Hawk/dove updating risk attitudes #36
Conversation
Codecov Report
@@ Coverage Diff @@
## main #36 +/- ##
==========================================
+ Coverage 48.58% 53.64% +5.05%
==========================================
Files 15 16 +1
Lines 566 645 +79
==========================================
+ Hits 275 346 +71
- Misses 291 299 +8 |
rlskoeser
force-pushed
the
hawkdove-update-risk
branch
from
e50342b to
0d5980c
Compare
simulatingrisk/charts/histogram.py
Outdated
| # You have to specify the dependencies as follows, so that the figure | ||
| # auto-updates when viz.model or viz.df is changed. |
There was a problem hiding this comment.
ooh, yes, thanks for catching
simulatingrisk/hawkdovevar/app.py
Outdated
| jupyterviz_params_var = jupyterviz_params.copy() | ||
| # remove parameter for agent risk level; |
There was a problem hiding this comment.
Why is this removed? not clear from immediate context but perhaps clear elsewhere
There was a problem hiding this comment.
good point, I should clarify that in the comment at least - it's a parameter that's only relevant for the single-risk level simulation because in the variable simulation risk attitudes are generated randomly; I don't currently have base/common parameters (common params would probably be clearer)
| @@ -13,18 +17,94 @@ def set_risk_level(self): | |||
| # generate a random risk level | |||
| self.risk_level = self.random.randint(0, num_neighbors) | |||
There was a problem hiding this comment.
self.random presumably comes from the base mesa class? Otherwise wondering what it refers to
There was a problem hiding this comment.
yes, mesa provides access to built-in random methods that works just like the main python library random module
It's in mentioned in their documentation best practices, it would allow using a random seed at model instantiation: https://mesa.readthedocs.io/en/stable/best-practices.html#randomization
| def adjust_risk(self): | ||
| # look at neighbors | ||
| # if anyone has more points | ||
| # either adopt their risk attitude or average theirs with yours | ||
|
|
||
| best = self.most_successful_neighbor | ||
| # if most successful neighbor has more points and a different | ||
| # risk attitude, adjust | ||
| if best.points > self.points and best.risk_level != self.risk_level: | ||
| # adjust risk based on model configuration | ||
| if self.model.risk_adjustment == "adopt": | ||
| # adopt neighbor's risk level | ||
| self.risk_level = best.risk_level | ||
| elif self.model.risk_adjustment == "average": | ||
| # average theirs with mine, then round to a whole number | ||
| # since this model uses discrete risk levels | ||
| self.risk_level = round( | ||
| statistics.mean([self.risk_level, best.risk_level]) | ||
| ) |
There was a problem hiding this comment.
Very, very cool, and slick. I remember talking about this and/or similar strategies of risk adjustment with David and Lara and you.
| assert ( | ||
| repr(agent) | ||
| == f"<HawkDoveSingleRiskAgent id={agent_id} r={risk_level} points=0>" | ||
| ) |
| assert model.risk_adjustment == "adopt" | ||
| assert model.adjust_round_n == 5 | ||
| assert model.hawk_odds == 0.2 | ||
| assert model.include_diagonals is False | ||
|
|
||
| # handle string none for solara app parameters | ||
| model = HawkDoveVariableRiskModel(5, risk_adjustment="none") | ||
| assert model.risk_adjustment is None | ||
|
|
||
| # complain about invalid adjustment type | ||
| with pytest.raises(ValueError, match="Unsupported risk adjustment 'bogus'"): | ||
| HawkDoveVariableRiskModel(3, risk_adjustment="bogus") |
There was a problem hiding this comment.
These easy access object property-methods really make using and testing quite easy and legible.
| ] | ||
|
|
||
|
|
||
| @pytest.mark.parametrize("params,expect_adjust_step", adjustment_testdata) |
There was a problem hiding this comment.
This is nice, so it runs the test 4 times according to the number of tuples in the adjustment_testdata list?
There was a problem hiding this comment.
yeah, isn't that cool? very useful sometimes
| neighbor = Mock(points=1500, risk_level=3) | ||
| with patch.object(HawkDoveVariableRiskAgent, "most_successful_neighbor", neighbor): |
* Preliminary work for hawk/dove adaptive risk attitudes #20 * Clean up duplicate code; add comments about adjusting parameters * Add tests for variable hawk/dove; improve error handling * Add a label for round-adjustment parameter * Fix typo in app about placeholder content * Load about app content relative to app file * Remove model dependency in matplotlib histogram (prevents from updating) * Remove unused requirements text files/directories * Clean up outdated comments and refactor common hawk/dove params
modify variable-risk hawk/dove game to add logic for updating risk levels based on neighbors
adjustment logic is adapted from previous implementation in the risk bet simulation