slight dependency updates + cleanup on FA

Author: TheEimer

Makefile

@@ -71,7 +71,7 @@ clean: clean-doc clean-build clean-examples
 
 # Build a distribution in ./dist
 build:
-	$(PYTHON) setup.py sdist
+	uv build
 
 doc:
 	$(MAKE) -C ${DOCDIR} docs
@@ -82,12 +82,12 @@ doc:
 # Publish to testpypi
 # Will echo the commands to actually publish to be run to publish to actual PyPi
 # This is done to prevent accidental publishing but provide the same conveniences
-publish: clean-build build
+publish: clean build
 	$(PIP) install twine
 	$(PYTHON) -m twine upload --verbose --repository testpypi ${DIST}/*
 	@echo
 	@echo "Test with the following line:"
 	@echo "pip install --index-url https://test.pypi.org/simple/ dacbench"
 	@echo
 	@echo "Once you have decided it works, publish to actual pypi with"
-	@echo "python -m twine upload dist/*"
+	@echo "python -m twine upload dist/*"

dacbench/__init__.py

@@ -6,7 +6,7 @@
 from dacbench.abstract_benchmark import AbstractBenchmark
 from dacbench.abstract_env import AbstractEnv, AbstractMADACEnv
 
-__all__ = ["AbstractEnv", "AbstractMADACEnv", "AbstractBenchmark"]
+__all__ = ["AbstractBenchmark", "AbstractEnv", "AbstractMADACEnv"]
 
 from gymnasium.envs.registration import register
 

dacbench/abstract_env.py

@@ -190,9 +190,9 @@ def reset_(
         if scheme is None:
             scheme = options.get("scheme", self.instance_updates)
         if instance is None:
-            instance = options.get("instance", None)
+            instance = options.get("instance")
         if instance_id is None:
-            instance_id = options.get("instance_id", None)
+            instance_id = options.get("instance_id")
         self.use_next_instance(instance, instance_id, scheme=scheme)
 
     def use_next_instance(self, instance=None, instance_id=None, scheme=None):

dacbench/agents/__init__.py

@@ -2,4 +2,4 @@
 from dacbench.agents.generic_agent import GenericAgent
 from dacbench.agents.simple_agents import RandomAgent, StaticAgent
 
-__all__ = ["StaticAgent", "RandomAgent", "GenericAgent", "DynamicRandomAgent"]
+__all__ = ["DynamicRandomAgent", "GenericAgent", "RandomAgent", "StaticAgent"]

dacbench/benchmarks/__init__.py

@@ -10,8 +10,8 @@
 from dacbench.benchmarks.toysgd_benchmark import ToySGDBenchmark
 
 __all__ = [
-    "LubyBenchmark",
     "FunctionApproximationBenchmark",
+    "LubyBenchmark",
     "ToySGDBenchmark",
     #    "FastDownwardBenchmark",
 ]
@@ -28,7 +28,7 @@
         "please follow the installation guide."
     )
 
-sgd_spec = importlib.util.find_spec("torch")
+sgd_spec = importlib.util.find_spec("torchvision")
 found = sgd_spec is not None
 if found:
     from dacbench.benchmarks.sgd_benchmark import SGDBenchmark

dacbench/benchmarks/function_approximation_benchmark.py

@@ -82,7 +82,14 @@ def __init__(self, config_path=None, config=None):
 
         for key in FUNCTION_APPROXIMATION_DEFAULTS:
             if key not in self.config:
-                self.config[key] = FUNCTION_APPROXIMATION_DEFAULTS[key]
+                if key == "observation_space_args" and "config_space" in self.config:
+                    obs_length = 1 + len(self.config["config_space"]) * 4
+                    self.config[key] = [
+                        np.array([-np.inf for _ in range(obs_length)]),
+                        np.array([np.inf for _ in range(obs_length)]),
+                    ]
+                else:
+                    self.config[key] = FUNCTION_APPROXIMATION_DEFAULTS[key]
 
     def get_environment(self):
         """Return Function Approximation env with current configuration.
@@ -197,6 +204,10 @@ def get_benchmark(self, dimension=None, seed=0):
                 "Slope (dimension 1)",
                 "Action",
             ]
+            self.config.observation_space_args = [
+                np.array([-np.inf for _ in range(4)]),
+                np.array([np.inf for _ in range(4)]),
+            ]
         if dimension == 2:
             self.config.instance_set_path = "sigmoid_2D3M_train.csv"
             self.config.test_set_path = "sigmoid_2D3M_test.csv"
@@ -220,6 +231,10 @@ def get_benchmark(self, dimension=None, seed=0):
                 "Action dim 1",
                 "Action dim 2",
             ]
+            self.config.observation_space_args = [
+                np.array([-np.inf for _ in range(7)]),
+                np.array([np.inf for _ in range(7)]),
+            ]
         if dimension == 3:
             self.config.instance_set_path = "sigmoid_3D3M_train.csv"
             self.config.test_set_path = "sigmoid_3D3M_test.csv"
@@ -250,6 +265,10 @@ def get_benchmark(self, dimension=None, seed=0):
                 "Action 2",
                 "Action 3",
             ]
+            self.config.observation_space_args = [
+                np.array([-np.inf for _ in range(10)]),
+                np.array([np.inf for _ in range(10)]),
+            ]
         if dimension == 5:
             self.config.instance_set_path = "sigmoid_5D3M_train.csv"
             self.config.test_set_path = "sigmoid_5D3M_test.csv"
@@ -294,6 +313,10 @@ def get_benchmark(self, dimension=None, seed=0):
                 "Action 4",
                 "Action 5",
             ]
+            self.config.observation_space_args = [
+                np.array([-np.inf for _ in range(16)]),
+                np.array([np.inf for _ in range(16)]),
+            ]
         self.config.seed = seed
         self.read_instance_set()
         self.read_instance_set(test=True)

dacbench/envs/__init__.py

@@ -12,15 +12,15 @@
 from dacbench.envs.toysgd import ToySGDEnv, ToySGDInstance
 
 __all__ = [
-    "LubyEnv",
-    "LubyInstance",
-    "luby_gen",
     "FunctionApproximationEnv",
     "FunctionApproximationInstance",
+    "LubyEnv",
+    "LubyInstance",
+    "TheoryEnv",
     #   "FastDownwardEnv",
     "ToySGDEnv",
     "ToySGDInstance",
-    "TheoryEnv",
+    "luby_gen",
 ]
 
 modcma_spec = importlib.util.find_spec("modcma")

dacbench/envs/theory.py

@@ -224,11 +224,8 @@ def crossover(
             locs_x[locs_xprime] = False
             obj = self.get_fitness_after_crossover(xprime, locs_x, locs_xprime)
 
-            if (
-                obj not in (self.fitness, xprime.fitness)
-                or (
-                    not np.array_equal(xprime.data[locs_xprime], self.data[locs_xprime])
-                )
+            if obj not in (self.fitness, xprime.fitness) or (
+                (not np.array_equal(xprime.data[locs_xprime], self.data[locs_xprime]))
                 and (not np.array_equal(self.data[locs_x], xprime.data[locs_x]))
             ):
                 n_evals += 1

dacbench/wrappers/__init__.py

@@ -12,11 +12,11 @@
     "ActionFrequencyWrapper",
     "EpisodeTimeWrapper",
     "InstanceSamplingWrapper",
+    "MultiDiscreteActionWrapper",
+    "ObservationWrapper",
+    "PerformanceTrackingWrapper",
+    "PolicyProgressWrapper",
     "PolicyProgressWrapper",
     "RewardNoiseWrapper",
     "StateTrackingWrapper",
-    "PerformanceTrackingWrapper",
-    "PolicyProgressWrapper",
-    "ObservationWrapper",
-    "MultiDiscreteActionWrapper",
 ]