continual-foragax 0.20.1__py3-none-any.whl → 0.22.0__py3-none-any.whl

{continual_foragax-0.20.1.dist-info → continual_foragax-0.22.0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: continual-foragax
-Version: 0.20.1
+Version: 0.22.0
 Summary: A continual reinforcement learning benchmark
 Author-email: Steven Tang <stang5@ualberta.ca>
 Requires-Python: >=3.8

{continual_foragax-0.20.1.dist-info → continual_foragax-0.22.0.dist-info}/RECORD

@@ -1,8 +1,8 @@
 foragax/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 foragax/colors.py,sha256=rqNPiywP4Nvr0POhsGpasRk-nMMTS3DOwFRUgperlUk,2065
-foragax/env.py,sha256=phGPMzwNKbwLLIyme1bjca3l2MLNKfYYBmly7JP8cLc,20498
-foragax/objects.py,sha256=iDFo_2CjpgErAm3QdQt5ixmQ8jdIvl7siIPqzGwVcGk,7665
-foragax/registry.py,sha256=5Ik61gIzTHc5A1_aOVbRXbVooRXz3GjmL7KdGsZTOTE,8642
+foragax/env.py,sha256=Q-96fMoA_51TIJky2JXApITQHXC-1QfdeB5VZvNwe0o,21362
+foragax/objects.py,sha256=8tBFMiquWCkhOpNndNmzovMjw7lE5P81OOlUvN2F65w,8301
+foragax/registry.py,sha256=pRBWGP18jd4NKl1H-rwDYaAJKUgRWVfENQ9pvTS0tAw,9462
 foragax/rendering.py,sha256=bms7wvBZTofoR-K-2QD2Ggeed7Viw8uwAEiEpEM3eSo,2768
 foragax/weather.py,sha256=KNAiwuFz8V__6G75vZIWQKPocLzXqxXn-Vt4TbHIpcA,1258
 foragax/data/ECA_non-blended_custom/TG_SOUID100897.txt,sha256=N7URbX6VlCZvCboUogYjMzy1I-0cfNPOn0QTLSHHfQ0,1776751
@@ -128,8 +128,8 @@ foragax/data/ECA_non-blended_custom/TG_SOUID156887.txt,sha256=juzTPgJoJxfqmZkorL
 foragax/data/ECA_non-blended_custom/elements.txt,sha256=OtcUBoDAHxuln79BPKGu0tsQxG_5G2BfAX3Ck130kEA,4507
 foragax/data/ECA_non-blended_custom/metadata.txt,sha256=nudnmOCy5cPJfSXt_IjyX0S5-T7NkCZREICZSimqeqc,48260
 foragax/data/ECA_non-blended_custom/sources.txt,sha256=1j3lSmINAoCMqPqFrHfZJriOz6sTYZNOhXzUwvTLas0,20857
-continual_foragax-0.20.1.dist-info/METADATA,sha256=UlkWfKTFOp5OPzmJ6ojc9I1aEtqCngdhNFcnX-WZ7gQ,4897
-continual_foragax-0.20.1.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-continual_foragax-0.20.1.dist-info/entry_points.txt,sha256=Qiu6iE_XudrDO_bVAMeA435h4PO9ourt8huvSHiuMPc,41
-continual_foragax-0.20.1.dist-info/top_level.txt,sha256=-z3SDK6RfLIcLI24n8rdbeFzlVY3hunChzlu-v1Fncs,8
-continual_foragax-0.20.1.dist-info/RECORD,,
+continual_foragax-0.22.0.dist-info/METADATA,sha256=WCVeg6996zpBsLWNghDpibCAs70CDaI8KStzCFnSPNM,4897
+continual_foragax-0.22.0.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+continual_foragax-0.22.0.dist-info/entry_points.txt,sha256=Qiu6iE_XudrDO_bVAMeA435h4PO9ourt8huvSHiuMPc,41
+continual_foragax-0.22.0.dist-info/top_level.txt,sha256=-z3SDK6RfLIcLI24n8rdbeFzlVY3hunChzlu-v1Fncs,8
+continual_foragax-0.22.0.dist-info/RECORD,,

foragax/env.py

@@ -10,6 +10,7 @@ from typing import Any, Dict, Tuple, Union
 
 import jax
 import jax.numpy as jnp
+import numpy as np
 from flax import struct
 from gymnax.environments import environment, spaces
 
@@ -66,13 +67,16 @@ class ForagaxEnv(environment.Environment):
 
     def __init__(
         self,
+        name: str = "Foragax-v0",
         size: Union[Tuple[int, int], int] = (10, 10),
         aperture_size: Union[Tuple[int, int], int] = (5, 5),
         objects: Tuple[BaseForagaxObject, ...] = (),
         biomes: Tuple[Biome, ...] = (Biome(object_frequencies=()),),
         nowrap: bool = False,
+        deterministic_spawn: bool = False,
     ):
         super().__init__()
+        self._name = name
         if isinstance(size, int):
             size = (size, size)
         self.size = size
@@ -81,6 +85,7 @@ class ForagaxEnv(environment.Environment):
             aperture_size = (aperture_size, aperture_size)
         self.aperture_size = aperture_size
         self.nowrap = nowrap
+        self.deterministic_spawn = deterministic_spawn
         objects = (EMPTY,) + objects
         if self.nowrap:
             objects = objects + (PADDING,)
@@ -103,12 +108,35 @@ class ForagaxEnv(environment.Environment):
         self.biome_object_frequencies = jnp.array(
             [b.object_frequencies for b in biomes]
         )
-        self.biome_starts = jnp.array(
+        self.biome_starts = np.array(
             [b.start if b.start is not None else (-1, -1) for b in biomes]
         )
-        self.biome_stops = jnp.array(
+        self.biome_stops = np.array(
             [b.stop if b.stop is not None else (-1, -1) for b in biomes]
         )
+        self.biome_sizes = np.prod(self.biome_stops - self.biome_starts, axis=1)
+        self.biome_masks = []
+        for i in range(self.biome_object_frequencies.shape[0]):
+            # Create mask for the biome
+            start = jax.lax.select(
+                self.biome_starts[i, 0] == -1,
+                jnp.array([0, 0]),
+                self.biome_starts[i],
+            )
+            stop = jax.lax.select(
+                self.biome_stops[i, 0] == -1,
+                jnp.array(self.size),
+                self.biome_stops[i],
+            )
+            rows = jnp.arange(self.size[1])[:, None]
+            cols = jnp.arange(self.size[0])
+            mask = (
+                (rows >= start[1])
+                & (rows < stop[1])
+                & (cols >= start[0])
+                & (cols < stop[0])
+            )
+            self.biome_masks.append(mask)
 
     @property
     def default_params(self) -> EnvParams:
@@ -196,57 +224,18 @@ class ForagaxEnv(environment.Environment):
         self, key: jax.Array, params: EnvParams
     ) -> Tuple[jax.Array, EnvState]:
         """Reset environment state."""
-        key, subkey = jax.random.split(key)
-
         object_grid = jnp.zeros((self.size[1], self.size[0]), dtype=int)
-
-        iter_key = subkey
+        key, iter_key = jax.random.split(key)
         for i in range(self.biome_object_frequencies.shape[0]):
             iter_key, biome_key = jax.random.split(iter_key)
-            # Generate random layout
-            grid_rand = jax.random.uniform(biome_key, (self.size[1], self.size[0]))
-
-            # Create mask for the biome
-            start = jax.lax.select(
-                self.biome_starts[i, 0] == -1,
-                jnp.array([0, 0]),
-                self.biome_starts[i],
-            )
-            stop = jax.lax.select(
-                self.biome_stops[i, 0] == -1,
-                jnp.array(self.size),
-                self.biome_stops[i],
-            )
-
-            rows = jnp.arange(self.size[1])[:, None]
-            cols = jnp.arange(self.size[0])
+            mask = self.biome_masks[i]
 
-            mask = (
-                (rows >= start[1])
-                & (rows < stop[1])
-                & (cols >= start[0])
-                & (cols < stop[0])
-            )
-
-            # Generate objects for this biome and update the main grid
-            biome_freqs = self.biome_object_frequencies[i]
-            empty_freq = 1.0 - jnp.sum(biome_freqs)
-            all_freqs = jnp.concatenate([jnp.array([empty_freq]), biome_freqs])
-
-            cumulative_freqs = jnp.cumsum(
-                jnp.concatenate([jnp.array([0.0]), all_freqs])
-            )
-
-            # Determine which object to place in each cell
-            # The last object ID will be used for any value of grid_rand >= cumulative_freqs[-1]
-            # so we don't need to cap grid_rand
-            obj_ids_for_biome = jnp.arange(len(all_freqs))
-            cell_obj_ids = (
-                jnp.searchsorted(cumulative_freqs, grid_rand, side="right") - 1
-            )
-            biome_objects = obj_ids_for_biome[cell_obj_ids]
-
-            object_grid = jnp.where(mask, biome_objects, object_grid)
+            if self.deterministic_spawn:
+                biome_objects = self.generate_biome_new(i, biome_key)
+                object_grid = object_grid.at[mask].set(biome_objects)
+            else:
+                biome_objects = self.generate_biome_old(i, biome_key)
+                object_grid = jnp.where(mask, biome_objects, object_grid)
 
         # Place agent in the center of the world and ensure the cell is empty.
         agent_pos = jnp.array([self.size[0] // 2, self.size[1] // 2])
@@ -260,6 +249,25 @@ class ForagaxEnv(environment.Environment):
 
         return self.get_obs(state, params), state
 
+    def generate_biome_old(self, i: int, biome_key: jax.Array):
+        biome_freqs = self.biome_object_frequencies[i]
+        grid_rand = jax.random.uniform(biome_key, (self.size[1], self.size[0]))
+        empty_freq = 1.0 - jnp.sum(biome_freqs)
+        all_freqs = jnp.concatenate([jnp.array([empty_freq]), biome_freqs])
+        cumulative_freqs = jnp.cumsum(jnp.concatenate([jnp.array([0.0]), all_freqs]))
+        biome_objects = jnp.searchsorted(cumulative_freqs, grid_rand, side="right") - 1
+        return biome_objects
+
+    def generate_biome_new(self, i: int, biome_key: jax.Array):
+        biome_freqs = self.biome_object_frequencies[i]
+        grid = jnp.linspace(0, 1, self.biome_sizes[i], endpoint=False)
+        biome_objects = len(biome_freqs) - jnp.searchsorted(
+            jnp.cumsum(biome_freqs[::-1]), grid, side="right"
+        )
+        flat_biome_objects = biome_objects.flatten()
+        shuffled_objects = jax.random.permutation(biome_key, flat_biome_objects)
+        return shuffled_objects
+
     def is_terminal(self, state: EnvState, params: EnvParams) -> jax.Array:
         """Foragax is a continuing environment."""
         return False
@@ -267,7 +275,7 @@ class ForagaxEnv(environment.Environment):
     @property
     def name(self) -> str:
         """Environment name."""
-        return "Foragax-v0"
+        return self._name
 
     @property
     def num_actions(self) -> int:
@@ -438,13 +446,17 @@ class ForagaxObjectEnv(ForagaxEnv):
 
     def __init__(
         self,
+        name: str = "Foragax-v0",
         size: Union[Tuple[int, int], int] = (10, 10),
         aperture_size: Union[Tuple[int, int], int] = (5, 5),
         objects: Tuple[BaseForagaxObject, ...] = (),
         biomes: Tuple[Biome, ...] = (Biome(object_frequencies=()),),
         nowrap: bool = False,
+        deterministic_spawn: bool = False,
     ):
-        super().__init__(size, aperture_size, objects, biomes, nowrap)
+        super().__init__(
+            name, size, aperture_size, objects, biomes, nowrap, deterministic_spawn
+        )
 
         # Compute unique colors and mapping for partial observability
         # Exclude EMPTY (index 0) from color channels

foragax/objects.py

@@ -242,6 +242,34 @@ GREEN_FAKE_2 = NormalRegenForagaxObject(
     mean_regen_delay=10,
     std_regen_delay=1,
 )
+BROWN_MOREL_UNIFORM = DefaultForagaxObject(
+    name="brown_morel",
+    reward=10.0,
+    collectable=True,
+    color=(63, 30, 25),
+    regen_delay=(90, 110),
+)
+BROWN_OYSTER_UNIFORM = DefaultForagaxObject(
+    name="brown_oyster",
+    reward=1.0,
+    collectable=True,
+    color=(63, 30, 25),
+    regen_delay=(9, 11),
+)
+GREEN_DEATHCAP_UNIFORM = DefaultForagaxObject(
+    name="green_deathcap",
+    reward=-5.0,
+    collectable=True,
+    color=(0, 255, 0),
+    regen_delay=(9, 11),
+)
+GREEN_FAKE_UNIFORM = DefaultForagaxObject(
+    name="green_fake",
+    reward=0.0,
+    collectable=True,
+    color=(0, 255, 0),
+    regen_delay=(9, 11),
+)
 
 
 def create_weather_objects(

foragax/registry.py

@@ -12,12 +12,16 @@ from foragax.env import (
 from foragax.objects import (
     BROWN_MOREL,
     BROWN_MOREL_2,
+    BROWN_MOREL_UNIFORM,
     BROWN_OYSTER,
+    BROWN_OYSTER_UNIFORM,
     GREEN_DEATHCAP,
     GREEN_DEATHCAP_2,
     GREEN_DEATHCAP_3,
+    GREEN_DEATHCAP_UNIFORM,
     GREEN_FAKE,
     GREEN_FAKE_2,
+    GREEN_FAKE_UNIFORM,
     LARGE_MOREL,
     LARGE_OYSTER,
     MEDIUM_MOREL,
@@ -147,6 +151,31 @@ ENV_CONFIGS: Dict[str, Dict[str, Any]] = {
         "biomes": None,
         "nowrap": True,
     },
+    "ForagaxTwoBiome-v8": {
+        "size": None,
+        "aperture_size": None,
+        "objects": (
+            BROWN_MOREL_UNIFORM,
+            BROWN_OYSTER_UNIFORM,
+            GREEN_DEATHCAP_UNIFORM,
+            GREEN_FAKE_UNIFORM,
+        ),
+        "biomes": None,
+        "nowrap": True,
+    },
+    "ForagaxTwoBiome-v9": {
+        "size": None,
+        "aperture_size": None,
+        "objects": (
+            BROWN_MOREL_UNIFORM,
+            BROWN_OYSTER_UNIFORM,
+            GREEN_DEATHCAP_UNIFORM,
+            GREEN_FAKE_UNIFORM,
+        ),
+        "biomes": None,
+        "nowrap": True,
+        "deterministic_spawn": True,
+    },
     "ForagaxTwoBiomeSmall-v1": {
         "size": (16, 8),
         "aperture_size": None,
@@ -217,7 +246,7 @@ def make(
     if nowrap is not None:
         config["nowrap"] = nowrap
 
-    if env_id == "ForagaxTwoBiome-v7":
+    if env_id in ("ForagaxTwoBiome-v7", "ForagaxTwoBiome-v8", "ForagaxTwoBiome-v9"):
         margin = aperture_size[1] // 2 + 1
         width = 2 * margin + 9
         config["size"] = (width, 15)
@@ -270,5 +299,6 @@ def make(
         raise ValueError(f"Unknown observation type: {observation_type}")
 
     env_class = env_class_map[observation_type]
+    config["name"] = env_id
 
     return env_class(**config)