments
Appropriate tests for the new classes have been written
Created a doc file in the further issues for spatial tournaments.

After considering all the thoughts on how to implement a spatial structure tournament, I have decided to
simple consider a list that includes the edges. The new class SpatialMatches() creates
chunks where players - nodes only play players that are connected to with an edge. A tournament
class SpatialTournament was also implemented to pass the edges argument and SpatialMatches
as the match_generator. An error message was also included. Just in case a player is not interacting with anyone . If you have such a player is better to not include him in the tournament at all.

Before creating these classes I had written some tests. For SpatialMatches, SpatialTournament , as well as some property tests. Also a test for comparing the results of a complete multi-graph and a round robin tournament.

In the result set, seems that in the loop statements the self interaction have been counted twice each time. So an if statement had to be altered to an elif.
Also there was one test in the result_tests that seem to have been passing by change, so I altered the
implementation a little bit.

Furthermore, I have added in the docs of the Axelrod library a small explanation and example of how to create spatial tournaments. All tests seem to be working fine !

This class can be used from the tournament to create matches with spatial structure topology. By spatial stucture, we mean a graph where the players are the nodes and they only play other players - nodes that are connected with a an edge.

SpatialMatches is class created in the match generator file in order to create matches where not all players play each other. But instead the players, that are illustrated as nodes in a graph, only play with players that are connected to by an edge. SpatialMatches inherit from RoundRobinMatches but there is need for an extra argument taken by the class which is an edge dictionary. To achieve that we use the python function super. Super takes the name of the class and a self argument, and here we will replace what was previously written as super(RoundRobinMatches, self) with super(SpatialMatches, self).

SpatialTournament is a class which will be running a Spatial Tournament by calling the SpatialMatches class we created before. It inherit for the class Tournament and these are some tests for the specific class. The tests for the SpatialTournament do not contain the hypothesis tests yet. They will be added shortly. We have run all the unit tests and we all 22 have passed.

The Spatial Tournament class is a class resposinble for creating a Spatial structure tournament, where the players are respresented on a graph and they only matches are bewteen players-nodes connected with an edge. It does that by using the SpatialMatch class and it inherit from the tournament class. It's take a extra argument which is a dictionary called edges, that represents which nodes- players are connected and will face each other in the tournament. Proper testes for this class have been written and all are passing.

__init__.py file is responsible for importing functions and classes and putting everything together for axelrod library. Now that Spatial Tournament is written is has to be added in to be imported as well. The same was not needed to be done for SpatialMatches because __init__.py already imports everything for the file Match, which contains SpatialMatches.

In case a tournament is created containing disconnected nodes-players an Error will be displayed. That is because if a player does not have any interaction with any other player it should be removed from the tournament as not active player could not have a score or rank in the tournament. A player not participating is different from a player scoring zero and this should be noted. Also a test for this constrain was added in the test_match_generator.

This test was not correct as the number of repetitions is not the same as the number of interactions.

These need to be refactored into test_resultsset.py

Currently the underlying graph is cyclic it needs to be random.

In a complete multigraph alla the nodes are connected to each other and to themselves. Which in a round robin topology. Therefore, a test to make sure that the results procuded by a complete multigraph and a round robin are equivalent.

After running the tests with Travis has returned the following error: Data generation is extremely slow. Decided to simply remove the tests.

marcharper · 2016-07-13T16:08:47Z

axelrod/match_generator.py

+            The game object used to score the match
+        repetitions : int
+            The number of repetitions of a given match
+        edges : dictionary


Can you explain what this dictionary looks like? In the tests below it looks like a list.

marcharper · 2016-07-13T16:10:28Z

Can we add a test that has a set of deterministic players on a complete graph compared to the non-spatial tournament? We should get the same results in both cases. (Not sure if this is already there.)

Nikoleta-v3 · 2016-07-13T16:14:07Z

First of all, it is a list ! I just happened to be calling it a dictionary for a long time... I will correct that, there seems to be more than once. Also for your second point. It has already been implemented, is the test for the complete multigraph.

drvinceknight · 2016-07-13T16:16:38Z

Also for your second point. It has already been implemented, is the test for the complete multigraph.

Here: https://github.com/Axelrod-Python/Axelrod/pull/654/files#diff-74e02134acd0ef3b29c84d6643837201R593 (line 593):

+        def test_complete_tournament(self, strategies, turns):
+            """
+            A test to check that a spatial tournament on the complete multigraph
+            gives the same results as the round robin.
+            """

marcharper · 2016-07-13T16:17:38Z

Great, sorry I'm not familiar with the term multigraph.

Is there an easy demo case that compares to the other tournaments in the documentation? There is an example that shows how many cooperators you need to overcome a defector, is there a topology that changes the number of cooperators needed (a cycle perhaps)? That would be a nice addition to the docs.

drvinceknight · 2016-07-13T16:21:55Z

Great, sorry I'm not familiar with the term multigraph.

It could be better to just rename it to just 'complete graph'?

There is an example that shows how many cooperators you need to overcome a defector

If I remember correctly, I think that example is no longer there. It was in the early form of the docs which were far too complicated, before we simplified the tutorials to being more modular. Not saying it's not worth including an example like that here, just pointing it out.

Nikoleta-v3 · 2016-07-13T16:28:06Z

I would be happy to work on something like that ! But maybe @drvinceknight is right, is not to technical for the specific section. It could go to http://axelrod-tournament.readthedocs.io/en/latest/, as an "Research Example - something" ?.

drvinceknight · 2016-07-13T16:32:58Z

But maybe @drvinceknight is right, is not to technical for the specific section. It could go to http://axelrod-tournament.readthedocs.io/en/latest/, as an "Research Example - something" ?.

I think I would prefer the tournament repo to stay as it is. I also think the docs should not try and do too much (as they did previously) but perhaps there is an interesting topolgy that could be shown.

Perhaps run two tournaments: one complete graph (remarking that this corresponds to the round robin), and then another that perhaps isolates Defector a bit (perhaps it only plays TitForTat which serves as a bridge/separator to the other strategies that continue to all play each other)? And then that would show how the cooperators do better?

marcharper · 2016-07-13T16:33:11Z

axelrod/match_generator.py

+
+        self.edges = edges
+        super(SpatialMatches, self).__init__(players, turns, game, repetitions)
+        """


Let's move the doc string up to just under the class SpatialMatches line. Can we clean up the language a bit? I'll put suggestions inline below.

drvinceknight · 2016-07-13T16:34:38Z

"Research Example - something" ?.

It could certainly be a great notebook to add to https://github.com/Axelrod-Python/Axelrod-notebooks

marcharper · 2016-07-13T16:37:49Z

axelrod/match_generator.py

+        as a 2D square lattice. We want to create a class that will not take as
+        an argument only a 2D lattice but any given graph.
+
+        After considering various python packages for graphs, we ended up using


Remove this paragraph?

marcharper · 2016-07-13T16:48:41Z

This is a nice contribution overall!

I put in lots of little comments above. It would be nice to have a more complex example somewhere in the docs or notebooks. It might also be interesting to augment the tournament repository with some new tournaments, say on a big cycle, a grid like in the literature, or some other interesting case.

Also two ideas for the future:

probabilistic ending + spatial games, and
Moran Process on directed graphs (I'll tackle this one)

Nikoleta-v3 · 2016-07-13T16:55:09Z

Thank you @marcharper ! I will go through the comments and make the corrections. Also as suggested I will create a few notebooks.

We were considering evolution in general but suggested that could be left for another time. I do believe from October I will spend a lot of time with Spatial tournaments so I appreciate the ideas !

drvinceknight · 2016-07-13T16:58:56Z

It might also be interesting to augment the tournament repository with some new tournaments, say on a big cycle, a grid like in the literature, or some other interesting case.

Sure: my main question would be how to pick the particular graph for the tournament repo (if a cycle: how do we order it?). Perhaps a discussion for another issue: have opened Axelrod-Python/tournament#36

probabilistic ending + spatial games, and

Yes. Maybe you could look in to that at some point @Nikoleta-v3 (not necessarily this Summer: there's a lot to do based on our chat this am although it would be pretty straightforward to implement).

I'll throw an issue together for it :)

Moran Process on directed graphs (I'll tackle this one)

Yes :) 👍

Which is for the TestResultSet_SpatialStructure is the test_resultset file.

Nikoleta-v3 · 2016-07-14T11:36:41Z

@marcharper I believe I have addressed all of your comments ! But I am still working on some notebooks for the documentation and for my dissertation as well.

Nikoleta-v3 added 18 commits July 13, 2016 10:47

Have tests for Spatial Match generator.

cb0e930

Fixing test that was passing by chance.

ba53219

This test was not correct as the number of repetitions is not the same as the number of interactions.

Fixed self interactions being counted twice.

eaee751

Adding three tests for spatial.

7733d86

These need to be refactored into test_resultsset.py

Adding test property based test for spatial.

27dc71e

Currently the underlying graph is cyclic it needs to be random.

Refactored tests.

6526c3b

Removed a print command that was accidentally left here.

613b924

Pep tests.

a71c65f

Created a documentation for the spatial tournament.

e525ca1

Last corrections on the spatial tournament docs.

cbc445d

Adding an extra test for a complete multigraph.

f94d022

In a complete multigraph alla the nodes are connected to each other and to themselves. Which in a round robin topology. Therefore, a test to make sure that the results procuded by a complete multigraph and a round robin are equivalent.

Removing property and property tests for Spatial tournament.

3a44249

After running the tests with Travis has returned the following error: Data generation is extremely slow. Decided to simply remove the tests.

marcharper reviewed Jul 13, 2016
View reviewed changes

Removing spatial tournament from import.

dfe7755

marcharper reviewed Jul 13, 2016
View reviewed changes

drvinceknight mentioned this pull request Jul 13, 2016

Include a spatial tournament Axelrod-Python/tournament#36

Open

drvinceknight mentioned this pull request Jul 13, 2016

Probabilistic ending + spatial tournament #655

Closed

Nikoleta-v3 added 6 commits July 14, 2016 00:04

Changing the word "dictionary" to "list".

16d7178

Fixing doc string for SpatialMatches class.

1e64dbc

Removed a forgotten test which is not being used.

0c65804

PEP8 corrections that were pointed out.

7a82a2c

Expanding the doc string of the test_match_lenghts.

5b3f81e

Which is for the TestResultSet_SpatialStructure is the test_resultset file.

Replacing "TestResult_SpatialStructure" to "TestResultSpatialStructure".

9fcaaa5

marcharper added ready-for-review 1-positive-review labels Jul 14, 2016

meatballs merged commit 1422925 into Axelrod-Python:master Jul 18, 2016

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

635 Implementing spatial tournaments #654

635 Implementing spatial tournaments #654

Nikoleta-v3 commented Jul 13, 2016 •

edited

marcharper Jul 13, 2016

marcharper commented Jul 13, 2016 •

edited

Nikoleta-v3 commented Jul 13, 2016

drvinceknight commented Jul 13, 2016

marcharper commented Jul 13, 2016

drvinceknight commented Jul 13, 2016

Nikoleta-v3 commented Jul 13, 2016

drvinceknight commented Jul 13, 2016

marcharper Jul 13, 2016

drvinceknight commented Jul 13, 2016

marcharper Jul 13, 2016

marcharper commented Jul 13, 2016

Nikoleta-v3 commented Jul 13, 2016

drvinceknight commented Jul 13, 2016

Nikoleta-v3 commented Jul 14, 2016

635 Implementing spatial tournaments #654

635 Implementing spatial tournaments #654

Conversation

Nikoleta-v3 commented Jul 13, 2016 • edited

marcharper Jul 13, 2016

Choose a reason for hiding this comment

marcharper commented Jul 13, 2016 • edited

Nikoleta-v3 commented Jul 13, 2016

drvinceknight commented Jul 13, 2016

marcharper commented Jul 13, 2016

drvinceknight commented Jul 13, 2016

Nikoleta-v3 commented Jul 13, 2016

drvinceknight commented Jul 13, 2016

marcharper Jul 13, 2016

Choose a reason for hiding this comment

drvinceknight commented Jul 13, 2016

marcharper Jul 13, 2016

Choose a reason for hiding this comment

marcharper commented Jul 13, 2016

Nikoleta-v3 commented Jul 13, 2016

drvinceknight commented Jul 13, 2016

Nikoleta-v3 commented Jul 14, 2016

Nikoleta-v3 commented Jul 13, 2016 •

edited

marcharper commented Jul 13, 2016 •

edited