Implemented Leyvraz, K86R from Axelrod's second #1153

gaffney2010 · 2017-12-14T08:37:13Z

Fingerprints below.

Pretty direct translation. Their funny if statements reduced a lot.

  IF (J1 * J .EQ. 1) GOTO 540

J1 and J

  IF (J2 * 2 + J1 + J * 2 + J .EQ. 1) GOTO 550

J2 cannot be 1, because then the whole left side would exceed 1. Ditto J. So this becomes J1=1.

  IF (J2 * 2 + J1 * 2 + J .EQ. 1) GOTO 560

Similarly J1 and J2 must equal zero. So this becomes J=1.

meatballs · 2017-12-14T09:52:57Z

axelrod/strategies/axelrod_second.py

+    2. If opponent Defected three turns ago, then Cooperate.
+    3. If opponent Defected two turns ago, then Defect.
+    4. If opponent Defected last turn, then Defect with prob 50%.
+    5. Otherwise (all Coopertaions), then Cooperate.


typo: Cooperations

meatballs · 2017-12-14T09:53:32Z

axelrod/strategies/axelrod_second.py

+        # Otherwise Defect, Cooperate, Cooperate, and increase `prob`
+        self.prob += 0.05
+        self.more_coop, self.last_generous_n_turns_ago = 2, 1
+        return self.try_return(D, lower_flags=False)



What happened here? This doesn't look like it should have changed.

Sorry, maybe I shouldn't have slipped this in here. I got rid of a useless else.

meatballs · 2017-12-14T09:58:36Z

axelrod/strategies/axelrod_second.py

+            return D
+        if recent_history[-1] == D:
+            return random_choice(0.5)
+        return C  # recent_history == [C, C, C]


Just a suggestion to consider rather than a request, but might this be easier to read with a dict mapping recent_history to action? Something like:

history_to_action = { [C, C, C]: C, [C, C, D]: random_choice(0.5) etc.... } return history_to_action[recent_history]

I think this actually is not ideal in this case (unless I'm misunderstanding):

[C, C, D]: random_choice(0.5)

Implies history_to_action would need to be redefined at every call of strategy to ensure the random sample is different which in turn means we'd be using the random module even when not needed (which bumps the seed and makes the strategy slower).

One way around it (if you did want to use a dict) would be to use the dict to map to probabilities:

{ (C, C, C): 1, (C, C, D): 0.5 }

and then use random_choice(history_to_action[recent_history]) (the random_choice is smart so if passed 0 or 1 doesn't sample).

I went ahead and implemented this. I agree that it's a little more readable.

drvinceknight · 2017-12-14T10:25:07Z

Strategy logic looks good and a correct implementation to me. Will leave the stylistic issue of the dict up to @gaffney2010 and @meatballs 👍

meatballs · 2017-12-14T10:27:04Z

axelrod/strategies/axelrod_second.py

+            (D, D, C): 1.0,
+            (D, D, D): 0.5
+        }
+


drvinceknight · 2017-12-14T10:28:41Z

axelrod/strategies/axelrod_second.py

+    1. If opponent Defected last two turns, then Defect with prob 75%.
+    2. If opponent Defected three turns ago, then Cooperate.
+    3. If opponent Defected two turns ago, then Defect.
+    4. If opponent Defected last turn, then Defect with prob 50%.


I think the docstrings don't match the description:

(D, D, D): 0.5

That doesn't correspond to If opponent Defected last turn, then Defect with prob 50%. as indicated for example by (D, C, D): 1.0?

Apologies: I see you said "ordered" rules.

You're right. I was hoping the tests would catch any issues. I'll go through this more carefully and re-run the fingerprints.

No sorry apologies again, that's still not exact then, according to the ordered description: (D, D, D) should defect with prob 75%.

Now that we have this dict, we could almost just describe the strategy by saying:

Depending on the 3 most recent actions by the opponent the strategy cooperates with probability: - If (C, C, C): probability 1; - If (C, C, D): probability 0.5; ...

and essentially write out the 8 cases that correspond to the dict (which indeed really helps with readability here 👍).

I think (D, D, D) should be 0.25, by the first rule. (These rules get applied in order.) These others seem okay; (D, C, D) is 1.0 by rule 2.

I think you are correct:

>>> import axelrod as axl >>> import axelrod_fortran as axlf >>> axl.seed(0) >>> players = (axl.Defector(), axlf.Player("k68r")) >>> match = axl.Match(players, turns=1000) >>> results = match.play() >>> match.cooperation() (0,240)

so k68r is cooperating 25% of the time against D, D, D.

Whereas your current branch (as of 890d930):

>>> players = (axl.Defector(), axl.Leyvraz()) >>> match = axl.Match(players, turns=1000) >>> axl.seed(0) >>> results = match.play() >>> match.cooperation() (0, 508)

A direct map of the previous implementation:

- if recent_history[-1] == D and recent_history[-2] == D: - return random_choice(0.25) - if recent_history[-3] == D: - return C - if recent_history[-2] == D: - return D - if recent_history[-1] == D: - return random_choice(0.5) - return C # recent_history == [C, C, C]

gives:

{ (C, D, D): 0.25, (D, D, D): 0.25, (D, C, C): 1, (D, C, D): 1, (D, D, C): 1, (C, D, C): 0, (C, C, D): 0.5, (C, C, C): 1 }

which I believe gives the same as you have, apart from, as you say (D, D, D).

Would be good to confirm the fps just in case 👍

gaffney2010 · 2017-12-14T10:47:07Z

Re-updated with fixed probability. I added some comments and re-ran FPs for more confidence here.

drvinceknight · 2017-12-14T10:51:22Z

FPs look good.

drvinceknight · 2017-12-14T10:53:43Z

axelrod/tests/strategies/test_axelrod_second.py

+        'manipulates_source': False,
+        'manipulates_state': False
+    }
+


Could we add a test_init_of_dict(self) just to add a second anchor for the internal dictionary (to help avoid any future changes that might re introduce the error we just caught).

def test_init_of_dict(self): player = self.player() expected_prob_coop = {...} self.assertEqual(player.prob_coop, expected_proob_coop)

drvinceknight · 2017-12-14T10:57:05Z

For completeness have rerun at dacc9c8:

>>> players = (axl.Defector(), axl.Leyvraz())
>>> match = axl.Match(players, turns=1000)
>>> axl.seed(0)
>>> results = match.play()
>>> match.cooperation()
(0, 248)

👍

drvinceknight · 2017-12-14T11:51:49Z

axelrod/strategies/axelrod_second.py

+            if len(opponent.history) >= go_back:
+                recent_history[-go_back] = opponent.history[-go_back]
+
+        return random_choice(self.prob_coop[tuple(recent_history)])


I've hunted down why the type testing is failing on travis. It doesn't know what size tuple tuple(recent_history) returns so you need to change this to be:

1508 return random_choice(self.prob_coop[(recent_history[-3], │ + 1509 recent_history[-2], │ + 1510 recent_history[-1])])

So specifically say this is a tuple of size 3, which is a bit of a mouthful.

drvinceknight · 2017-12-14T21:43:00Z

Thanks for making all the changes 👍

gaffney2010 added 2 commits December 14, 2017 04:04

Implemented Leyvraz, K86R from Axelrod's second

c342743

Changed indentation in docstrings.

e60c63c

meatballs requested changes Dec 14, 2017

View reviewed changes

meatballs reviewed Dec 14, 2017

View reviewed changes

axelrod/strategies/axelrod_second.py Outdated

(D, D, C): 1.0,

(D, D, D): 0.5

}

Copy link

Member

meatballs Dec 14, 2017

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nice!

drvinceknight reacted with thumbs up emoji

meatballs approved these changes Dec 14, 2017

View reviewed changes

drvinceknight requested changes Dec 14, 2017

View reviewed changes

meatballs added the ready-to-merge label Dec 14, 2017

drvinceknight removed the ready-to-merge label Dec 14, 2017

Fixed typo in docstrings.

aa5fa8a

Moved probabilities in Leyvraz to dict for readability.

890d930

drvinceknight requested changes Dec 14, 2017

View reviewed changes

Fixed a proability in Leyvraz.

dacc9c8

drvinceknight requested changes Dec 14, 2017

View reviewed changes

Changed to tuple of explicit length.

835764a

drvinceknight approved these changes Dec 14, 2017

View reviewed changes

drvinceknight merged commit bc9cd5a into Axelrod-Python:master Dec 14, 2017

drvinceknight mentioned this pull request Dec 14, 2017

After next (>v3.10.0) release: strategies to update Axelrod-Python/axelrod-fortran#72

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Implemented Leyvraz, K86R from Axelrod's second #1153

Implemented Leyvraz, K86R from Axelrod's second #1153

gaffney2010 commented Dec 14, 2017

meatballs Dec 14, 2017

meatballs Dec 14, 2017

gaffney2010 Dec 14, 2017

meatballs Dec 14, 2017

drvinceknight Dec 14, 2017

drvinceknight Dec 14, 2017 •

edited

gaffney2010 Dec 14, 2017

drvinceknight commented Dec 14, 2017

meatballs Dec 14, 2017

drvinceknight Dec 14, 2017

drvinceknight Dec 14, 2017

gaffney2010 Dec 14, 2017

drvinceknight Dec 14, 2017

gaffney2010 Dec 14, 2017

drvinceknight Dec 14, 2017 •

edited

drvinceknight Dec 14, 2017 •

edited

drvinceknight Dec 14, 2017

gaffney2010 commented Dec 14, 2017

drvinceknight commented Dec 14, 2017

drvinceknight Dec 14, 2017 •

edited

drvinceknight commented Dec 14, 2017

drvinceknight Dec 14, 2017

drvinceknight commented Dec 14, 2017

Implemented Leyvraz, K86R from Axelrod's second #1153

Implemented Leyvraz, K86R from Axelrod's second #1153

Conversation

gaffney2010 commented Dec 14, 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

drvinceknight Dec 14, 2017 • edited

Choose a reason for hiding this comment

Choose a reason for hiding this comment

drvinceknight commented Dec 14, 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

drvinceknight Dec 14, 2017 • edited

Choose a reason for hiding this comment

drvinceknight Dec 14, 2017 • edited

Choose a reason for hiding this comment

Choose a reason for hiding this comment

gaffney2010 commented Dec 14, 2017

drvinceknight commented Dec 14, 2017

drvinceknight Dec 14, 2017 • edited

Choose a reason for hiding this comment

drvinceknight commented Dec 14, 2017

Choose a reason for hiding this comment

drvinceknight commented Dec 14, 2017

drvinceknight Dec 14, 2017 •

edited

drvinceknight Dec 14, 2017 •

edited

drvinceknight Dec 14, 2017 •

edited

drvinceknight Dec 14, 2017 •

edited