40 finish first version of gym interface

.github/workflows/ci.yml

@@ -0,0 +1,90 @@
+name: ci
+
+on:
+  push:
+    branches:
+      - master
+  pull_request:
+    branches:
+      - master
+
+# https://jacobian.org/til/github-actions-poetry/
+jobs:
+  flake8-lint:
+    runs-on: ubuntu-latest
+    name: Lint
+    steps:
+      - name: Check out source repository
+        uses: actions/checkout@v2
+      - name: Set up Python environment
+        uses: actions/setup-python@v1
+        with:
+          python-version: "3.9"
+      - name: flake8 Lint
+        uses: py-actions/flake8@v1
+        with:
+          ignore: "E203,E402,E501,F401,F841"
+          exclude: "logs/*,data/*,furuta/logging/protobuf/*"
+
+  test:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v2
+
+      # If you wanted to use multiple Python versions, you'd have specify a matrix in the job and
+      # reference the matrixe python version here.
+      - uses: actions/setup-python@v2
+        with:
+          python-version: 3.9
+
+      # Cache the installation of Poetry itself, e.g. the next step. This prevents the workflow
+      # from installing Poetry every time, which can be slow. Note the use of the Poetry version
+      # number in the cache key, and the "-0" suffix: this allows you to invalidate the cache
+      # manually if/when you want to upgrade Poetry, or if something goes wrong. This could be
+      # mildly cleaner by using an environment variable, but I don't really care.
+      - name: cache poetry install
+        uses: actions/cache@v2
+        with:
+          path: ~/.local
+          key: poetry-1.4.0-0
+
+      # Install Poetry. You could do this manually, or there are several actions that do this.
+      # `snok/install-poetry` seems to be minimal yet complete, and really just calls out to
+      # Poetry's default install script, which feels correct. I pin the Poetry version here
+      # because Poetry does occasionally change APIs between versions and I don't want my
+      # actions to break if it does.
+      #
+      # The key configuration value here is `virtualenvs-in-project: true`: this creates the
+      # venv as a `.venv` in your testing directory, which allows the next step to easily
+      # cache it.
+      - uses: snok/install-poetry@v1
+        with:
+          version: 1.4.0
+          virtualenvs-create: true
+          virtualenvs-in-project: true
+
+      # Cache your dependencies (i.e. all the stuff in your `pyproject.toml`). Note the cache
+      # key: if you're using multiple Python versions, or multiple OSes, you'd need to include
+      # them in the cache key. I'm not, so it can be simple and just depend on the poetry.lock.
+      - name: cache deps
+        id: cache-deps
+        uses: actions/cache@v2
+        with:
+          path: .venv
+          key: pydeps-${{ hashFiles('**/poetry.lock') }}
+
+      # Install dependencies. `--no-root` means "install all dependencies but not the project
+      # itself", which is what you want to avoid caching _your_ code. The `if` statement
+      # ensures this only runs on a cache miss.
+      - run: poetry install --no-interaction --no-root
+        if: steps.cache-deps.outputs.cache-hit != 'true'
+
+      # Now install _your_ project. This isn't necessary for many types of projects -- particularly
+      # things like Django apps don't need this. But it's a good idea since it fully-exercises the
+      # pyproject.toml and makes that if you add things like console-scripts at some point that
+      # they'll be installed and working.
+      - run: poetry install --no-interaction
+
+      # And finally run tests. I'm using pytest and all my pytest config is in my `pyproject.toml`
+      # so this line is super-simple. But it could be as complex as you need.
+      - run: poetry run pytest

.gitignore

@@ -1,4 +1,5 @@
 **/videos/*
+**/outputs/*
 **/wandb/*
 **/runs/*
 **/models/*
@@ -124,7 +125,7 @@ celerybeat.pid
 # Environments
 .env
 .venv
-env/
+# env/ bc we use env for gym env config but have no env/ dir for actual venvs
 venv/
 ENV/
 env.bak/

furuta/__init__.py

@@ -2,9 +2,9 @@
 
 register(
     id="FurutaReal-v0",
-    entry_point="furuta_gym.envs:FurutaReal",
+    entry_point="furuta.rl.envs.furuta_real:FurutaReal",
 )
 register(
     id="FurutaSim-v0",
-    entry_point="furuta_gym.envs:FurutaSim",
+    entry_point="furuta.rl.envs.furuta_sim:FurutaSim",
 )

furuta/controls/controllers.py

@@ -11,11 +11,15 @@ def compute_command(self, position: float):
     @staticmethod
     def build_controller(parameters: dict):
         controller_type = parameters["controller_type"]
-        match controller_type:
-            case "PIDController":
-                return PIDController(parameters)
-            case _:
-                raise ValueError(f"Invalid controller type: {controller_type}")
+        # match controller_type:
+        #     case "PIDController":
+        #         return PIDController(parameters)
+        #     case _:
+        #         raise ValueError(f"Invalid controller type: {controller_type}")
+        if controller_type == "PIDController":
+            return PIDController(parameters)
+        else:
+            raise ValueError(f"Invalid controller type: {controller_type}")
 
 
 class PIDController(Controller):

furuta/logging/protobuf/pendulum_state.proto

@@ -6,7 +6,5 @@ message PendulumState {
   float motor_angle_velocity = 3;
   float pendulum_angle_velocity = 4;
   float reward = 5;
-  bool done = 6;
-  float action = 7;
-  float corrected_action = 8;
+  float action = 6;
 }

furuta/rl/__init__.py

@@ -1,3 +0,0 @@
-# from furuta_gym.envs.furuta_base import FurutaBase   # noqa F420
-# from furuta_gym.envs.furuta_real import FurutaReal  # noqa F420
-# from furuta_gym.envs.furuta_sim import FurutaSim   # noqa F420

furuta/rl/algos.py

@@ -6,11 +6,11 @@
 # check if they all have the train freq param
 # check if they have other tuple args
 # check if it would be cleaner for sb3 to accept list instead of tuple?
-# TODO also does having the sb3 import means i need sb3 when importing from everywhere else in the package?
 class SAC(stable_baselines3.SAC):
     def __init__(self, **kwargs):
         # sb3 expects tuple, omegaconf returns list
         # so we need to convert kwarg train_freq from tuple to list
-        kwargs.update({"train_freq": tuple(kwargs["train_freq"])})
+        if "train_freq" in kwargs and isinstance(kwargs["train_freq"], list):
+            kwargs.update({"train_freq": tuple(kwargs["train_freq"])})
 
         super().__init__(**kwargs)