just finite_state_machines #956

eric-s-s · 2017-04-01T18:39:16Z

#884
(that last commit should read "fixed type_hints". resolving a merge when two people had done almost the same work at almost the same time, was an ... unexpected surprise)

did a lot of fixing to FSM and its testing strategy. it looked like this should be its own PR separate from the other three i was planning to group it with. there are some issues with some of the FSMPlayers. They have states that are impossible to reach.

I tried to come up with a decent testing strategy, and need another opinion(s).

marcharper · 2017-04-01T20:27:00Z

It's ok if there are unreachable states, I think. Some of these are the output of ML training algorithms (anything that has Evolved in the name), and reachability wasn't enforced. That said, if we can show that removing some of the states does not affect the strategy at all then I'm in favor of simplifying them.

marcharper · 2017-04-01T20:28:41Z

axelrod/strategies/finite_state_machines.py

@@ -101,14 +123,15 @@ class Fortress3(FSMPlayer):
    }

    def __init__(self) -> None:
-        transitions = (
-            (1, D, 2, D),
+        transitions = [


What's the reason for the change to a list instead of a tuple?

the FSM and FSMPlayer used a list. then it got changed to tuple in all the subsequent ones. i spent about 15 minutes scratching my head trying to figure out why mypy was complaining before i noticed the difference. i used list because that's what i saw first and they all needed to be the same.

should i change all to tuples or leave as is?

My preference is tuples since everything involved is immutable.

ok. will do.

mypy gave me trouble on the change. it only worked it the type hints were a generalized tuple. it did have the nice advantage that i could set the defaults in the kwargs and not after the fact.

marcharper · 2017-04-01T20:30:40Z

axelrod/tests/strategies/test_finite_state_machines.py

@@ -174,13 +265,46 @@ class TestPredator(TestFSMPlayer):
        'manipulates_source': False,
        'manipulates_state': False
    }
+    """
+    transitions = [
+                    (0, C, 0, D),  NO ROUTE


These states are still active on the first move? There's just no route back?

This is a named strategy from the literature, FYI

they are not active on the first move because "initial_state" is set to one. should initial state get set to 0?

DOI:10.1109/CEC.2006.1688322.

i was wondering what that was!

It's supposed to start at 0, good catch!

marcharper · 2017-04-01T20:32:07Z

axelrod/strategies/finite_state_machines.py

+            return False
+        return (self._state, self._state_transitions) == (other.state, other.state_transitions)
+
+    def __ne__(self, other):


Do we not get __ne__ for free by defining __eq__?

i know in python2 that was not the case and made for some odd behavior. here are the links for py2 and py3. it LOOKS like they changed it. my paranoid reaction is to leave the lines and not risk the bug, but i'll definitely leave that decision in your hands.

python 2

python 3

I believe it's no longer necessary in py3, can we remove it (less lines of code to maintain :)) and just have a test that checks we have the __ne__ behaviour?

marcharper · 2017-04-01T20:32:56Z

axelrod/strategies/finite_state_machines.py

@@ -55,16 +81,16 @@ class FSMPlayer(Player):
        'manipulates_state': False
    }

-    def __init__(self, transitions=None, initial_state=None,
-                 initial_action=None) -> None:
+    def __init__(self, transitions: List[Tuple[int, Any, int, Any]] =None, initial_state: int =None,


Please check the spacing on the function definition

do you mean def something(num: int =5): should be def something(num: int = 5)? if that's it, i wasn't sure of the spacing convention for type hints.

Yes, that's the convention we've been trying to use.

eric-s-s · 2017-04-02T04:48:49Z

@marcharper

That said, if we can show that removing some of the states does not affect the strategy at all then I'm in favor of simplifying them.

should they be commented out with a note about the ML? should they be deleted and numbers be left as is? The class names would be confusing otherwise.

here's a test i wrote, that prints out proof of inaccessibility.

    def test_all_states_reachable(self):
        player = self.player()
        initial_state = player.initial_state
        transitions = player.fsm.state_transitions
        
        called_states = set(pair[0] for pair in transitions.values())
        called_states.add(initial_state)
        
        owned_states = set(pair[0] for pair in transitions.keys())

        un_callable = owned_states.difference(called_states)
        if un_callable:
            print('for player: {}, the following states are un-callable: {}'.format(self.player, un_callable))

output was.

for player: <class 'axelrod.strategies.finite_state_machines.EvolvedFSM16Noise05'>, the following states are un-callable: {9, 7}

and

for player: <class 'axelrod.strategies.finite_state_machines.EvolvedFSM16'>, the following states are un-callable: {9, 4}

marcharper · 2017-04-02T16:34:14Z

I vote to just delete the unreachable states, and let's leave that test in for future submissions. If something goes wrong we can look back in the git history for the deleted transitions.

drvinceknight · 2017-04-02T19:12:42Z

axelrod/tests/strategies/test_finite_state_machines.py

-        if un_callable_states:
-            raise AssertionError('The following states are un-callable: {}'.format(un_callable_states))
+        error_message = 'The following states are un-reachable: {}'.format(list(un_callable_states))
+        self.assertEqual(error_message, 'The following states are un-reachable: []')


What's this supposed to be testing? I'm a bit confused by the need for this line...

from @marcharper

I vote to just delete the unreachable states, and let's leave that test in for future submissions.

it's supposed to test whether an FSM has states that can't actually be reached, as was found for the two EvolvedFSM classes . i thought the lines i originally wrote were clear:

if un_callable_states: raise AssertionError('The following states are un-callable: {}'.format(un_callable_states)

but the coverage tool won't allow them, because the "if" clause is never called. and now that i wrote down my explanation, i see what to do and feel kinda dumb. please wait while i fix that. just a moment.

drvinceknight · 2017-04-02T20:02:36Z

axelrod/tests/strategies/test_finite_state_machines.py

-        error_message = 'The following states are un-reachable: {}'.format(list(un_callable_states))
-        self.assertEqual(error_message, 'The following states are un-reachable: []')
+        extra_info = 'The following states are un-reachable: {}'.format(un_callable_states)
+        self.assertEqual(un_callable_states, set(), msg=extra_info)


That makes sense :D 👍

drvinceknight · 2017-04-04T08:55:59Z

axelrod/tests/strategies/test_finite_state_machines.py

+            # (7, D, 15, C), states 7 or 9
+            (8, C, 2, C),
+            (8, D, 4, D),
+            # (9, C, 15, D),


I'm hoping to work through this properly today, can these commented out states simply be removed from the docstring or do we need them here for completeness?

thanks for the catch! i forgot them, and they need to be changed to tuples regardless. i'll await a decision and act on it.

Let's remove them

drvinceknight · 2017-04-04T20:42:07Z

This looks good to me but will leave @marcharper to confirm as he's started looking at it (I'm fine for this to be merged Marc).

drvinceknight · 2017-04-11T07:40:12Z

I believe this PR addresses a bug present in v2.8.0:

With v2.8.0:

>>> axl.seed(0)
>>> C, D = axl.Actions.C, axl.Actions.D 
>>> transitions = [(0, C, 0, C), (0, D, 3, C),                                                                   
...                (1, C, 5, D), (1, D, 0, C),                                                                   
...                (2, C, 3, C), (2, D, 2, D),                                                                   
...                (3, C, 4, D), (3, D, 6, D),                                                                   
...                (4, C, 3, C), (4, D, 1, D),                                                                   
...                (5, C, 6, C), (5, D, 3, D),                                                                   
...                (6, C, 6, D), (6, D, 6, D),                                                                   
...                (7, C, 7, D), (7, D, 5, C)]
>>> pair = (axl.FSMPlayer(transitions=transitions), axl.Random())
>>> match = axl.Match(pair, turns=5)
>>> match.play()
---------------------------------------------------------------------------
KeyError                                  Traceback (most recent call last)
<ipython-input-3-e0ed4b66431c> in <module>()
     11 pair = (axl.FSMPlayer(transitions=transitions), axl.Random())
     12 match = axl.Match(pair, turns=20)
---> 13 match.play()

/home/vince/anaconda3/envs/moran/lib/python3.6/site-packages/axelrod/match.py in play(self)
    119             while turn < self.turns:
    120                 turn += 1
--> 121                 self.players[0].play(self.players[1], self.noise)
    122             result = list(
    123                 zip(self.players[0].history, self.players[1].history))

/home/vince/anaconda3/envs/moran/lib/python3.6/site-packages/axelrod/player.py in play(self, opponent, noise)
    166     def play(self, opponent, noise=0):
    167         """This pits two players against each other."""
--> 168         s1, s2 = self.strategy(opponent), opponent.strategy(self)
    169         if noise:
    170             s1, s2 = self._add_noise(noise, s1, s2)

/home/vince/anaconda3/envs/moran/lib/python3.6/site-packages/axelrod/strategies/finite_state_machines.py in strategy(self, opponent)
     73             return self.initial_action
     74         else:
---> 75             action = self.fsm.move(opponent.history[-1])
     76             # Record the state for testing purposes, this isn't necessary
     77             # for the strategy to function

/home/vince/anaconda3/envs/moran/lib/python3.6/site-packages/axelrod/strategies/finite_state_machines.py in move(self, opponent_action)
     29     def move(self, opponent_action: Action) -> Action:
     30         """Computes the response move and changes state."""
---> 31         next_state, next_action = self.state_transitions[(self.state, opponent_action)]
     32         self.state = next_state
     33         return next_action

KeyError: (None, 'D')

With current master at 09d13eb:

>>> axl.seed(0)
>>> C, D = axl.Actions.C, axl.Actions.D 
>>> transitions = [(0, C, 0, C), (0, D, 3, C),                                                                   
...                (1, C, 5, D), (1, D, 0, C),                                                                   
...                (2, C, 3, C), (2, D, 2, D),                                                                   
...                (3, C, 4, D), (3, D, 6, D),                                                                   
...                (4, C, 3, C), (4, D, 1, D),                                                                   
...                (5, C, 6, C), (5, D, 3, D),                                                                   
...                (6, C, 6, D), (6, D, 6, D),                                                                   
...                (7, C, 7, D), (7, D, 5, C)]
>>> pair = (axl.FSMPlayer(transitions=transitions), axl.Random())
>>> match = axl.Match(pair, turns=5)
>>> match.play()
[('C', 'D'),
 ('C', 'D'),
 ('C', 'C'),
 ('D', 'C'),
 ('C', 'D')]

drvinceknight · 2017-04-11T07:40:35Z

I might throw that in as a test so we know it doesn't reappear.

I found a bug working with v2.8.0, but Axelrod-Python#956 seems to have clear it up. I'm just adding a test to confirm it.

drvinceknight · 2017-04-11T08:06:51Z

I've opened #967

eric-s-s added 8 commits March 31, 2017 15:18

mypy vs FSM grrrrrr

31d785c

before bed

43a3d62

switch branch

c6e7c8e

set up fsm tests. branch switch

83a8457

i now despise FSM

2ab3332

PR only for FSM

f5ff05a

need to fix merge issue with master

0ce7efe

fixed type_hints mess

eedea65

marcharper requested changes Apr 1, 2017

View reviewed changes

eric-s-s added 2 commits April 2, 2017 22:14

possibly resolved all issues. switching branches

350b0e4

keeping current with *args to **kwargs

1f988d1

eric-s-s added 4 commits April 3, 2017 02:19

made required changes. mypy pickiness forced simplified type hints

c674567

merge master to deal with repr difference.

18906dd

i think that works

eea189c

did not see that coming. fixed coverage

143383d

drvinceknight reviewed Apr 2, 2017

View reviewed changes

sigh. brain not working

abfd8ef

drvinceknight reviewed Apr 2, 2017

View reviewed changes

drvinceknight reviewed Apr 4, 2017

View reviewed changes

small change to test comment strings

07a43a2

marcharper approved these changes Apr 6, 2017

View reviewed changes

marcharper merged commit 09d13eb into Axelrod-Python:master Apr 6, 2017

drvinceknight added a commit to drvinceknight/Axelrod that referenced this pull request Apr 11, 2017

Add an extra test for the FSM players

4b76d14

I found a bug working with v2.8.0, but Axelrod-Python#956 seems to have clear it up. I'm just adding a test to confirm it.

drvinceknight mentioned this pull request Apr 11, 2017

Add an extra test for the FSM players #967

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

just finite_state_machines #956

just finite_state_machines #956

eric-s-s commented Apr 1, 2017 •

edited

Loading

marcharper commented Apr 1, 2017 •

edited

Loading

marcharper Apr 1, 2017

eric-s-s Apr 2, 2017 •

edited

Loading

marcharper Apr 2, 2017

eric-s-s Apr 2, 2017

eric-s-s Apr 2, 2017

marcharper Apr 1, 2017

marcharper Apr 1, 2017

eric-s-s Apr 2, 2017

marcharper Apr 2, 2017

marcharper Apr 1, 2017

eric-s-s Apr 2, 2017 •

edited

Loading

drvinceknight Apr 2, 2017

marcharper Apr 1, 2017

eric-s-s Apr 2, 2017

marcharper Apr 2, 2017

eric-s-s commented Apr 2, 2017 •

edited

Loading

marcharper commented Apr 2, 2017

drvinceknight Apr 2, 2017

eric-s-s Apr 2, 2017 •

edited

Loading

drvinceknight Apr 2, 2017

drvinceknight Apr 4, 2017

eric-s-s Apr 4, 2017

marcharper Apr 4, 2017

drvinceknight commented Apr 4, 2017

drvinceknight commented Apr 11, 2017 •

edited

Loading

drvinceknight commented Apr 11, 2017

drvinceknight commented Apr 11, 2017

just finite_state_machines #956

just finite_state_machines #956

Conversation

eric-s-s commented Apr 1, 2017 • edited Loading

marcharper commented Apr 1, 2017 • edited Loading

Choose a reason for hiding this comment

eric-s-s Apr 2, 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

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

eric-s-s Apr 2, 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

Choose a reason for hiding this comment

eric-s-s commented Apr 2, 2017 • edited Loading

marcharper commented Apr 2, 2017

Choose a reason for hiding this comment

eric-s-s Apr 2, 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

Choose a reason for hiding this comment

drvinceknight commented Apr 4, 2017

drvinceknight commented Apr 11, 2017 • edited Loading

drvinceknight commented Apr 11, 2017

drvinceknight commented Apr 11, 2017

eric-s-s commented Apr 1, 2017 •

edited

Loading

marcharper commented Apr 1, 2017 •

edited

Loading

eric-s-s Apr 2, 2017 •

edited

Loading

eric-s-s Apr 2, 2017 •

edited

Loading

eric-s-s commented Apr 2, 2017 •

edited

Loading

eric-s-s Apr 2, 2017 •

edited

Loading

drvinceknight commented Apr 11, 2017 •

edited

Loading