Merge pull request #41 from rdnfn/dev/general

v0.4.1

Author: rdnfn

.gitignore

@@ -117,5 +117,5 @@ wandb/
 docs/generated/
 
 #beobench
-beobench_results/
+beobench_results*
 notebooks/archive

.pre-commit-config.yaml

@@ -1,6 +1,6 @@
 repos:
   - repo: https://github.com/psf/black
-    rev: 21.9b0
+    rev: 22.3.0
     hooks:
       - id: black
         language_version: python3.9

HISTORY.rst

@@ -2,6 +2,14 @@
 History
 =======
 
+0.4.1 (2022-03-30)
+------------------
+
+* Feature: enable package extras to be given in development mode
+* Feature: add support for arm64/aarch64-based development by forcing experiment containers to run as amd64 containers on those systems (#32)
+* Fix: add gym to extended package requirements
+
+
 0.4.0 (2022-03-28)
 ------------------
 

MANIFEST.in

@@ -6,9 +6,10 @@ include README.rst
 include PYPI_README.rst
 
 recursive-include tests *
-recursive-exclude * __pycache__
-recursive-exclude * *.py[co]
 
 recursive-include beobench/experiment/definitions *
+recursive-exclude beobench/experiment/definitions/archive/ *
 
 recursive-exclude docs *.rst conf.py Makefile make.bat *.jpg *.png *.gif
+recursive-exclude * __pycache__
+recursive-exclude * *.py[co]

beobench/__init__.py

@@ -2,7 +2,7 @@
 
 __author__ = """rdnfn"""
 __email__ = "-"
-__version__ = "0.4.0"
+__version__ = "0.4.1"
 
 from beobench.utils import restart
 from beobench.experiment.scheduler import run

beobench/cli.py

@@ -44,7 +44,7 @@ def cli():
 )
 @click.option(
     "--local-dir",
-    default="./beobench_results/ray_results",
+    default="./beobench_results",
     help="Local directory to write results to.",
     type=click.Path(exists=False, file_okay=False, dir_okay=True),
 )

beobench/data/configs/rewex01.yaml

@@ -2,23 +2,38 @@
 # Run with the command
 # beobench run -c beobench/experiment/definitions/rewex01.yaml -d . --use-gpu --docker-shm-size 28gb
 
+# Some of the descriptions of RLlib config values are taken from
+# https://docs.ray.io/en/latest/rllib/rllib-training.html
+
 # agent config
 agent:
   origin: rllib # either path to agent script or name of agent library (rllib)
   config: # given to ray.tune.run() as arguments (since rllib set before)
     run_or_experiment: PPO
     stop:
-      timesteps_total: 400000
+      timesteps_total: 35040
     config:
-      lr: 0.005
+      lr: 0.0005
       model:
         fcnet_activation: relu
         fcnet_hiddens: [256,256,256,256]
         post_fcnet_activation: tanh
       batch_mode: complete_episodes
       gamma: 0.999
-      horizon: 1000
-      metrics_smoothing_episodes: 5
+      # Number of steps after which the episode is forced to terminate. Defaults
+      # to `env.spec.max_episode_steps` (if present) for Gym envs.
+      horizon: 96
+      # Calculate rewards but don't reset the environment when the horizon is
+      # hit. This allows value estimation and RNN state to span across logical
+      # episodes denoted by horizon. This only has an effect if horizon != inf.
+      soft_horizon: True
+      # Number of timesteps collected for each SGD round. This defines the size
+      # of each SGD epoch.
+      train_batch_size: 94 # single day of 15min steps
+      # Total SGD batch size across all devices for SGD. This defines the
+      # minibatch size within each epoch.
+      sgd_minibatch_size: 24
+      metrics_smoothing_episodes: 1
       framework: torch
       log_level: "WARNING"
       num_workers: 1 # this is required for energym to work (can fail silently otherwise)

beobench/data/configs/rewex01_test01.yaml

@@ -1,6 +1,6 @@
 # REWEX Experiment 01
 # Run with the command
-# beobench run -c beobench/data/configs/rewex01_test01.yaml -d . --use-gpu --docker-shm-size 28gb
+# beobench run -c beobench/data/configs/rewex01_test01.yaml -d .[extended] --use-gpu --docker-shm-size 28gb
 
 # agent config
 agent:

beobench/data/configs/rewex01_test02.yaml

@@ -0,0 +1,19 @@
+# REWEX Experiment 02
+# Run with the command
+# beobench run -c beobench/data/configs/rewex01_test02.yaml -d .[extended] --use-gpu --docker-shm-size 28gb
+
+# agent config
+agent:
+  origin: beobench/data/agents/random_agent.py # either path to agent script or name of agent library (rllib)
+# environment config
+env:
+  name: MixedUseFanFCU-v0
+  gym: energym
+  config:
+    days: 365
+    energym_environment: MixedUseFanFCU-v0
+    gym_kwargs:
+      max_episode_length: 35040
+      normalize: true
+      step_period: 15
+    weather: GRC_A_Athens

beobench/experiment/containers.py

@@ -2,6 +2,8 @@
 
 import subprocess
 import os
+import pathlib
+import shutil
 
 # To enable compatiblity with Python<=3.6 (e.g. for sinergym dockerfile)
 try:
@@ -16,8 +18,9 @@
 def build_experiment_container(
     build_context: str,
     use_no_cache: bool = False,
-    version="latest",
-    enable_rllib=False,
+    local_dir: pathlib.Path = None,
+    version: str = "latest",
+    enable_rllib: bool = False,
 ) -> None:
     """Build experiment container from beobench/integrations/boptest/Dockerfile.
 
@@ -27,26 +30,49 @@ def build_experiment_container(
             of existing beobench integration (e.g. `boptest`). See the official docs
             https://docs.docker.com/engine/reference/commandline/build/ for more info.
         use_no_cache (bool, optional): wether to use cache in build. Defaults to False.
+        version (str, optional): version to add to container tag. Defaults to "latest".
+        enable_rllib (bool, optional): whether to install rllib. Defaults to False.
     """
 
     # Flags are shared between gym image build and gym_and_beobench image build
     flags = []
+
+    # Using buildx to enable platform-specific builds
+    build_commands = ["docker", "buildx", "build"]
+
+    # On arm64 machines force experiment containers to be amd64
+    # This is only useful for development purposes.
+    # (example: M1 macbooks)
+    if os.uname().machine in ["arm64", "aarch64"]:
+        flags += ["--platform", "linux/amd64"]
+
     if use_no_cache:
         flags.append("--no-cache")
 
+    # pylint: disable=invalid-name
     AVAILABLE_INTEGRATIONS = [
         "boptest",
         "sinergym",
         "energym",
-    ]  # pylint: disable=invalid-name
+    ]
 
     if build_context in AVAILABLE_INTEGRATIONS:
         image_name = f"beobench_{build_context}"
         integration_name = build_context
-        build_context = (
-            f"https://github.com/rdnfn/"
-            f"beobench_contrib.git#main:gyms/{build_context}"
+
+        # TODO: remove tmp git dir once buildkit version in docker cli updated
+        tmp_git_dir = (local_dir / "tmp" / "beobench_contrib").absolute()
+        subprocess.check_call(
+            [
+                "git",
+                "clone",
+                "https://github.com/rdnfn/beobench_contrib.git",
+                tmp_git_dir,
+            ]
         )
+
+        build_context = str(tmp_git_dir / "gyms" / integration_name)
+
         print(
             (
                 f"Recognised integration named {integration_name}: using build"
@@ -64,17 +90,14 @@ def build_experiment_container(
 
     # Part 1: build base experiment image
     args = [
-        "docker",
-        "build",
+        *build_commands,
         "-t",
         base_image_tag,
-        "-f",
-        "Dockerfile",  # change to non-default name
         *flags,
         build_context,
     ]
     env = os.environ.copy()
-    env["DOCKER_BUILDKIT"] = "0"
+    print("Running command: " + " ".join(args))
     subprocess.check_call(
         args,
         env=env,  # this enables accessing dockerfile in subdir
@@ -92,14 +115,13 @@ def build_experiment_container(
     # Which extras to install beobench container
     # e.g. using pip install beobench[extras]
     if enable_rllib:
-        beobench_extras = '"extended,rllib"'
+        beobench_extras = "extended,rllib"
     else:
         beobench_extras = "extended"
     # Load dockerfile into pipe
     with subprocess.Popen(["cat", complete_dockerfile], stdout=subprocess.PIPE) as proc:
         beobench_build_args = [
-            "docker",
-            "build",
+            *build_commands,
             "-t",
             complete_image_tag,
             "-f",
@@ -118,6 +140,8 @@ def build_experiment_container(
             env=env,  # this enables accessing dockerfile in subdir
         )
 
+    # TODO: remove tmp git dir once buildkit version in docker cli updated
+    shutil.rmtree(tmp_git_dir)
     print("Experiment gym image build finished.")
 
     return complete_image_tag