CooperateISO - #1498
Open
adrianhutter wants to merge 31 commits into
Open
Conversation
adrianhutter
marked this pull request as ready for review
August 5, 2026 12:15
drvinceknight
requested changes
Aug 7, 2026
drvinceknight
left a comment
drvinceknight
left a comment
Member
There was a problem hiding this comment.
This looks really great, I enjoyed the preprint (congrats on the work)!
My requests are mainly stylistic as we tend to try and avoid inline comments (there are a number in the tests that I didn't comment on that could perhaps be moved to docstrings).
| @@ -1,4 +1,7 @@ | |||
| sphinx>=7.0.0,<9.0.0 | |||
Member
There was a problem hiding this comment.
Can you help me understand why this change is needed? (This might be because of an upstream change.)
|
|
||
| # The theme to use for HTML and HTML Help pages. See the documentation for | ||
| # a list of builtin themes. | ||
| on_rtd = os.environ.get("READTHEDOCS", None) == "True" |
Member
There was a problem hiding this comment.
Could you help me understand why this change? (It might be something got stale that I'm not remembering.)
| return opponent.history[-1] | ||
|
|
||
|
|
||
| # We describe memory-1 strategies as length-4 arrays, quantifying the probability of cooperation in the states [CC, CD, DC, DD]. |
Member
There was a problem hiding this comment.
Suggested change
| # We describe memory-1 strategies as length-4 arrays, quantifying the probability of cooperation in the states [CC, CD, DC, DD]. |
Comment on lines
+87
to
+114
| # Apply p_noise only to own strategy, not to opponent | ||
| # (the opponent strategy already includes noise effects). | ||
| own = my_strategy + p_noise * (1.0 - 2.0 * my_strategy) | ||
|
|
||
| # Flip CD/DC for opponent. | ||
| opp = opp_strategy[[0, 2, 1, 3]] | ||
|
|
||
| # Build the transition matrix. | ||
| trans_mat = np.array( | ||
| [ | ||
| own * opp, | ||
| own * (1.0 - opp), | ||
| (1.0 - own) * opp, | ||
| (1.0 - own) * (1.0 - opp), | ||
| ] | ||
| ).T | ||
|
|
||
| R, P, S, T = RPST | ||
| rewards = np.array([R, S, T, P], dtype=float) | ||
|
|
||
| # Don't include init state in summed rewards. | ||
| inv = np.linalg.inv(np.eye(4) - (1.0 - p_end) * trans_mat) | ||
|
|
||
| # Calculate expected reward, | ||
| reward = init_state @ (inv @ rewards - rewards) | ||
|
|
||
| # Avg. reward per step | ||
| return p_end * float(reward) / (1.0 - p_end) |
Member
There was a problem hiding this comment.
Could you remove this inline comments and put the corresponding explanation in the docstring please.
Comment on lines
+130
to
+148
| # Clamp to possible values, given noise | ||
| opp = np.clip(opponent, p_noise, 1.0 - p_noise) | ||
|
|
||
| # Setup initial state | ||
| init_state = np.zeros(4, dtype=np.float32) | ||
| init_state[init_state_idx] = 1.0 | ||
|
|
||
| # Define the objective function to minimize (negative reward) | ||
| def objective(params: np.ndarray) -> float: | ||
| return -get_reward(params, opp, init_state, p_end, p_noise, RPST) | ||
|
|
||
| x0 = np.array([0.5, 0.5, 0.5, 0.5]) | ||
| bounds = [(0.0, 1.0), (0.0, 1.0), (0.0, 1.0), (0.0, 1.0)] | ||
| result = minimize( | ||
| objective, x0, method="L-BFGS-B", bounds=bounds, options={"maxiter": 50} | ||
| ) | ||
|
|
||
| # result.fun is the minimum loss (-reward), result.x are the optimal parameters | ||
| return -result.fun, result.x |
Member
There was a problem hiding this comment.
I think we can remove the inline comments here
Comment on lines
+183
to
+193
| # Track the opponent's rate of cooperation (numerator, denominator) for each state. | ||
| # Assume we have seen the opponent play following TfT once in each state, | ||
| # to make the opponent-model well-defined from the start. | ||
| self.ewma_CC = [1.0, 1.0] | ||
| self.ewma_CD = [1.0, 1.0] | ||
| self.ewma_DC = [0.0, 1.0] | ||
| self.ewma_DD = [0.0, 1.0] | ||
|
|
||
| # Initial cooperation probabilities (num / den) | ||
| self.opp_model = [1.0, 0.0, 1.0, 0.0] | ||
| self.my_policy = [1.0, 0.0, 1.0, 0.0] |
Member
There was a problem hiding this comment.
Can you move these inline comments to the docstring of the __init__
Comment on lines
+300
to
+301
| # Estimate of the opponent's rate of playing D after C, taking noise | ||
| # into account. |
Member
There was a problem hiding this comment.
Suggested change
| # Estimate of the opponent's rate of playing D after C, taking noise | |
| # into account. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Adding 3 strategies (LongtermTfT, ISO, CooperateISO) from https://arxiv.org/abs/2303.03519.
Made a few changes relative to the versions used in the paper. Most importantly, ISO now uses
scipy.optimizeinstead of torch's Adam. This works equally well, is faster, and avoids having to import torch.Ran the CooperateISO implementation from this PR against all strategies in Axelrod-4.14.0, and verified that it outperforms EvolvedLookerUp2_2_2 and DBS at both noise levels 0% and 10%.
Fix Read the Docs build failures by updating documentation requirements and conf.py: