Add new strategy ShortMem #857

janga1997 · 2017-02-08T18:14:05Z

#379
From the suggested list of strategies by @marcharper from this paper.

I'm sure there a few things I missed. @drvinceknight Do take a look whenever free.
I will work on the tests if this is okay so far.

drvinceknight · 2017-02-08T18:21:12Z

axelrod/strategies/shortmem.py

+C, D = Actions.C, Actions.D
+
+class ShortMem(Player):
+    """A player who only ever defects."""


I don't think this docstring is accurate right?

Also note that we require a few more details in the docstring, you can see an example here: http://axelrod.readthedocs.io/en/latest/tutorials/contributing/strategy/writing_the_new_strategy.html

drvinceknight · 2017-02-08T18:23:48Z

axelrod/strategies/shortmem.py

+    def strategy(opponent: Player) -> Action:
+
+        memoryDepth = self.classifier['memoryDepth']
+        cooperateRatio = array.count('C')/memoryDepth


You haven't defined array so I expect this strategy would fail to play.

Note that we're going to need some tests for the strategy too, you can see some information on the tests here: http://axelrod.readthedocs.io/en/latest/tutorials/contributing/strategy/writing_test_for_the_new_strategy.html

We're happy to give a hand with tests if you'd like one. Essentially we want code that checks that all the logic you write below works :)

Also, note that a Player has a cooperations attribute that counts the number of cooperations.

I thought about that, but we need the number of cooperations and defections in the last 10 steps only, and to the extent that I understand it, it records the Cs and Ds for the whole play.

drvinceknight · 2017-02-08T18:24:34Z

axelrod/strategies/shortmem.py

+        elif defectRatio - cooperateRatio > 0.3:
+            return D
+        else:
+            return TitForTat().strategy(opponent)


Could you code this explicity please. Something like:

else: return opponent.history[-1]

janga1997 · 2017-02-08T19:00:05Z

@drvinceknight I've been meaning to get a little feedback. In this case, is FIFO really relevant to the what this strategy does, because I don't think it is.

drvinceknight · 2017-02-08T19:01:02Z

I've been meaning to get a little feedback. In this case, is FIFO really relevant to the what this strategy does, because I don't think it is.

Sorry, what do you mean by FIFO?

janga1997 · 2017-02-08T19:03:36Z

It's an acronym for first in, first out, where the oldest (first) entry, or 'head' of the queue, is processed first. This term was specifically mentioned in the description of this strategy in the paper.

drvinceknight · 2017-02-08T19:04:10Z

It's an acronym for first in, first out, where the oldest (first) entry, or 'head' of the queue, is processed first. This term was specifically mentioned in the description of this strategy in the paper.

I think it's ok to simply say "the last ten moves".

drvinceknight

The main thing that is missing are tests :) Let us know if we can assist with them.

drvinceknight · 2017-02-08T19:06:53Z

axelrod/strategies/shortmem.py

+    def strategy(opponent: Player) -> Action:
+
+        memoryDepth = self.classifier['memoryDepth']
+        array = self.history[:-11:-1]


As order doesn't matter you can just take: self.history[-10:].

>>> a = list(range(20)) >>> a[-10:] [10,11,12,13,14,15,16,17,18,19]

However note that I think you mean for this to be opponent.history.

drvinceknight · 2017-02-08T19:07:44Z

axelrod/strategies/shortmem.py

+        cooperateRatio = array.count('C')/memoryDepth
+        defectRatio = array.count('D')/memoryDepth
+
+        if len(self.history) <= memoryDepth:


You could move this to before the counting (a tiny efficiency improvement).

drvinceknight · 2017-02-08T19:08:07Z

axelrod/strategies/shortmem.py

+    """
+    A player starts by always cooperating for the first 10 moves.
+
+    The opponent answers are stored in the memory. The memory is FIFO and the maximum size of the memory is 10 results. From the tenth round on, the program analyzes the memory, and compare the number of defects and cooperates of the opponent, based in percentage. If cooperation occurs 30% more than defection, it will cooperate.


Could you stick to PEP8 here, so keep the characters to less than 80 per line please :)

drvinceknight · 2017-02-08T19:10:15Z

axelrod/strategies/shortmem.py

+
+    The opponent answers are stored in the memory, whose maximum size is 10 results. From the tenth round on, the program analyzes the memory, and compare the number of defects and cooperates of the opponent, based in percentage. If cooperation occurs 30% more than defection, it will cooperate.
+    If defection occurs 30% more than cooperation, the program will defect. Otherwise, the program follows the TitForTat algorithm.
+


Can you add something like:

Names - Short Mem: [Axelrod1980]_ """

Except replace Axelrod1980 by a key that you'd put in https://github.com/Axelrod-Python/Axelrod/blob/master/docs/reference/bibliography.rst

janga1997 · 2017-02-08T19:49:19Z

@drvinceknight I hope its okay now?

Nikoleta-v3 · 2017-02-08T20:10:19Z

@janga1997 Hello I am also a contributor to the library! Did you remember to add the reference like suggested.

Except replace Axelrod1980 by a key that you'd put in https://github.com/Axelrod-Python/Axelrod/blob/master/docs/reference/bibliography.rst

janga1997 · 2017-02-08T20:15:38Z

@Nikoleta-v3 no I didn't yet. I will do it tomorrow along with the tests, as it late night here. Thanks .

janga1997 · 2017-02-09T10:21:45Z

@drvinceknight @Nikoleta-v3 Do review whenever you're free. I believe I've covered the necessary test cases.

Nikoleta-v3 · 2017-02-09T12:39:20Z

docs/reference/bibliography.rst

@@ -37,3 +37,4 @@ documentation.
 .. [Stewart2012] Stewart, a. J., & Plotkin, J. B. (2012). Extortion and cooperation in the Prisoner’s Dilemma. Proceedings of the National Academy of Sciences, 109(26), 10134–10135. http://doi.org/10.1073/pnas.1208087109
 .. [Szabó1992] Szabó, G., & Fáth, G. (2007). Evolutionary games on graphs. Physics Reports, 446(4-6), 97–216. http://doi.org/10.1016/j.physrep.2007.04.004
 .. [Tzafestas2000] Tzafestas, E. (2000). Toward adaptive cooperative behavior. From Animals to Animals: Proceedings of the 6th International Conference on the Simulation of Adaptive Behavior {(SAB-2000)}, 2, 334–340.
+.. [Andre2013] Andre L. C., Honovan P., Felipe T. and Frederico G. (2013). Iterated Prisoner’s Dilemma - An extended analysis, http://abricom.org.br/wp-content/uploads/2016/03/bricsccicbic2013_submission_202.pdf


I am not sure if the reference is generated correctly because there is an extra : in shortmem.py line 15. Could you please check. 😄

ShortMem: [Andre:2013]_

janga1997 · 2017-02-09T13:05:43Z

@Nikoleta-v3 Is everything fine with the tests and the strategy itself?

Nikoleta-v3 · 2017-02-09T13:11:35Z

Everything else looks good to me! Happy to be working with you 😄
The core contributors need to give a 👍 now.

drvinceknight

One tiny tiny thing left from me. Thanks @Nikoleta-v3 and @janga1997 this is looking good!

drvinceknight · 2017-02-09T13:21:32Z

axelrod/strategies/shortmem.py

+    """
+    A player starts by always cooperating for the first 10 moves.
+
+    The opponent answers are stored in the memory, whose maximum size is 10 results. From the tenth round on, the program analyzes the memory, and compare the number of defects and cooperates of the opponent, based in percentage. If cooperation occurs 30% more than defection, it will cooperate.


This all looks good to me, could you keep the number of characters on each line to less than 80 please? So it just needs this docstring to line break a bit earlier.

drvinceknight · 2017-02-09T13:31:29Z

How exactly do you make a word wrap at the the 80th column.

It would depend on your editor. There are ways to set it up so it does it automatically but you can also create a visual queue that just shows where the 80 character limit is and then you can line break manually.

marcharper

Thanks for the contribution! I think there's one error and some style issues, and otherwise almost ready.

marcharper · 2017-02-09T15:18:06Z

axelrod/strategies/shortmem.py

+        if len(opponent.history) <= 10:
+            return C
+
+        array = opponent.history[:-11:-1]


Take a look at this line, the arrays are off by one.

>>> l = list(range(100)) >>> l[:-11:-1] [99, 98, 97, 96, 95, 94, 93, 92, 91, 90] >>> l[-11:] [89, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99]

Correct me if I am wrong, since we are talking about the last ten rounds here, doesn't l[-11:] produce a slice with 11 elements?

I think you can just use:

>>> a = range(20) >>> list(a[-10:]) [10, 11, 12, 13, 14, 15, 16, 17, 18, 19] >>> len(list(a[-10:])) 10

If you go for a[-10:] I think that improves the readability/ambiguity and good call on the having a test that checks this.

I understand about the readability/ambiguity issue, but what exactly is a test that checks this?

Is it self.responses_test([C], [C] * 4 + [D] *6, [D] * 4 + [C] * 6) ?

Sorry, @drvinceknight's suggestion is better. A test to check would fail on 9 trailing elements but not 10.

Sorry for being difficult, but I really don't see the reason for an additional test.

The strategy is to return C from 1-9 moves, and from the 10th move, the rest of the logic kicks in.
I don't understand what you and @drvinceknight mean by

A test to check would fail on 9 trailing elements but not 10

Why would the test fail on 9 trailing elements? It would return C as the strategy describes.

Sorry for being difficult, but I really don't see the reason for an additional test.

Don't worry you're not being difficult, it's good to talk these things over :)

I would definitely suggest you change:

opponent.history[:-11:-1]

to be

opponent.history[-10:]

This just helps with readability.

In terms of the test, I suggest you just add the following one:

self.responses_test([C], [D] *9, [D] * 9) # This checks that when there are 9 previous plays you cooperate self.responses_test([D], [D] *10, [D] * 10) # This checks that when there are 10 previous plays the logic kicks in

marcharper · 2017-02-09T15:19:12Z

axelrod/strategies/shortmem.py

+    """
+    A player starts by always cooperating for the first 10 moves.
+
+    The opponent answers are stored in the memory, whose maximum size is 10


Technically all the history is saved in memory in our library. I would just say that the player analyzes the last ten rounds of play without any details about what's in memory.

answers -> actions

marcharper · 2017-02-09T15:20:22Z

axelrod/strategies/shortmem.py

+            return C
+
+        array = opponent.history[:-11:-1]
+        cooperateRatio = array.count('C')


These aren't ratios, cooperation_counts and defection_counts would be better names, or simply C_counts.

marcharper · 2017-02-09T15:20:51Z

axelrod/strategies/shortmem.py

+    compare the number of defects and cooperates of the opponent, based in
+    percentage. If cooperation occurs 30% more than defection, it will
+    cooperate.
+    If defection occurs 30% more than cooperation, the program will defect.


program --> player

marcharper · 2017-02-09T15:22:19Z

axelrod/tests/unit/test_shortmem.py

@@ -0,0 +1,34 @@
+"""Tests for the Cooperator strategy."""


Cooperator -> shortmem

Please add a test to catch the array length issue above (array is of length 9 instead of 10 but none of the tests were tripped).

I replied about the array length issue above.

marcharper · 2017-02-09T15:22:38Z

axelrod/tests/unit/test_shortmem.py

+        self.responses_test([C], [C] * 4, [D] * 4)
+        self.responses_test([C], [C] * 4 + [D] *6, [D] * 4 + [C] * 6)
+
+        #Cooperate if in the last ten moves, Cooperations - Defections >= 30


30% and see below

I didn't understand what you mean.

We're not calculating percentages here. I get that it's pointless to divide by 10 on both sides but it's not an accurate description, and it might be confusing to look at this code in six months. So I'm asking that the comments elaborate more about what the code's intention.

marcharper · 2017-02-09T15:23:32Z

axelrod/strategies/shortmem.py

+
+    The opponent answers are stored in the memory, whose maximum size is 10
+    results. From the tenth round on, the program analyzes the memory, and
+    compare the number of defects and cooperates of the opponent, based in


defections and cooperations

marcharper · 2017-02-09T15:24:46Z

axelrod/strategies/shortmem.py

+        cooperateRatio = array.count('C')
+        defectRatio = array.count('D')
+
+        if cooperateRatio - defectRatio >= 3:


FYI standard Python style is camel_case instead of CamelCase for variables, search for PEP8 (the style guide).

drvinceknight · 2017-02-10T09:33:04Z

axelrod/strategies/shortmem.py

+        if len(opponent.history) <= 10:
+            return C
+
+        array = opponent.history[:-11:-1]


Sorry for being difficult, but I really don't see the reason for an additional test.

Don't worry you're not being difficult, it's good to talk these things over :)

I would definitely suggest you change:

opponent.history[:-11:-1]

to be

opponent.history[-10:]

This just helps with readability.

In terms of the test, I suggest you just add the following one:

self.responses_test([C], [D] *9, [D] * 9) # This checks that when there are 9 previous plays you cooperate self.responses_test([D], [D] *10, [D] * 10) # This checks that when there are 10 previous plays the logic kicks in

drvinceknight · 2017-02-10T09:35:07Z

axelrod/strategies/shortmem.py

+        C_counts = array.count('C')
+        D_counts = array.count('D')
+
+        if C_counts - D_counts >= 3:


Another question (not sure how we missed this), is the logic correct here?

You're looking at the difference in counts (subtraction) but the description is saying that there should be 30% more cooperations than defections? So I'm not sure the logic is actually correct.

I think this test should be:

if C_counts / D_counts >= 1.3:

That checks that there are 30% more cooperations.

I am starting to doubt it a bit myself.
But doesn't the term more than imply subtraction?

After thinking about it, it seems your logic makes more sense w.r.t. the given statement.

When I first thought about it, I made the assumption that percentages are to be calculated w.r.t the sum of C and D (10). My mistake. I would have never found it myself.

But doesn't the term more than imply subtraction?

My suggestion is incorrect, because we're only considering 10 turns this is actually correct. Apologies.

@janga1997 I actually think I'm wrong. I think you logic was correct before.

…to shortMem

drvinceknight · 2017-02-10T16:56:39Z

axelrod/strategies/shortmem.py

+        if D_counts == 0:
+            return C
+
+        if C_counts / (D_counts * 1.0) >= 1.3:


I think the logic was actually correct before (my mistake) I suggest you revert: a7b8d9f

This reverts commit a7b8d9f.

janga1997 · 2017-02-10T17:19:20Z

@drvinceknight Is that everything?

drvinceknight · 2017-02-10T17:22:32Z

@drvinceknight Is that everything?

This looks good to me. I'll leave @marcharper to merge it :) 👍

This was missed in #857.

Add new strategy ShortMem

05631e4

drvinceknight requested changes Feb 8, 2017

View reviewed changes

Fix errors

129dcc2

Make changes to docstring

104e9af

drvinceknight requested changes Feb 8, 2017

View reviewed changes

janga1997 added 2 commits February 9, 2017 00:40

Edit for efficiency

91ae444

Add credit to author

7130692

janga1997 added 4 commits February 9, 2017 15:43

Make changes to ShortMem

de373a9

Include ShortMem in _strategies.py

807a881

Add tests for ShortMem

b1f93b8

Add paper to bibliography

92b00be

Nikoleta-v3 reviewed Feb 9, 2017

View reviewed changes

Fix error in docstring

85c0986

drvinceknight requested changes Feb 9, 2017

View reviewed changes

Word wrap to 80 columns

51ca1e2

drvinceknight approved these changes Feb 9, 2017

View reviewed changes

drvinceknight added the ready-to-merge label Feb 9, 2017

marcharper requested changes Feb 9, 2017

View reviewed changes

Fix style issues

d5a561b

drvinceknight mentioned this pull request Feb 10, 2017

Add type hints to handshake, inverse, mathematicalconstants #858

Merged

drvinceknight requested changes Feb 10, 2017

View reviewed changes

drvinceknight removed the ready-to-merge label Feb 10, 2017

Merge branch 'master' of https://github.com/Axelrod-Python/Axelrod in…

dfdb808

…to shortMem

janga1997 force-pushed the shortMem branch from 9a36186 to dfdb808 Compare February 10, 2017 11:48

Fix logical error in strategy and test

a7b8d9f

drvinceknight requested changes Feb 10, 2017

View reviewed changes

janga1997 added 2 commits February 10, 2017 22:43

Revert "Fix logical error in strategy and test"

cd8668f

This reverts commit a7b8d9f.

Fix Readability

014744e

drvinceknight approved these changes Feb 10, 2017

View reviewed changes

drvinceknight added the ready-to-merge label Feb 10, 2017

marcharper approved these changes Feb 11, 2017

View reviewed changes

marcharper merged commit b174daa into Axelrod-Python:master Feb 11, 2017

drvinceknight added a commit that referenced this pull request Feb 11, 2017

Add shortmem to autodoc

abd5811

This was missed in #857.

drvinceknight mentioned this pull request Feb 11, 2017

Add shortmem to autodoc #861

Merged

marcharper pushed a commit that referenced this pull request Feb 13, 2017

Add shortmem to autodoc

6413ccb

This was missed in #857.


		The opponent answers are stored in the memory, whose maximum size is 10 results. From the tenth round on, the program analyzes the memory, and compare the number of defects and cooperates of the opponent, based in percentage. If cooperation occurs 30% more than defection, it will cooperate.
		If defection occurs 30% more than cooperation, the program will defect. Otherwise, the program follows the TitForTat algorithm.

Add new strategy ShortMem #857

Add new strategy ShortMem #857

Conversation

janga1997 commented Feb 8, 2017 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

janga1997 Feb 8, 2017 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

janga1997 commented Feb 8, 2017

drvinceknight commented Feb 8, 2017

janga1997 commented Feb 8, 2017

drvinceknight commented Feb 8, 2017

drvinceknight left a 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

janga1997 commented Feb 8, 2017

Nikoleta-v3 commented Feb 8, 2017

janga1997 commented Feb 8, 2017

janga1997 commented Feb 9, 2017

Choose a reason for hiding this comment

janga1997 commented Feb 9, 2017

Nikoleta-v3 commented Feb 9, 2017

drvinceknight left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

drvinceknight commented Feb 9, 2017

marcharper left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

drvinceknight Feb 9, 2017 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

marcharper Feb 10, 2017 • edited Loading

Choose a reason for hiding this comment

janga1997 Feb 10, 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

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

janga1997 commented Feb 10, 2017

drvinceknight commented Feb 10, 2017

janga1997 commented Feb 8, 2017 •

edited

Loading

janga1997 Feb 8, 2017 •

edited

Loading

drvinceknight Feb 9, 2017 •

edited

Loading

marcharper Feb 10, 2017 •

edited

Loading

janga1997 Feb 10, 2017 •

edited

Loading