Add state to action analysis. #870
Conversation
Add two functions to interaction_utils: - `interaction_utils.compute_state_to_action_distribution`: computes list of counter objects of mapping a `(state, action)` to a count. - `interaction_utils.compute_normalised_state_to_action_distribution`: computes list of counter objects of mapping a `(state, action)` to a normalised count (for a given state). There have then been incorporated in `axelrod.result_set` and in particular in the `result_set.summarise()` output. Which now includes the average rates for each strategy in a tournament. So for memory 1 strategies this means we can measure their parameters and for non memory 1 strategies we can infer what the parameters correspond to the data.
axelrod/interaction_utils.py
Outdated
| def compute_state_to_action_distribution(interactions): | ||
| """ | ||
| Returns the count of each state to action | ||
| for a set of interactions. |
There was a problem hiding this comment.
I think this needs a little more explanation as to what the function does. It wasn't clear to me until I looked at how it's being used in the result set. Perhaps an example would be useful?
There was a problem hiding this comment.
Yeah good shout. I'll clear that up and find the others. I think I got lazy :)
axelrod/interaction_utils.py
Outdated
| """ | ||
| Returns the normalised count of each state to action | ||
| for a set of interactions. | ||
|
|
There was a problem hiding this comment.
Same here. A little more explanation of what this is doing might be in order.
|
Let me know if you think it needs more than those last two commits :) Just jumping in the plane, will pick this up later :D |
axelrod/interaction_utils.py
Outdated
| Implying that from a state of (C, D) (the first player having played C and | ||
| the second playing D) the player in question then played C. | ||
|
|
||
| The following counter object, implies that the player in question was in |
There was a problem hiding this comment.
Pedantic: "The following Counter object implies that"
axelrod/interaction_utils.py
Outdated
| return None | ||
|
|
||
| distributions = [Counter([(state, outcome[j]) for state, outcome in zip(interactions, | ||
| interactions[1:])]) |
There was a problem hiding this comment.
Can we make the spacing here more aesthetically pleasing?
axelrod/interaction_utils.py
Outdated
|
|
||
| ((C, D), C) | ||
|
|
||
| Implying that from a state of (C, D) (the first player having played C and |
There was a problem hiding this comment.
Pedantic: lowercase "impying"
axelrod/interaction_utils.py
Outdated
| Implying that from a state of (C, D) (the first player having played C and | ||
| the second playing D) the player in question then played C. | ||
|
|
||
| The following counter object, implies that the player in question was only |
axelrod/interaction_utils.py
Outdated
| ---------- | ||
| normalised_state_to_C_distributions : List of Counter Object | ||
| List of Counter objects where the keys are the states and actions and | ||
| the values the normalized counts.. The |
axelrod/interaction_utils.py
Outdated
| normalized_distribution = [] | ||
| for player in range(2): | ||
| counter = {} | ||
| for state in [('C', 'C'), ('C', 'D'), ('D', 'C'), ('D', 'D')]: |
There was a problem hiding this comment.
C, D as in Actions.C, Actions.D rather than 'C' and 'D'
axelrod/result_set.py
Outdated
| """ | ||
| Returns | ||
| ---------- | ||
|
|
| norm_counter = Counter() | ||
| for state in [(C, C), (C, D), (D, C), (D, D)]: | ||
| total = counter[(state, C)] + counter[(state, D)] | ||
| if total > 0: |
There was a problem hiding this comment.
if total == 0 then we add nothing to the list? just checking that is the intention rather than None or an empty dict.
There was a problem hiding this comment.
If total == 0 we still add the empty norm_counter so the behaviour is consistant :)
The test cases catch this occurrence :)
| @@ -22,11 +22,27 @@ class TestMatch(unittest.TestCase): | |||
| Counter({('D', 'C'): 2}), | |||
| Counter({('C', 'C'): 1, ('C', 'D'): 1}), | |||
| None] | |||
| state_to_action_distribution = [[Counter({(('C', 'D'), 'D'): 1}), | |||
There was a problem hiding this comment.
Can we use C and D instead of 'C' and 'D'?
| self.state_to_action_distribution): | ||
| self.assertEqual(dist, | ||
| iu.compute_state_to_action_distribution(inter)) | ||
| inter = [(C, D), (D, C), (C, D), (D, C), (D, D), (C, C), (C, D)] |
|
Looks fine overall, I added some minor comments about docstrings and |
- 'C'/'D' -> C/D - Adding type docstrings.
I think I've caught all of them and answered your question above, happy to further improve it though if you think of anything else :) 👍 |
Add two functions to interaction_utils:
interaction_utils.compute_state_to_action_distribution: computeslist of counter objects of mapping a
(state, action)to a count.interaction_utils.compute_normalised_state_to_action_distribution: computeslist of counter objects of mapping a
(state, action)to a normalisedcount (for a given state).
There have then been incorporated in
axelrod.result_setand in particular inthe
result_set.summarise()output. Which now includes the averagerates for each strategy in a tournament. So for memory 1 strategies this
means we can measure their parameters and for non memory 1 strategies we
can infer what the parameters correspond to the data.