continual-foragax 0.36.1__tar.gz → 0.38.0__tar.gz

{continual_foragax-0.36.1 → continual_foragax-0.38.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: continual-foragax
-Version: 0.36.1
+Version: 0.38.0
 Summary: A continual reinforcement learning benchmark
 Author-email: Steven Tang <stang5@ualberta.ca>
 Requires-Python: >=3.8

{continual_foragax-0.36.1 → continual_foragax-0.38.0}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "continual-foragax"
-version = "0.36.1"
+version = "0.38.0"
 description = "A continual reinforcement learning benchmark"
 readme = "README.md"
 authors = [
@@ -30,7 +30,7 @@ build-backend = "setuptools.build_meta"
 [tool]
 [tool.commitizen]
 name = "cz_conventional_commits"
-version = "0.36.1"
+version = "0.38.0"
 tag_format = "$version"
 version_files = ["pyproject.toml"]
 

{continual_foragax-0.36.1 → continual_foragax-0.38.0}/src/continual_foragax.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: continual-foragax
-Version: 0.36.1
+Version: 0.38.0
 Summary: A continual reinforcement learning benchmark
 Author-email: Steven Tang <stang5@ualberta.ca>
 Requires-Python: >=3.8

{continual_foragax-0.36.1 → continual_foragax-0.38.0}/src/foragax/env.py

@@ -132,6 +132,7 @@ class ForagaxEnv(environment.Environment):
         observation_type: str = "object",
         dynamic_biomes: bool = False,
         biome_consumption_threshold: float = 0.9,
+        dynamic_biome_spawn_empty: float = 0.0,
         max_expiries_per_step: int = 1,
     ):
         super().__init__()
@@ -156,6 +157,7 @@ class ForagaxEnv(environment.Environment):
         self.teleport_interval = teleport_interval
         self.dynamic_biomes = dynamic_biomes
         self.biome_consumption_threshold = biome_consumption_threshold
+        self.dynamic_biome_spawn_empty = dynamic_biome_spawn_empty
         if max_expiries_per_step < 1:
             raise ValueError("max_expiries_per_step must be at least 1")
         self.max_expiries_per_step = max_expiries_per_step
@@ -888,7 +890,7 @@ class ForagaxEnv(environment.Environment):
         biome_freqs = self.biome_object_frequencies[biome_idx]
         biome_mask = self.biome_masks_array[biome_idx]
 
-        key, spawn_key, color_key, params_key = jax.random.split(key, 4)
+        key, spawn_key, color_key, params_key, dropout_key = jax.random.split(key, 5)
 
         # Generate object IDs using deterministic or random spawn
         if deterministic:
@@ -927,6 +929,13 @@ class ForagaxEnv(environment.Environment):
                 jnp.searchsorted(cumulative_freqs, grid_rand, side="right") - 1
             )
 
+        # Apply random dropout if configured
+        if self.dynamic_biome_spawn_empty > 0:
+            dropout_mask = jax.random.bernoulli(
+                dropout_key, 1.0 - self.dynamic_biome_spawn_empty, object_grid.shape
+            )
+            object_grid = jnp.where(dropout_mask, object_grid, 0)
+
         # Initialize color grid
         color_grid = jnp.full((self.size[1], self.size[0], 3), 255, dtype=jnp.uint8)
 

{continual_foragax-0.36.1 → continual_foragax-0.38.0}/src/foragax/objects.py

@@ -241,6 +241,7 @@ class FourierObject(BaseForagaxObject):
         reward_delay: int = 0,
         max_reward_delay: Optional[int] = None,
         regen_delay: Optional[Tuple[int, int]] = None,
+        reward_repeat: int = 1,
     ):
         if max_reward_delay is None:
             max_reward_delay = reward_delay
@@ -257,6 +258,7 @@ class FourierObject(BaseForagaxObject):
         self.base_magnitude = base_magnitude
         self.reward_delay_val = reward_delay
         self.regen_delay_range = regen_delay
+        self.reward_repeat = reward_repeat
 
     def get_state(self, key: jax.Array) -> jax.Array:
         """Generate random Fourier series parameters.
@@ -321,7 +323,7 @@ class FourierObject(BaseForagaxObject):
         y_max = params[2]
 
         # Normalize time to [0, 2π] using the object's period
-        t = 2.0 * jnp.pi * (clock % period) / period
+        t = 2.0 * jnp.pi * ((clock // self.reward_repeat) % period) / period
 
         # Extract interleaved coefficients: [a1, b1, a2, b2, ...]
         ab_coeffs = params[3:]
@@ -715,6 +717,7 @@ def create_fourier_objects(
     base_magnitude: float = 1.0,
     reward_delay: int = 0,
     regen_delay: Optional[Tuple[int, int]] = None,
+    reward_repeat: int = 1,
 ):
     """Create HOT and COLD FourierObject instances.
 
@@ -733,6 +736,7 @@ def create_fourier_objects(
         color=(0, 0, 0),
         reward_delay=reward_delay,
         regen_delay=regen_delay,
+        reward_repeat=reward_repeat,
     )
 
     cold = FourierObject(
@@ -742,6 +746,7 @@ def create_fourier_objects(
         color=(0, 0, 0),
         reward_delay=reward_delay,
         regen_delay=regen_delay,
+        reward_repeat=reward_repeat,
     )
 
     return hot, cold

{continual_foragax-0.36.1 → continual_foragax-0.38.0}/src/foragax/registry.py

@@ -128,6 +128,32 @@ ENV_CONFIGS: Dict[str, Dict[str, Any]] = {
         "deterministic_spawn": True,
         "dynamic_biomes": True,
     },
+    "ForagaxDiwali-v4": {
+        "size": (15, 15),
+        "aperture_size": None,
+        "objects": None,
+        "biomes": (
+            # Hot biome
+            Biome(start=(0, 2), stop=(15, 6), object_frequencies=(0.5, 0.0)),
+            # Cold biome
+            Biome(start=(0, 9), stop=(15, 13), object_frequencies=(0.0, 0.5)),
+        ),
+        "nowrap": False,
+        "deterministic_spawn": True,
+        "dynamic_biomes": True,
+        "biome_consumption_threshold": 1000,
+    },
+    "ForagaxDiwali-v5": {
+        "size": (15, 15),
+        "aperture_size": None,
+        "objects": None,
+        "biomes": (Biome(start=(0, 0), stop=(15, 15), object_frequencies=(0.2,)),),
+        "nowrap": False,
+        "deterministic_spawn": True,
+        "dynamic_biomes": True,
+        "biome_consumption_threshold": 1000,
+        "dynamic_biome_spawn_empty": 0.1,
+    },
     "ForagaxTwoBiome-v1": {
         "size": (15, 15),
         "aperture_size": None,
@@ -575,6 +601,20 @@ def make(
             reward_delay=reward_delay,
             regen_delay=(9, 11),
         )[:1]
+    if env_id == "ForagaxDiwali-v4":
+        config["objects"] = create_fourier_objects(
+            num_fourier_terms=10,
+            reward_delay=reward_delay,
+            regen_delay=(9, 11),
+            reward_repeat=100,
+        )
+    if env_id == "ForagaxDiwali-v5":
+        config["objects"] = create_fourier_objects(
+            num_fourier_terms=10,
+            reward_delay=reward_delay,
+            regen_delay=(9, 11),
+            reward_repeat=100,
+        )[:1]
 
     if env_id == "ForagaxSineTwoBiome-v1":
         biome1_oyster, biome1_deathcap, biome2_oyster, biome2_deathcap = (

{continual_foragax-0.36.1 → continual_foragax-0.38.0}/tests/test_foragax.py

@@ -2811,13 +2811,14 @@ def test_fourier_reward_mathematical_properties():
     key = jax.random.key(42)
 
     # Create a simple single-term Fourier series: a1*cos(t) + b1*sin(t)
+    # We use period=8.0 so that clock=1 corresponds to phase 2π/8 = π/4
     simple_params = jnp.array(
         [
-            2 * jnp.pi,
+            8.0,  # period
             -1.0,
-            1.0,  # period=2π, min=-1, max=1
+            1.0,  # min=-1, max=1
             0.5,
-            0.5,  # a1=0.5, b1=0.5 (should give 0.5*(cos(t) + sin(t)))
+            0.5,  # a1=0.5, b1=0.5
             0.0,
             0.0,
             0.0,
@@ -2828,7 +2829,8 @@ def test_fourier_reward_mathematical_properties():
         dtype=jnp.float32,
     )
 
-    # At t=0: 0.5*(cos(0) + sin(0)) = 0.5*(1 + 0) = 0.5
+    # At t=0 (clock=0): phase=0
+    # 0.5*(cos(0) + sin(0)) = 0.5*(1 + 0) = 0.5
     reward_t0_simple = fourier_obj.reward(0, key, simple_params)
     # The actual reward is normalized: 2*(raw - min)/(max - min) - 1, then * base_magnitude
     # raw = 0.5, min=-1, max=1, so: 2*(0.5 - (-1))/(1 - (-1)) - 1 = 2*(1.5)/2 - 1 = 1.5 - 1 = 0.5
@@ -2837,9 +2839,10 @@ def test_fourier_reward_mathematical_properties():
         f"Expected ~1.0, got {reward_t0_simple}"
     )
 
-    # At t=π/4: 0.5*(cos(π/4) + sin(π/4)) = 0.5*(0.707 + 0.707) = 0.5*1.414 ≈ 0.707
+    # At t=1 (clock=1): phase = 2π * 1/8 = π/4
+    # 0.5*(cos(π/4) + sin(π/4)) = 0.5*(0.707 + 0.707) = 0.5*1.414 ≈ 0.707
+    reward_t_quarter = fourier_obj.reward(1, key, simple_params)
     t_quarter = jnp.pi / 4
-    reward_t_quarter = fourier_obj.reward(t_quarter, key, simple_params)
     raw_quarter = 0.5 * (jnp.cos(t_quarter) + jnp.sin(t_quarter))
     expected_quarter = (
         2 * (raw_quarter - (-1)) / (1 - (-1)) - 1