Update Melting Pot wrapper to use pypi release (#111)

Co-authored-by: Elliot Tower <etower@farama.org>
Co-authored-by: Mark Towers <mark.m.towers@gmail.com>

Author: 3 people

.github/workflows/optional-test-meltingpot.yml

@@ -13,7 +13,7 @@ jobs:
       # Melting Pot
       - run: |
           docker build -f bin/meltingpot.Dockerfile \
-            --build-arg PYTHON_VERSION='3.9' \
+            --build-arg PYTHON_VERSION='3.10' \
             --tag shimmy-meltingpot-docker .
       - name: Run meltingpot tests
         run: docker run shimmy-meltingpot-docker pytest tests/test_meltingpot.py

bin/meltingpot.Dockerfile

@@ -1,9 +1,9 @@
 # A Dockerfile that sets up a full shimmy install with test dependencies
 # adapted from https://github.com/deepmind/meltingpot/blob/main/.devcontainer/Dockerfile
 
-# if PYTHON_VERSION is not specified as a build argument, set it to 3.9.
+# if PYTHON_VERSION is not specified as a build argument, set it to 3.10.
 ARG PYTHON_VERSION
-ARG PYTHON_VERSION=${PYTHON_VERSION:-3.9}
+ARG PYTHON_VERSION=${PYTHON_VERSION:-3.10}
 FROM python:$PYTHON_VERSION
 
 SHELL ["/bin/bash", "-o", "pipefail", "-c"]
@@ -38,40 +38,4 @@ RUN if [ -f "pyproject.toml" ]; then \
         mkdir -p bin && mv docker_entrypoint bin/docker_entrypoint; \
     fi
 
-# Install Melting Pot dependencies
-RUN apt-get update \
-    && DEBIAN_FRONTEND=noninteractive apt-get -qq -y install \
-    build-essential \
-    curl \
-    ffmpeg \
-    git
-
-# Install lab2d (appropriate version for architecture)
-RUN if [ "$(uname -m)" != 'x86_64' ]; then \
-        echo "No Lab2d wheel available for $(uname -m) machines." >&2 \
-        exit 1; \
-    elif [ "$(uname -s)" = 'Linux' ]; then \
-        pip install https://github.com/deepmind/lab2d/releases/download/release_candidate_2022-03-24/dmlab2d-1.0-cp39-cp39-manylinux_2_31_x86_64.whl ;\
-    else \
-        pip install https://github.com/deepmind/lab2d/releases/download/release_candidate_2022-03-24/dmlab2d-1.0-cp39-cp39-macosx_10_15_x86_64.whl ;\
-    fi
-
-# Download Melting Pot assets
-RUN mkdir -p /workspaces/meltingpot/meltingpot \
-    && curl -SL https://storage.googleapis.com/dm-meltingpot/meltingpot-assets-2.1.0.tar.gz \
-    | tar -xz --directory=/workspaces/meltingpot/meltingpot
-
-# Clone Melting Pot repository and install dependencies
-RUN git clone https://github.com/deepmind/meltingpot.git
-RUN cp -r meltingpot/ /workspaces/meltingpot/ && rm -R meltingpot/
-
-# Checkout the last commit with 3.9 support that passed CI (July 17 2023)
-# Newer versions and pypi wheels caused issues
-RUN cd /workspaces/meltingpot/meltingpot && git checkout ed2e6e79ca49a14a22aa4b6117ac407f39fbef81
-
-RUN pip install -e /workspaces/meltingpot/meltingpot
-
-# Set Python path for meltingpot
-ENV PYTHONPATH "${PYTHONPATH}:/workspaces/meltingpot/meltingpot/"
-
 ENTRYPOINT ["/usr/local/shimmy/bin/docker_entrypoint"]

docs/environments/meltingpot.md

@@ -25,12 +25,12 @@ Shimmy provides compatibility wrappers to convert all [Melting Pot](https://gith
 To install `shimmy` and required dependencies:
 
 ```
-pip install shimmy[melting-pot]
+pip install shimmy[meltingpot]
 ```
 
-Melting Pot is not distributed via [pypi](https://pypi.org/) and must be installed manually. We provide an [installation script](https://github.com/Farama-Foundation/Shimmy/blob/main/scripts/install_melting_pot.sh) (compatible with macOS and linux). For troubleshooting,  refer to the official [installation instructions](https://github.com/deepmind/meltingpot#installation).
+We also provide a [Dockerfile](https://github.com/Farama-Foundation/Shimmy/blob/main/bin/meltingpot.Dockerfile) for reproducibility and cross-platform compatibility (see
 
-We also provide a [Dockerfile](https://github.com/Farama-Foundation/Shimmy/blob/main/bin/meltingpot.Dockerfile) for reproducibility and cross-platform compatibility (see [Installation](/content/getting_started/))
+[Installation](/content/getting_started/))
 
 ```
 curl https://raw.githubusercontent.com/Farama-Foundation/Shimmy/main/bin/meltingpot.Dockerfile | docker build -t meltingpot -f - . && docker run -it meltingpot

scripts/install_melting_pot.sh

@@ -46,7 +46,7 @@ def get_version():
     ],
     "dm-lab": ["dm-env>=1.6"],
     "openspiel": ["open_spiel>=1.2", "pettingzoo>=1.23"],
-    "meltingpot": ["pettingzoo>=1.23"],
+    "meltingpot": ["pettingzoo>=1.23", "dm-meltingpot>=2.2.0; python_version > '3.9'"],
     "bsuite": ["bsuite>=0.3.5"],
 }
 extras["all"] = [

setup.py

@@ -2,7 +2,6 @@
 from __future__ import annotations
 
 import functools
-from itertools import repeat
 from typing import TYPE_CHECKING, Any
 
 import dm_control.composer
@@ -43,27 +42,30 @@ def _unravel_ma_timestep(
     observations: dict[AgentID, Any] = dict(zip(agents, list_observations))
 
     # sometimes deepmind decides not to reward people
-    rewards: dict[AgentID, float] = dict(zip(agents, repeat(0.0)))
+    rewards: dict[AgentID, float] = {agent: 0.0 for agent in agents}
+
     if timestep.reward:
         rewards = dict(zip(agents, timestep.reward))
 
     # expand everything else
-    terminations: dict[AgentID, bool] = dict(zip(agents, repeat(term)))
-    truncations: dict[AgentID, bool] = dict(zip(agents, repeat(trunc)))
-
-    # duplicate infos
-    info = {
-        "timestep.discount": timestep.discount,
-        "timestep.step_type": timestep.step_type,
+    terminations: dict[AgentID, bool] = {agent: term for agent in agents}
+    truncations: dict[AgentID, bool] = {agent: trunc for agent in agents}
+
+    # duplicate infos across agents
+    infos = {
+        agent: {
+            "timestep.discount": timestep.discount,
+            "timestep.step_type": timestep.step_type,
+        }
+        for agent in agents
     }
-    info: dict[AgentID, Any] = dict(zip(agents, repeat(info)))
 
     return (
         observations,
         rewards,
         terminations,
         truncations,
-        info,
+        infos,
     )
 
 

shimmy/dm_control_multiagent_compatibility.py

@@ -8,6 +8,7 @@
 from __future__ import annotations
 
 import functools
+from itertools import repeat
 from typing import TYPE_CHECKING, Any, Optional
 
 import dm_env
@@ -20,7 +21,8 @@
 import shimmy.utils.meltingpot as utils
 
 if TYPE_CHECKING:
-    import meltingpot.python
+    import meltingpot
+    from meltingpot.utils.substrates import substrate
 
 
 class MeltingPotCompatibilityV0(ParallelEnv, EzPickle):
@@ -46,15 +48,15 @@ class MeltingPotCompatibilityV0(ParallelEnv, EzPickle):
 
     def __init__(
         self,
-        env: meltingpot.python.utils.substrates.substrate.Substrate | None = None,
+        env: substrate.Substrate | None = None,
         substrate_name: str | None = None,
         max_cycles: int = MAX_CYCLES,
         render_mode: str | None = None,
     ):
         """Wrapper that converts a Melting Pot environment into a PettingZoo environment.
 
         Args:
-            env (Optional[meltingpot.python.utils.substrates.substrate.Substrate]): existing Melting Pot environment to wrap
+            env (Optional[substrate.Substrate]): existing Melting Pot environment to wrap
             substrate_name (Optional[str]): name of Melting Pot substrate to load (instead of existing environment)
             max_cycles (Optional[int]): maximum number of cycles before truncation
             render_mode (Optional[str]): rendering mode
@@ -158,7 +160,7 @@ def reset(
         self,
         seed: int | None = None,
         options: dict | None = None,
-    ) -> tuple[ObsDict, dict[str, Any]]:
+    ) -> tuple[ObsDict, dict[AgentID, Any]]:
         """reset.
 
         Resets the environment.
@@ -176,11 +178,17 @@ def reset(
 
         observations = utils.timestep_to_observations(timestep)
 
-        return observations, {
-            "step-type": timestep.step_type,
-            "discount": timestep.discount,
+        # duplicate infos across agents
+        infos = {
+            agent: {
+                "timestep.discount": timestep.discount,
+                "timestep.step_type": timestep.step_type,
+            }
+            for agent in self.agents
         }
 
+        return observations, infos
+
     def step(
         self, actions: ActionDict
     ) -> tuple[

shimmy/meltingpot_compatibility.py

@@ -18,26 +18,22 @@ def load_meltingpot(substrate_name: str):
         substrate_name: str
 
     Returns:
-        env: meltingpot.python.utils.substrates.substrate.Substrate
+        env: meltingpot.utils.substrates.substrate.Substrate
     """
-    import meltingpot.python
+    import meltingpot
     from ml_collections import config_dict
 
     # Create env config
     substrate_name = substrate_name
-    player_roles = meltingpot.python.substrate.get_config(
-        substrate_name
-    ).default_player_roles
+    player_roles = meltingpot.substrate.get_config(substrate_name).default_player_roles
     env_config = {
         "substrate": substrate_name,
         "roles": player_roles,
     }
 
     # Build substrate from pickle
     env_config = config_dict.ConfigDict(env_config)
-    env = meltingpot.python.substrate.build(
-        env_config["substrate"], roles=env_config["roles"]
-    )
+    env = meltingpot.substrate.build(env_config["substrate"], roles=env_config["roles"])
     return env
 
 

shimmy/utils/meltingpot.py

@@ -10,8 +10,8 @@
 
 pytest.importorskip("meltingpot")
 
-import meltingpot.python
-from meltingpot.python.configs.substrates import SUBSTRATES
+import meltingpot
+from meltingpot.substrate import SUBSTRATES
 
 from shimmy.meltingpot_compatibility import MeltingPotCompatibilityV0
 from shimmy.utils.meltingpot import load_meltingpot