Add a new DynamicTwoTitsForTat strategy #1030

FakeNameSE · 2017-05-28T18:26:39Z

Adds the new strategy DynamicTwoTitsForTat, based off of the TwoTitsForTat stategy by adding a probability of forgiveness based off of the ratio of the opponent's cooperations to total moves (so their rough probability of cooperation).

Link to original request: #1027

…ing TwoTitsForTat stategy by adding a probability of forgiveness based off of the ratio of the opponent's cooperations to total moves (so their rough probability of cooperation). Signed-off-by: FakeNameSE <grantlycee@gmail.com>

FakeNameSE · 2017-05-28T18:50:26Z

My edits passed the doctest and my unittest locally.

drvinceknight

This is looking really good @FakeNameSE!

A couple of little things from me. The tests are failing at an integration test that checks that all deterministic strategies act the same way when repeated. However this strategy does not but that's simply because it hasn't been classified correctly (I've noted where it should have "stochastic": True).

I think I'm ok with the name because it is punishing a defection with 2 defections. 👍

Let me know if anything I've asked for isn't clear. Looking forward to getting this strategy in :)

drvinceknight · 2017-05-28T19:00:32Z

axelrod/strategies/titfortat.py

+        if D in opponent.history[-2:]:
+            # Probability of turning the other cheek based off of 
+            # opponent's probability of defection
+            if random.random() < (opponent.cooperations / len(opponent.history)):


Replace with random_choice(opponent.cooperations / len(opponent.history))

There are internal reasons why we use random_choice. It is implemented in a way that it won't sample a random number if the probability is 0 or 1, which in turn means that the random state won't be offset when strategies act deterministicaly :)

For example: if we were to play a Defector there would not actually be anything random here (the probability would always be 0) , so random_choice would not actually sample any random numbers.

Got it. Unfortunately, after changing that I now get:

FAIL: test_strategy (axelrod.tests.strategies.test_titfortat.TestDynamicTwoTitsForTat) ---------------------------------------------------------------------- Traceback (most recent call last): File "/home/grant/Axelrod/axelrod/tests/strategies/test_titfortat.py", line 132, in test_strategy self.second_play_test(rCC=C, rCD=D, rDC=C, rDD=D) File "/home/grant/Axelrod/axelrod/tests/strategies/test_player.py", line 484, in second_play_test rCD, C, D, seed=seed) File "/home/grant/Axelrod/axelrod/tests/strategies/test_player.py", line 361, in test_responses test_class.assertEqual(s1, response) AssertionError: 'C' != 'D' - C + D

I think your player tests are testing things sufficiently so I'd just remove the second_play_test here (the fact that it failed is based on the history being different and the probabilities being sampled correctly).

Thank you, I now get

====================================================================== FAIL: test_strategy (axelrod.tests.strategies.test_titfortat.TestDynamicTwoTitsForTat) ---------------------------------------------------------------------- Traceback (most recent call last): File "/home/grant/Axelrod/axelrod/tests/strategies/test_titfortat.py", line 139, in test_strategy self.versus_test(opponent, expected_actions=actions) File "/home/grant/Axelrod/axelrod/tests/strategies/test_player.py", line 535, in versus_test self.assertEqual(match.play(), expected_actions) AssertionError: Lists differ: [('C', 'D'), ('C', 'C'), ('C', 'C'), ('C', 'D'), ('C', 'C')] != [('C', 'D'), ('D', 'C'), ('D', 'C'), ('C', 'D'), ('D', 'C')] First differing element 1: ('C', 'C') ('D', 'C') - [('C', 'D'), ('C', 'C'), ('C', 'C'), ('C', 'D'), ('C', 'C')] + [('C', 'D'), ('D', 'C'), ('D', 'C'), ('C', 'D'), ('D', 'C')]

I always get (C,C) for the last iteration.

Please push your code, it's difficult for me to debug it without seeing it (there could be an error in the strategy now or a number of other things). I'm going to sleep now but will look at it in the morning. We'll figure it out.

How should I push it? Otherwise, it is here: https://github.com/FakeNameSE/Axelrod

That link shows me the same branch at the same stage of what I'm seeing on this PR. Commit your changes to the same branch and push them (they'll automatically appear here). Let me know if you don't know what I mean and I'll give a detailed explanation when I have a moment.

Sorry, I forgot to, Done

drvinceknight · 2017-05-28T19:01:35Z

axelrod/tests/strategies/test_titfortat.py

+        # Will defect twice when last turn of opponent was defection.
+        opponent = axelrod.MockPlayer(actions=[D, C, C, D, C])
+        actions = [(C, D), (D, C), (D, C), (C, D), (D, C)]
+        self.versus_test(opponent, expected_actions=actions)


The mock player test will need to be seeded: http://axelrod.readthedocs.io/en/stable/tutorials/contributing/strategy/writing_test_for_the_new_strategy.html

self.versus_test(axelrod.Alternator(), expected_actions=actions, seed=0)

(I expect the others won't as their probabilities are 0 or 1 right?)

Once seeded can you find two seeds where the player acts differently for different seeds (thus showing that it's stochastic).

I'm not sure I follow. I choose a different seed, and then do I repeat the same test?

I'm not sure I follow. I choose a different seed, and then do I repeat the same test?

That's right with different actions.

To illustrate:

>>> p = 0.5 # for example opponent.history = [C, C, D, D] >>> axl.seed(0) >>> axl.random_choice(p) 'D' >>> axl.seed(1) >>> axl.random_choice(p) 'C'

So with different seeds the random choice gives different results.

So in your case you'd have something like:

actions = [..., (D, D)] self.versus_test(opponent=opponent, expected_actions=actions, seed=1) actions = [..., (C, D)] # Different actions self.versus_test(opponent=opponent, expected_actions=actions, seed=0) # Different seed

This just shows that the strategy does act randomly.

drvinceknight · 2017-05-28T19:07:02Z

axelrod/strategies/titfortat.py

+    name = 'Dynamic Two Tits For Tat'
+    classifier = {
+        'memory_depth': 2,  # Long memory, memory-2
+        'stochastic': False,


This strategy is stochastic (it acts randomly).

drvinceknight · 2017-05-28T19:09:02Z

axelrod/strategies/titfortat.py

+            return C
+        if D in opponent.history[-2:]:
+            # Probability of turning the other cheek based off of 
+            # opponent's probability of defection


This comment is ambiguous. I would remove it.

drvinceknight · 2017-05-28T19:09:41Z

axelrod/strategies/titfortat.py

+    opponent with a dynamic bias based off of the opponents ratio of 
+    cooperations to total moves (so their current probability of 
+    cooperating towards cooporating regardless of the move 
+    (aka: forgiveness)."""


Missing a closing bracket.

drvinceknight · 2017-05-28T19:11:40Z

axelrod/strategies/titfortat.py

@@ -92,6 +94,39 @@ class TwoTitsForTat(Player):
    def strategy(opponent: Player) -> Action:
        return D if D in opponent.history[-2:] else C

+class DynamicTwoTitsForTat(Player):
+    """A player starts by cooperating and then mimics previous move by 


This strategy does not mimic the previous move by the opponent. It seeks revenge if the opponent has defected in the previous two rounds.

drvinceknight · 2017-05-28T19:15:51Z

axelrod/strategies/titfortat.py

+    opponent with a dynamic bias based off of the opponents ratio of 
+    cooperations to total moves (so their current probability of 
+    cooperating towards cooporating regardless of the move 
+    (aka: forgiveness)."""


Can you add (at the end of this docstring):

Names: - DynamicTwoTitsForTat: Original name by <your name if you wish>

Similar to http://axelrod.readthedocs.io/en/stable/reference/all_strategies.html#axelrod.strategies.ann.EvolvedANN for example.

drvinceknight · 2017-05-28T19:20:05Z

axelrod/strategies/titfortat.py

@@ -1,6 +1,8 @@
 from axelrod.actions import Actions, Action
 from axelrod.player import Player
 from axelrod.strategy_transformers import TrackHistoryTransformer, FinalTransformer
+from axelrod.random_ import random_choice
+import random


Once you use random_choice, remove import random.

drvinceknight · 2017-05-28T21:08:18Z

I'll take a look in the morning. Could either be that you haven't found a particular seed (two different seeds could give the same outcome of course) or an error in the strategy somewhere. Could of course be something else that I've missed.

…

On Sun, 28 May 2017, 22:02 FakeNameSE, ***@***.***> wrote: ***@***.**** commented on this pull request. ------------------------------ In axelrod/strategies/titfortat.py <#1030 (comment)> : > + 'long_run_time': False, + 'inspects_source': False, + 'manipulates_source': False, + 'manipulates_state': False + } + + @staticmethod + def strategy(opponent): + # First move + if len(opponent.history) == 0: + # Make sure we cooporate first turn + return C + if D in opponent.history[-2:]: + # Probability of turning the other cheek based off of + # opponent's probability of defection + if random.random() < (opponent.cooperations / len(opponent.history)): opponent = axelrod.MockPlayer(actions=[D, C, D, D, C]) actions = [(C, D), (C,C), (C,D), (C,D), (C,C)] self.versus_test(opponent, expected_actions=actions, seed=2) still gives me the same thing, regardless of the seed — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#1030 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ACCGWgBVfD8RgHFmMiOluLJi-a4YscZqks5r-eDlgaJpZM4NouBh> .

drvinceknight · 2017-05-29T06:26:42Z

axelrod/strategies/titfortat.py

+            return C
+        if D in opponent.history[-2:]:
+            # Probability of cooperating regardless
+            if random_choice(opponent.cooperations / len(opponent.history)):


random_choice does not return a boolean but returns either a C or a D. This should be:

125 if D in opponent.history[-2:]: 126 # Probability of cooperating regardless 127 return random_choice(opponent.cooperations / len(opponent.history)) 128 return C

I have checked this locally and it all works, you'll need to adjust the tests and the actions. For example the following are passing tests for this strategy:

134 opponent = axelrod.MockPlayer(actions=[D, C, D, D, C]) 135 actions = [(C, D), (D, C), (C, D), (D, D), (D, C)] 136 self.versus_test(opponent, expected_actions=actions, seed=1) 137 138 actions = [(C, D), (D, C), (D, D), (D, D), (C, C)] 139 self.versus_test(opponent, expected_actions=actions, seed=2)

As well as those two I'd just add tests with the Defector and Cooperator (the case when this strategy does not act randomly):

140 141 actions = [(C, C), (C, C), (C, C), (C, C), (C, C)] 142 self.versus_test(axelrod.Cooperator(), expected_actions=actions) 143 144 actions = [(C, D), (D, D), (D, D), (D, D), (D, D)] 145 self.versus_test(axelrod.Defector(), expected_actions=actions)

drvinceknight · 2017-05-29T06:27:20Z

axelrod/strategies/titfortat.py

+    defectiions by opponent with a dynamic bias based off of the 
+    opponents ratio of cooperations to total moves (so their current 
+    probability of cooperating towards cooporating regardless of the 
+    move (aka: forgiveness)).


Blank line before Names please.

Signed-off-by: FakeNameSE <grantlycee@gmail.com>

FakeNameSE · 2017-05-29T21:00:03Z

At last, thank you for your help. Everything is fixed now

drvinceknight · 2017-05-30T08:04:38Z

At last, thank you for your help. Everything is fixed now

Great! Thanks for the work on it. I've given it a 👍, we have a two core reviewer policy so another of the core devs will take another look when they have a moment (they might raise some other points).

Looking forward to getting it in 👍

meatballs · 2017-05-30T08:43:54Z

axelrod/strategies/titfortat.py

+    defectiions by opponent with a dynamic bias based off of the 
+    opponents ratio of cooperations to total moves (so their current 
+    probability of cooperating towards cooporating regardless of the 
+    move (aka: forgiveness)).


typos and grammar:
defectiions -> defections
opponents ratio -> opponent's ratio
based off of the -> based on

"its opponent's defectiions by opponent"

Should this be either 'its opponent's defections' or 'defections by opponent'?

"(so their current probability of cooperating towards cooporating regardless of the move (aka: forgiveness))."

Sorry, but I don't understand what that line means.

meatballs · 2017-05-30T08:50:04Z

axelrod/strategies/titfortat.py

+    @staticmethod
+    def strategy(opponent):
+        # First move
+        if len(opponent.history) == 0:


this would be better as if not opponent.history:

FYI @FakeNameSE this gives some good background: https://stackoverflow.com/questions/53513/best-way-to-check-if-a-list-is-empty

(Apologies, as is was my suggestion, thanks to @meatballs for pointing it out!)

FakeNameSE · 2017-05-30T13:17:31Z

Sorry about the typos. Should be fixed now.

meatballs · 2017-05-30T14:47:36Z

axelrod/strategies/titfortat.py

-    probability of cooperating towards cooporating regardless of the 
-    move (aka: forgiveness)).
+    defections with defections, but with a dynamic bias towards cooperating 
+    based off of the opponent's ratio of cooperations to total moves 


"based off of the..." -> "based on the..."

meatballs · 2017-05-30T14:48:09Z

Looking good. Just the one minor docstring improvement left for me.

thanks!

meatballs · 2017-05-30T16:56:13Z

Many thanks @FakeNameSE !!

I'll send you an invitation to join the project team and you can then have the logo displayed on your github profile by visiting https://github.com/orgs/Axelrod-Python/people and changing your membership from private to public.

We also have a chat room for the project: https://gitter.im/Axelrod-Python/Axelrod

FakeNameSE · 2017-05-30T18:12:24Z

Thank you.

drvinceknight · 2017-05-30T18:14:42Z

Thank you.

Thanks for the contribution! 👍

FakeNameSE changed the title ~~Added a new DynamicTwoTitsForTat strategy which modifies the preexist…~~ Add a new DynamicTwoTitsForTat strategy May 28, 2017

drvinceknight requested changes May 28, 2017

View reviewed changes

Some attempts to fix.

4c89284

drvinceknight requested changes May 29, 2017

View reviewed changes

FakeNameSE added 2 commits May 29, 2017 15:19

Add blank line before names

f07bcb7

Fixed tests and DynamicsTwoTitsForTat strategy

e9abbe8

Signed-off-by: FakeNameSE <grantlycee@gmail.com>

drvinceknight approved these changes May 30, 2017

View reviewed changes

drvinceknight added the ready-to-merge label May 30, 2017

meatballs requested changes May 30, 2017

View reviewed changes

meatballs added ready-for-review and removed ready-to-merge labels May 30, 2017

Fix typos and make empty list check more pythonic

d014e39

meatballs reviewed May 30, 2017

View reviewed changes

Small docstring fix

1971111

meatballs approved these changes May 30, 2017

View reviewed changes

meatballs merged commit 9edd676 into Axelrod-Python:master May 30, 2017

Add a new DynamicTwoTitsForTat strategy #1030

Add a new DynamicTwoTitsForTat strategy #1030

Conversation

FakeNameSE commented May 28, 2017

FakeNameSE commented May 28, 2017

drvinceknight left a comment • edited

Choose a reason for hiding this comment

Choose a reason for hiding this comment

drvinceknight May 28, 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

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

drvinceknight commented May 28, 2017 via email

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

FakeNameSE commented May 29, 2017 • edited

drvinceknight commented May 30, 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

FakeNameSE commented May 30, 2017

Choose a reason for hiding this comment

meatballs commented May 30, 2017

meatballs commented May 30, 2017

FakeNameSE commented May 30, 2017

drvinceknight commented May 30, 2017

drvinceknight left a comment •

edited

drvinceknight May 28, 2017 •

edited

FakeNameSE commented May 29, 2017 •

edited