Terminology for agent versus "standard" values in extensive games by tturocy · Pull Request #642 · gambitproject/gambit

tturocy · 2025-11-24T12:42:23Z

This contains work for solving #617 .

Checklist of to-dos:

In Python and C++ the existing GetLiapValue/liap_value and GetMaxRegret/max_regret are prefixed by Agent/agent.
Existing tests and examples are updated to use agent terminology (but see action below!)
New functions enumpure_agent_solve and liap_agent_solve implemented and documented (and cross-linked with their non-agent versions)
GUI uses only the non-agent forms (for the avoidance of confusion to less experienced users; explicitly working with the agent form is an advanced topic suitable for CLI/pygambit use)
gambit-liap updated to follow convention of gambit-enumpure - default to using strategic form, use agent only if prompted explicitly to do so.
Implement liap_value and max_regret (and player regret) for MixedBehaviorProfile. These should be identical to their strategic equivalent.
Review all tests to see whether the agent or "regular" version should be used
Review all tutorials/user guide to see whether the agent or "regular" version should be used as examples (most likely the "regular" version is appropriate for general introductory tutorials)
Add tutorial/example covering Myerson's figure 4.2 example to illustrate the difference between standard Nash and agent Nash.

…ax_regret` -> `agent_max_regret` in Python

…` for expositional clarity.

…ure`

rahulsavani

Quite a lot to check here, but I believe everything is fine. Two minor things, the second of which is more general than this PR:

enumpure_solve now has no use_strategic flag, but the NashComputationResult stills shows it as follows: NashComputationResult(method='enumpure', rational=True, use_strategic=True; similarly for enumpure_agent_solve where use_strategic is then False in the NashComputationResult. This might be fine as it indicates that type of the profiles under equilibria, but then perhaps that should be indicated another way rather than via this defunct argument?
test_tutorials.py initially failed for me. On further inspection that was because it used a notebook checkpoint. I am not sure we want that behavior if it can be customized. If you agree we could ask Ed about looking into this.

tturocy · 2025-11-25T16:36:33Z

use_strategic=True is set in the output object for all methods that use the strategic form, even those that do not have sequence form (or agent form) equivalents.

review-notebook-app · 2025-12-01T09:11:38Z

Check out this pull request on

See visual diffs & provide feedback on Jupyter Notebooks.

Powered by ReviewNB

# Conflicts: # doc/tutorials/03_stripped_down_poker.ipynb # src/games/stratmixed.h # src/pygambit/nash.pxi # src/tools/liap/liap.cc # tests/test_behav.py # tests/test_nash.py

…profiles

tturocy · 2025-12-08T22:18:54Z

@rahulsavani We now have the new, Nash-correct max_regret and liap_value implemented for mixed behaviour profiles. Or, at least, it should be correct - so you can now add your tests to this branch.

rahulsavani · 2025-12-11T13:37:04Z

@rahulsavani We now have the new, Nash-correct max_regret and liap_value implemented for mixed behaviour profiles. Or, at least, it should be correct - so you can now add your tests to this branch.

It seems liap_value still comes back erroneously as 0 for the agent-but-not-full-Nash profile for Myserson's fig 4.2:

In [1]: import pygambit as gbt
   ...: g = gbt.read_efg('tests/test_games/myerson_fig_4_2.efg')
   ...: p = g.mixed_behavior_profile([[[0, 1], [0, 1]], [[1, 0]]])
   ...: print(p.max_regret())
   ...: print(p.agent_max_regret())
   ...: print(p.liap_value()) # incorrectly zero
   ...: print(p.agent_liap_value())
1.0
0.0
0.0
0.0

…Error

# Conflicts: # tests/test_game.py

tturocy · 2025-12-17T14:32:28Z

@rahulsavani We now have the new, Nash-correct max_regret and liap_value implemented for mixed behaviour profiles. Or, at least, it should be correct - so you can now add your tests to this branch.

It seems liap_value still comes back erroneously as 0 for the agent-but-not-full-Nash profile for Myserson's fig 4.2:
In [1]: import pygambit as gbt
   ...: g = gbt.read_efg('tests/test_games/myerson_fig_4_2.efg')
   ...: p = g.mixed_behavior_profile([[[0, 1], [0, 1]], [[1, 0]]])
   ...: print(p.max_regret())
   ...: print(p.agent_max_regret())
   ...: print(p.liap_value()) # incorrectly zero
   ...: print(p.agent_liap_value())
1.0
0.0
0.0
0.0

This was not one of my more clever moments. The C++ implementation is correct, it's just that liap_value was literally calling the wrong function (it was calling agent liap internally).

The branch has been updated with this fix. I've also merged in other work on clarifying mixed strategy profile cacheing proactively to fix other potential cache invalidation problems we haven't stumbled upon yet.

Assuming everything is working now, are we ready to merge or do we want to have the example set up as well?

rahulsavani · 2025-12-17T14:48:14Z

Assuming everything is working now, are we ready to merge or do we want to have the example set up as well?

I will do some checks, and I can also do both final tasks, reviewing the tutorials/user guide and adding a tutorial on Myerson fig 4.2.

rahulsavani · 2025-12-17T14:52:29Z

@tturocy: Before merging, it would be good if you could review the new consistency checks in tests/test_behav.py, already pushed.

rahulsavani · 2025-12-17T17:06:14Z

@tturocy: I did a first draft of the notebook. Will check it carefully in the morning. I could incorporate any suggestions for changes then in case you get a chance to look before.

…et.ipynb

rahulsavani · 2025-12-18T08:40:51Z

@edwardchalstrey1: There is now an advanced draft of a new notebook related to this PR, see:

04_agent_versus_non_agent_regret.ipynb

If you get a chance to take a look, feedback would be welcome. We should then also decide how we want to present it/link to it in the user guide.

edwardchalstrey1 · 2025-12-18T10:56:07Z

doc/tutorials/advanced_tutorials/agent_versus_non_agent_regret.ipynb

@@ -0,0 +1,705 @@
+{


Change 3 to 4

Reply via ReviewNB

edwardchalstrey1 · 2025-12-18T10:56:07Z

doc/tutorials/advanced_tutorials/agent_versus_non_agent_regret.ipynb

@@ -0,0 +1,705 @@
+{


Is it worth offering a brief description of the game and if possible, what you might intuitively expect from the equilibria computation?

Reply via ReviewNB

It would definitely be nice. I will check Myerson's text in case a useful description is given, but as I understand it, it is a pathological example constructed specifically to show the difference between agent and non-agent versions of these notions.

edwardchalstrey1 · 2025-12-18T10:56:07Z

doc/tutorials/advanced_tutorials/agent_versus_non_agent_regret.ipynb

@@ -0,0 +1,705 @@
+{


At the moment, the tutorial that explains starting points is in the advanced_tutorials folder, if this assumes the reader has that knowledge already perhaps we need to reorder, or simply put this tutorial in the advanced folder and link to the other one here?

Reply via ReviewNB

Oh, this is definitely not for non-expert users. As per my other reply to your comment, the agent versions are not actually particularly useful as I see it, but we have them already so we chose to keep them.

edwardchalstrey1 · 2025-12-18T10:56:07Z

doc/tutorials/advanced_tutorials/agent_versus_non_agent_regret.ipynb

@@ -0,0 +1,705 @@
+{


If the intended audience isn't likely to already know this, perhaps explain why computing these values is useful

Reply via ReviewNB

Good idea -- it is meant to already be clear that we are interested in these being zero, as that gives a Nash eq, but I agree it will be useful to add that they are both measures of how close to equilibrium we are.

edwardchalstrey1 · 2025-12-18T10:59:31Z

@edwardchalstrey1: There is now an advanced draft of a new notebook related to this PR, see:

04_agent_versus_non_agent_regret.ipynb

If you get a chance to take a look, feedback would be welcome. We should then also decide how we want to present it/link to it in the user guide.

@rahulsavani nice, great to see draw_tree in action already! I have added some comments via ReviewNB - in short, it seems like this is intended to be a tutorial 4 read in the sequence of beginner tutorials, based on how you are describing the concepts/terminology throughout, so I suggested a few points of clarification or where detail could be added. However, I also note that this could go in the advanced tutorials instead, which would make this less necessary.

rahulsavani · 2025-12-18T12:23:38Z

@rahulsavani nice, great to see draw_tree in action already! I have added some comments via ReviewNB - in short, it seems like this is intended to be a tutorial 4 read in the sequence of beginner tutorials, based on how you are describing the concepts/terminology throughout, so I suggested a few points of clarification or where detail could be added. However, I also note that this could go in the advanced tutorials instead, which would make this less necessary.

@edwardchalstrey1 :

Yes, draw_tree is great!

Thanks for the comments -- I put this new one in advanced_tutorials after all.

I also finished the penultimate task which involved replacing agent_max_regret and agent_liap_solve with the standard versions in 03_stripped_down_poker.ipynb. If you can check my work there that would be great.

edwardchalstrey1 · 2025-12-18T13:24:55Z

doc/tutorials/03_stripped_down_poker.ipynb

@@ -39,7 +39,7 @@
  },


Maybe add an extra sentence saying "We set use_strategic=True because..."

Reply via ReviewNB

Excellent comment -- on reflection we should not use it, so I removed it!

…ker.ipynb

rahulsavani · 2025-12-18T16:18:35Z

@tturocy: I am done with this for now, this is ready to merge from my side, unless you have further suggestions for changes.

# Conflicts: # src/tools/liap/liap.cc

…or to analysis. We are going to remove `.sort_infosets()` separately; for the moment this adjusts the tests to add an explicit call to clear the errors for clarity.

# Conflicts: # tests/games.py # tests/test_nash.py

tturocy added 6 commits November 24, 2025 11:23

For mixed behaviors renamed liap_value -> agent_liap_value and `m…

966bec8

…ax_regret` -> `agent_max_regret` in Python

Rename agent functions in MixedBehaviorProfile

89c182c

Separated out enumpure_agent_solve function for clarity.

1b839d5

Add liap_agent_solve and remove that functionality from `liap_solve…

a1f6946

…` for expositional clarity.

Change interface of gambit-liap to be more similar to `gambit-enump…

8cca508

…ure`

Change GUI behavior to use only strategic for enumpure and liap

8512263

tturocy requested a review from rahulsavani November 24, 2025 12:42

tturocy linked an issue Nov 24, 2025 that may be closed by this pull request

ENH: Make behavior profile terminology more transparent and consistent #617

Closed

rahulsavani approved these changes Nov 24, 2025

View reviewed changes

tturocy mentioned this pull request Dec 5, 2025

Modernisation and consistency of regret and liap values #672

Merged

tturocy added 2 commits December 8, 2025 21:25

Merge branch 'master' into dev_terminology

2f69c57

# Conflicts: # doc/tutorials/03_stripped_down_poker.ipynb # src/games/stratmixed.h # src/pygambit/nash.pxi # src/tools/liap/liap.cc # tests/test_behav.py # tests/test_nash.py

Implement new (correct) max_regret and liap_value on mixed behaviour …

b8026c3

…profiles

tturocy requested a review from rahulsavani December 8, 2025 22:19

fixed typo: self.__max_regret -> self._max_regret

0db7bef

rahulsavani mentioned this pull request Dec 16, 2025

Add tests for MixedBehaviorProfile regrets #707

Closed

rahulsavani and others added 8 commits December 16, 2025 18:55

tests for max_regret/infoset regret on mixed behavior profiles

fef1ab1

added tests/test_games/myerson_fig_4_2.efg

65f6a12

Added comprehensive tests for GameStructureChangedError (#700)

ef1482b

agent_{liap_value,max_regret} added to tests for GameStructureChanged…

8e92042

…Error

Merge branch 'master' into dev_terminology

2a4cfbf

# Conflicts: # tests/test_game.py

Merge branch 'master' into dev_terminology

a89032a

Merge branch 'master' into dev_terminology

48460cd

Fix incorrect function called in MixedBehaviorProfile.liap_value

cae6595

notebook showing agent versus non-agent notions

f4de44e

rahulsavani added 3 commits December 17, 2025 17:26

missing file

1794228

using liap_solve and enummixed_solv in 04_agent_versus_non_agent_regr…

f0cec0b

…et.ipynb

improve 04_agent_versus_non_agent_regret.ipynb

2ea0a22

edwardchalstrey1 reviewed Dec 18, 2025

View reviewed changes

rahulsavani added 3 commits December 18, 2025 11:57

move to advanced_tutorials/agent_versus_non_agent_regret.ipynb

fd63135

fix relative import

a494268

agent -> non-agent max_regret/liap_solve in 03_stripped_down_poker.ipynb

698907b

edwardchalstrey1 reviewed Dec 18, 2025

View reviewed changes

removed use_strategic from call to logit_solve in 03_stripped_down_po…

cd2d917

…ker.ipynb

tturocy and others added 11 commits December 19, 2025 11:36

Updated ChangeLog.

c95c11c

Merge branch 'master' into dev_terminology

70a8bfa

# Conflicts: # src/tools/liap/liap.cc

fixing up enumpoly tests

e3b2fd0

problem cases shown in tests

c03048c

test_regrets_tmp

00da58d

clearer example of (one) problem

edcd647

test demonstrating issue is to do with infoset order

a5de7d4

test demonstrating issue is to do with infoset order

7d5628e

Previous test failures were due to not calling .sort_infosets() pri…

c7b3628

…or to analysis. We are going to remove `.sort_infosets()` separately; for the moment this adjusts the tests to add an explicit call to clear the errors for clarity.

Merge branch 'master' into dev_terminology

8375d98

# Conflicts: # tests/games.py # tests/test_nash.py

Tidy up tests now that information set sorting is obsolete.

5f782ce

tturocy marked this pull request as ready for review January 2, 2026 09:57

tturocy merged commit b5ab74f into master Jan 2, 2026
26 checks passed

tturocy deleted the dev_terminology branch January 2, 2026 10:01

satyam102006 mentioned this pull request Feb 21, 2026

Add tests for MixedBehaviorProfile regrets (Resolves #707) #781

Closed

Conversation

tturocy commented Nov 24, 2025 • edited by rahulsavani Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

rahulsavani left a comment

Choose a reason for hiding this comment

Uh oh!

tturocy commented Nov 25, 2025

Uh oh!

review-notebook-app bot commented Dec 1, 2025

Uh oh!

tturocy commented Dec 8, 2025

Uh oh!

rahulsavani commented Dec 11, 2025

Uh oh!

tturocy commented Dec 17, 2025

Uh oh!

rahulsavani commented Dec 17, 2025

Uh oh!

rahulsavani commented Dec 17, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

rahulsavani commented Dec 17, 2025

Uh oh!

rahulsavani commented Dec 18, 2025

Uh oh!

edwardchalstrey1 Dec 18, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

edwardchalstrey1 Dec 18, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

rahulsavani Dec 18, 2025

Choose a reason for hiding this comment

Uh oh!

edwardchalstrey1 Dec 18, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

rahulsavani Dec 18, 2025

Choose a reason for hiding this comment

Uh oh!

edwardchalstrey1 Dec 18, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

rahulsavani Dec 18, 2025

Choose a reason for hiding this comment

Uh oh!

edwardchalstrey1 commented Dec 18, 2025

Uh oh!

rahulsavani commented Dec 18, 2025

Uh oh!

edwardchalstrey1 Dec 18, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

rahulsavani Dec 18, 2025

Choose a reason for hiding this comment

Uh oh!

rahulsavani commented Dec 18, 2025

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

tturocy commented Nov 24, 2025 •

edited by rahulsavani

Loading

rahulsavani commented Dec 17, 2025 •

edited

Loading

edwardchalstrey1 Dec 18, 2025 •

edited

Loading

edwardchalstrey1 Dec 18, 2025 •

edited

Loading

edwardchalstrey1 Dec 18, 2025 •

edited

Loading

edwardchalstrey1 Dec 18, 2025 •

edited

Loading

edwardchalstrey1 Dec 18, 2025 •

edited

Loading