noregret 0.0.0.dev9__tar.gz → 0.0.0.dev10__tar.gz

{noregret-0.0.0.dev9 → noregret-0.0.0.dev10}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: noregret
-Version: 0.0.0.dev9
+Version: 0.0.0.dev10
 Summary: No-regret learning dynamics
 Home-page: https://github.com/uoftcprg/noregret
 Author: Universal, Open, Free, and Transparent Computer Poker Research Group

{noregret-0.0.0.dev9 → noregret-0.0.0.dev10}/noregret/__init__.py

@@ -17,6 +17,7 @@ from noregret.games import (
     RockPaperScissors,
     RockPaperScissorsPlus,
     RockPaperSuperscissors,
+    Simulation,
     StagHunt,
     StrategyProfile,
     to_extensive_form_game,
@@ -67,7 +68,7 @@ from noregret.solvers import (
     stochastic_regret_minimization,
     symmetric_regret_minimization,
 )
-from noregret.utilities import import_object, sample, tuple_or_none
+from noregret.utilities import import_object, tuple_or_none
 
 BM = BlumMansour
 """Alias for :class:`noregret.BlumMansour`."""
@@ -121,6 +122,8 @@ RM = RegretMatching
 """Alias for :class:`noregret.RegretMatching`."""
 rm = regret_minimization
 """Alias for :func:`noregret.regret_minimization`."""
+Sim = Simulation
+"""Alias for :class:`noregret.Simulation`."""
 stochastic_rm = stochastic_regret_minimization
 """Alias for :func:`noregret.stochastic_regret_minimization`."""
 symmetric_rm = symmetric_regret_minimization
@@ -194,10 +197,11 @@ __all__ = (
     'RockPaperScissors',
     'RockPaperScissorsPlus',
     'RockPaperSuperscissors',
-    'sample',
     'SequenceFormPolytope',
     'SequenceFormPolytopeRegretMinimizer',
     'Serializable',
+    'Sim',
+    'Simulation',
     'StagHunt',
     'stochastic_regret_minimization',
     'StochasticRegretMinimizer',

{noregret-0.0.0.dev9 → noregret-0.0.0.dev10}/noregret/games/__init__.py

@@ -2,6 +2,7 @@
 from noregret.games.black_box import (
     BlackBoxGame,
     open_spiel_game,
+    Simulation,
     StrategyProfile,
     UniformStrategyProfile,
 )
@@ -53,6 +54,7 @@ __all__ = (
     'RockPaperScissors',
     'RockPaperScissorsPlus',
     'RockPaperSuperscissors',
+    'Simulation',
     'StagHunt',
     'StrategyProfile',
     'to_extensive_form_game',

{noregret-0.0.0.dev9 → noregret-0.0.0.dev10}/noregret/games/black_box.py

@@ -2,6 +2,7 @@
 from abc import ABC, abstractmethod
 from dataclasses import dataclass, field
 from functools import partial
+from typing import Any
 
 from ordered_set import OrderedSet
 from pyspiel import exploitability, GameType, load_game
@@ -9,6 +10,39 @@ from pyspiel import exploitability, GameType, load_game
 from noregret.kernels import Kernel
 
 
+@dataclass
+class Simulation:
+    """Class for simulations."""
+    kernel: Kernel
+    """Kernel."""
+    players: list[int]
+    """Players."""
+    decision_points: list[str | None]
+    """Decision points."""
+    actions: list[str]
+    """Actions."""
+    utilities: Any
+    """Utilities."""
+
+    def sequences(self, player=None):
+        """Return sequences given an optional player.
+
+        :param player: Optional player.
+        :return: Sequences.
+        """
+        for i, j, a in zip(self.players, self.decision_points, self.actions):
+            if i is not None and (player is None or i == player):
+                yield j, a
+
+    def utility(self, player):
+        """Return the utility given a player.
+
+        :param player: Player.
+        :return: Utility.
+        """
+        return self.utilities[player]
+
+
 @dataclass
 class BlackBoxGame(ABC):
     """Abstract base class for black box games."""
@@ -140,11 +174,53 @@ class BlackBoxGame(ABC):
         return np.array(ps, dtype)
 
     def exploitability(self, strategy_profile):
+        """Return exploitability given a strategy profile.
+
+        :param strategy_profile: Strategy profile.
+        :return: Exploitability.
+        """
         if not self.is_two_player or not self.is_zero_sum:
             raise ValueError('not 2p0s')
 
         raise NotImplementedError
 
+    def simulate(self, strategy_profile):
+        """Run a simulation given a strategy profile.
+
+        :param strategy_profile: Strategy profile.
+        :return: Simulation.
+        """
+        np = self.kernel.numpy
+        is_ = []
+        js = []
+        as_ = []
+        h = self.root_node
+
+        while A := self.actions(h):
+            i = self.player(h)
+
+            if i is None:
+                j = None
+                ps = self.chance_probabilities(h)
+            else:
+                j = self.information_set(h)
+                ps = strategy_profile(h)
+
+            a = np.random.choice(A, p=ps).item()
+            h = self.apply(h, a)
+
+            is_.append(i)
+            js.append(j)
+            as_.append(a)
+
+        is_ = tuple(is_)
+        js = tuple(js)
+        as_ = tuple(as_)
+        us = self.utilities(h)
+        simulation = Simulation(self.kernel, is_, js, as_, us)
+
+        return simulation
+
 
 @dataclass
 class _OpenSpielBlackBoxGame(BlackBoxGame):

{noregret-0.0.0.dev9 → noregret-0.0.0.dev10}/noregret/regret_minimizers/stochastic.py

@@ -10,7 +10,6 @@ from noregret.regret_minimizers.probability_simplices import (
     ProbabilitySimplexRegretMinimizer,
     RegretMatching,
 )
-from noregret.utilities import sample
 
 
 @dataclass
@@ -105,7 +104,7 @@ class StochasticRegretMinimizer(ABC):
                 us[j] = np.array(u_primes, dtype)
                 u += us[j] @ ps
             else:
-                a = sample(A, ps)
+                a = np.random.choice(A, p=ps).item()
                 h_prime = self.game.apply(h, a)
                 u += self._external_sampling(i, us, h_prime)
 
@@ -135,7 +134,7 @@ class StochasticRegretMinimizer(ABC):
             else:
                 ps = self._action_probabilities(h)
 
-            k = sample(range(len(A)), ps)
+            k = np.random.choice(len(A), p=ps)
             a = A[k]
             h_prime = self.game.apply(h, a)
             p_prime = ps[k] * p

{noregret-0.0.0.dev9 → noregret-0.0.0.dev10}/noregret/tests/test_games.py

@@ -6,6 +6,7 @@ import noregret as nr
 
 class GameTestCaseMixin(ABC):
     KER = None
+    GAMES = None
 
     @abstractmethod
     def uniform_strategy_profile(self, game):
@@ -145,12 +146,43 @@ class ExtensiveFormGameTestCase(GameTestCaseMixin, TestCase):
                 self.assertEqual(sfp.parent_sequences, sfp2.parent_sequences)
 
 
+class SimulationTestCase(TestCase):
+    KER = nr.FPKer()
+
+    def test_sequences(self):
+        np = self.KER.numpy
+        dtype = self.KER.data_type
+        sim = nr.Sim(
+            self.KER,
+            (0, None, 0, 1),
+            ('', None, 'ab', 'b'),
+            ('a', 'b', 'c', 'd'),
+            np.array([1, -1], dtype),
+        )
+
+        self.assertEqual(
+            tuple(sim.sequences()),
+            (('', 'a'), ('ab', 'c'), ('b', 'd')),
+        )
+        self.assertEqual(tuple(sim.sequences(0)), (('', 'a'), ('ab', 'c')))
+        self.assertEqual(tuple(sim.sequences(1)), (('b', 'd'),))
+
+    def test_utility(self):
+        np = self.KER.numpy
+        dtype = self.KER.data_type
+        sim = nr.Sim(self.KER, (), (), (), np.array([1, -1], dtype))
+
+        self.assertEqual(sim.utility(0), 1)
+        self.assertEqual(sim.utility(1), -1)
+
+
 class BlackBoxGameTestCase(TestCase):
     KER = nr.FPKer()
     GAMES = (
         nr.open_spiel_game(KER, 'kuhn_poker'),
         nr.open_spiel_game(KER, 'leduc_poker'),
     )
+    SEED = 42
 
     def test_actions_and_children(self):
         for game in self.GAMES:
@@ -210,6 +242,16 @@ class BlackBoxGameTestCase(TestCase):
 
             self.assertAlmostEqual(epsilon, epsilon2)
 
+    def test_simulation(self):
+        np = self.KER.numpy
+
+        for game in self.GAMES:
+            np.random.seed(self.SEED)
+
+            sigma = nr.UniformStrategyProfile(self.KER, game)
+
+            game.simulate(sigma)
+
 
 if __name__ == '__main__':
     main()  # pragma: no cover

{noregret-0.0.0.dev9 → noregret-0.0.0.dev10}/noregret/tests/test_regret_minimization.py

@@ -1,6 +1,5 @@
 from functools import partial
 from math import inf
-from random import seed
 from unittest import main, TestCase
 
 import noregret as nr
@@ -218,9 +217,11 @@ class StochasticRegretMinimizationTestCase(TestCase):
     SEED = 42
 
     def test_external_sampling(self):
+        np = self.KER.numpy
+
         assert self.GAME.is_two_player and self.GAME.is_zero_sum
 
-        seed(self.SEED)
+        np.random.seed(self.SEED)
 
         R = nr.MCCFR(self.KER, self.GAME)
         sigma = nr.stochastic_rm(
@@ -235,9 +236,11 @@ class StochasticRegretMinimizationTestCase(TestCase):
         self.assertLess(epsilon, self.TARGET_EXPLOITABILITY)
 
     def test_outcome_sampling(self):
+        np = self.KER.numpy
+
         assert self.GAME.is_two_player and self.GAME.is_zero_sum
 
-        seed(self.SEED)
+        np.random.seed(self.SEED)
 
         R = nr.MCCFR(
             self.KER,

{noregret-0.0.0.dev9 → noregret-0.0.0.dev10}/noregret/utilities.py

@@ -1,6 +1,5 @@
 """Module for utilities."""
 from importlib import import_module
-from random import choices
 
 
 def import_object(object_path):
@@ -35,16 +34,3 @@ def tuple_or_none(values):
     :return: Tuple or ``None``.
     """
     return None if values is None else tuple(values)
-
-
-def sample(values, probabilities):
-    """Sample a random value as per the probabilities.
-
-    >>> sample(range(5), [0, 0, 1, 0, 0])
-    2
-
-    :param values: Values to be sampled from.
-    :param probabilities: The probabilities of sampling each value.
-    :return: The sampled value.
-    """
-    return choices(values, probabilities)[0]

{noregret-0.0.0.dev9 → noregret-0.0.0.dev10}/noregret.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: noregret
-Version: 0.0.0.dev9
+Version: 0.0.0.dev10
 Summary: No-regret learning dynamics
 Home-page: https://github.com/uoftcprg/noregret
 Author: Universal, Open, Free, and Transparent Computer Poker Research Group

{noregret-0.0.0.dev9 → noregret-0.0.0.dev10}/setup.py

@@ -4,7 +4,7 @@ from setuptools import find_packages, setup
 
 setup(
     name='noregret',
-    version='0.0.0.dev9',
+    version='0.0.0.dev10',
     description='No-regret learning dynamics',
     long_description=open('README.rst').read(),
     long_description_content_type='text/x-rst',