Reconcile Moran processes #1043

drvinceknight · 2017-06-10T22:12:40Z

Use spatial Moran Processes throughout.

Also add ability to pass a prob_end to the Matches.

In terms of performance this actually seems faster:

%%timeit
# Branch implementation of Moran Process
for players in [[axl.Cooperator()] * 3 + [axl.Defector()] * 3,
                [axl.Random(), axl.Grudger(), axl.TitFor2Tats()],
                [s() for s in axl.demo_strategies]]:

    for seed in range(100):
        axl.seed(seed)
        mp = axl.MoranProcess(players)
        mp.play()
        winner = mp.winning_strategy_name

output: 1 loop, best of 3: 8.7 s per loop

whereas current master:

%%timeit
# Master branch implementation of Moran Process
for players in [[axl.Cooperator()] * 3 + [axl.Defector()] * 3,
                [axl.Random(), axl.Grudger(), axl.TitFor2Tats()],
                [s() for s in axl.demo_strategies]]:

    for seed in range(100):
        axl.seed(seed)
        mp = axl.MoranProcess(players)
        mp.play()
        winner = mp.winning_strategy_name

Output: 1 loop, best of 3: 9.92 s per loop

This complements #1042.

Use spatial Moran Processes throughout Also add ability to pass a prob_end to the Matches.

marcharper · 2017-06-10T23:10:02Z

docs/tutorials/further_topics/moran_processes_on_graphs.rst

@@ -4,7 +4,7 @@ Moran Process on Graphs
 =======================

 The library also provides a graph-based Moran process [Shakarian2013]_ with
-:code:`MoranProcessGraph`.  To use this class you must supply at least one
+:code:`MoranProcess`.  To use this class you must supply at least one


class --> feature since the class can be used without a graph.

marcharper · 2017-06-10T23:12:55Z

axelrod/moran.py

                 deterministic_cache=None, mutation_rate=0., mode='bd',
-                 match_class=Match):
+                 match_class=Match, interaction_graph=None,


Do we need match_class any longer?

I don't think so, I'll take it out.

marcharper · 2017-06-10T23:13:39Z

axelrod/moran.py

+            interaction_graph = complete_graph(len(players), loops=False)
+            if reproduction_graph is None:
+                reproduction_graph = complete_graph(len(players))
+        if reproduction_graph is None:


Is this the right default? I think we should consider using the same default in all cases.

This is the same default as it was implemented in MoranProcessGraph but I'll change it. I've got a further idea for how I'd like to refactor this as well.

marcharper · 2017-06-10T23:14:07Z

axelrod/moran.py

@@ -124,15 +159,21 @@ def mutate(self, index):
        return new_player

    def death(self, index=None):
-        """Selects the player to be removed. Note that the in the birth-death


This docstring is still relevant, no?

Yup, not sure why I got rid of it.

marcharper · 2017-06-10T23:16:22Z

One concern is that without the non-graph implementation we don't have a canonical implementation to compare to. We should consider adding some tests that really demonstrate that it is equivalent (e.g. explicitly tests the interaction pairs for various population sizes).

drvinceknight · 2017-06-11T10:26:02Z

We should consider adding some tests that really demonstrate that it is equivalent (e.g. explicitly tests the interaction pairs for various population sizes).

Sure. I'm going to add to the tests a bit anyway, I'll try to incorporate your suggestion but I might need more guidance. 👍

- Add ability for a graph to add loops. - Ensure reproduction graph is same as inter graph - Remove Match class option from Moran process. - Uncomment property based test. Slipped by in a previous commit on this branch as a mistake. - Fix docstrings and add type hints. Addresses #808. Changed `fitness_proportionate_selection` to make mypy happy. - s/random.seed/axelrod.seed - Fix docs - Add numerous tests. No specific interactions tests: I feel this is fairly well covered now with some behaviour specifically tested as well.

drvinceknight · 2017-06-11T16:34:14Z

I've just pushed a bunch of changes. Amongst other things I've added numpy docstrings and type hints. I've also gone through and added more unit type tests. I took a look at adding specific interaction tests but I'm not sure it's necessary:

Each method is now tested (it wasn't before);
There are some tests that do test the expected overall outcomes.

I'm not at all averse to adding more tests if you're happy to spell them out for me (or perhaps they can be left as a further PR). 👍

marcharper · 2017-06-11T17:30:55Z

axelrod/moran.py

@@ -389,14 +395,15 @@ class ApproximateMoranProcess(MoranProcess):
    Instead of playing the matches, the result is sampled
    from a dictionary of player tuples to distribution of match outcomes
    """
-    def __init__(self, players, cached_outcomes, mutation_rate=0.):
+    def __init__(self, players: List[Player], cached_outcomes: dict,


We need all the parameters to pass through?

Are you asking about the turns prob_end etc? In which case: no we don't. They are not used as all match results sampled from the cached_outcomes.

If you're asking about the graphs. I didn't want to tackle that here. It would be an easy thing to add.

It would be an easy thing to add.

But let's do that at a later date.

marcharper · 2017-06-11T17:42:08Z

axelrod/tests/unit/test_moran.py

        mp = MoranProcess(players)
        populations = mp.play()
        self.assertEqual(len(mp), 9)
        self.assertEqual(len(populations), 9)
        self.assertEqual(populations, mp.populations)
        self.assertEqual(mp.winning_strategy_name, str(axelrod.Defector()))

-    def test_standard_fixation(self):


Why remove this test?

Because it was for a mock match class (it was a test for the match_class attribute).

(It shouldn't have been called test_standard_fixation)

Actually it tested the constant fitness case (or else, that was the original intent).

Most of the tests in here were testing multiple things (which I've tried to address). This test also tested that you could pass a different match class and has been removed because I've removed the match_class attribute (as you suggested earlier).

If you wanted to test constant fitness we can do that by passing a cache where all strategies have the same score.

Actually, I don't understand what that test was doing at all with the Mock class (apart from testing that it worked). I've fixed the py3.6 problems and about to push with a version of this test (just using a normal match).

It was returning a constant score (fitness) in a 2:1 ratio to see if the outcome was in the ballpark of what would be expected for an r=2 Moran process. This appears to be what the new test essentially does.

Cool :) Let me know if you think there's anything else that we can add.

marcharper · 2017-06-11T17:42:29Z

docs/tutorials/further_topics/moran_processes_on_graphs.rst

-Graphs are undirected by default. Various intermediates such as the list of
-neighbors are cached for efficiency by the graph object.
+Graphs are undirected by default but you can pass :code:`directed=True` to
+create a directed graph.. Various intermediates such as the list of neighbors


drvinceknight · 2017-06-11T18:18:36Z

Failures are py3.6 vs py3.5... Working on basically ensuring the dicts are ordered as expected.

drvinceknight · 2017-06-11T19:07:44Z

5c26eee gets tests to pass on both py3.6 and py3.5
b3960ca adds a constant fitness test case

Reconcile Moran processes

f16e44b

Use spatial Moran Processes throughout Also add ability to pass a prob_end to the Matches.

marcharper reviewed Jun 10, 2017

View reviewed changes

drvinceknight added the ready-for-review label Jun 11, 2017

Formatting fixes

20ed823

marcharper reviewed Jun 11, 2017

View reviewed changes

Remove a typo.

bc87892

drvinceknight added 2 commits June 11, 2017 19:46

Sort everything to ensure py3.5/3.6 compatibility.

5c26eee

Add a constant fitness test.

b3960ca

Uncomment a test.

ce39975

marcharper approved these changes Jun 12, 2017

View reviewed changes

drvinceknight added ready-to-merge and removed ready-for-review labels Jun 12, 2017

meatballs merged commit 49fc54b into master Jun 13, 2017

meatballs deleted the reconcile-moran-b branch June 13, 2017 07:48

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Reconcile Moran processes #1043

Reconcile Moran processes #1043

drvinceknight commented Jun 10, 2017

marcharper Jun 10, 2017

marcharper Jun 10, 2017

drvinceknight Jun 11, 2017

marcharper Jun 10, 2017

drvinceknight Jun 11, 2017

marcharper Jun 10, 2017

drvinceknight Jun 11, 2017

marcharper commented Jun 10, 2017

drvinceknight commented Jun 11, 2017

drvinceknight commented Jun 11, 2017

marcharper Jun 11, 2017

drvinceknight Jun 11, 2017

drvinceknight Jun 11, 2017

marcharper Jun 11, 2017

drvinceknight Jun 11, 2017 •

edited

Loading

marcharper Jun 11, 2017

drvinceknight Jun 11, 2017

drvinceknight Jun 11, 2017

drvinceknight Jun 11, 2017 •

edited

Loading

marcharper Jun 11, 2017

drvinceknight Jun 11, 2017

marcharper Jun 11, 2017

drvinceknight commented Jun 11, 2017

drvinceknight commented Jun 11, 2017

Reconcile Moran processes #1043

Reconcile Moran processes #1043

Conversation

drvinceknight commented Jun 10, 2017

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

marcharper commented Jun 10, 2017

drvinceknight commented Jun 11, 2017

drvinceknight commented Jun 11, 2017

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 Jun 11, 2017 • edited Loading

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 Jun 11, 2017 • edited Loading

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 Jun 11, 2017

drvinceknight commented Jun 11, 2017

drvinceknight Jun 11, 2017 •

edited

Loading

drvinceknight Jun 11, 2017 •

edited

Loading