gryf/coach
1
0
Fork 0
mirror of https://github.com/gryf/coach.git synced 2025-12-17 19:20:19 +01:00
Code Issues Projects Releases Wiki Activity

introduce dockerfiles. (#169)

Browse Source 
* introduce dockerfiles.

* ensure golden tests are run not just collected.

* Skip CI download of dockerfiles.

* add StarCraft environment and tests.

* add minimaps starcraft validation parameters.

* Add functional test running (from Ayoob)

* pin mujoco_py version to a 1.5 compatible release.

* fix config syntax issue.

* pin remaining mujoco_py install calls.

* Relax pin of gym version in gym Dockerfile.

* update makefile based on functional test filtering.
This commit is contained in:
Scott Leishman
2019-04-03 09:33:17 -07:00
committed by anabwan
parent 0b808f0794
commit f173e69187
11 changed files with 459 additions and 33 deletions
Download Patch File Download Diff File Expand all files Collapse all files

116
.circleci/config.yml
View File

@@ -30,11 +30,6 @@ aliases:
aws eks update-kubeconfig --name coach-aws-cicd aws eks update-kubeconfig --name coach-aws-cicd
sudo curl -o /usr/local/bin/kubectl https://storage.googleapis.com/kubernetes-release/release/$(curl -s https://storage.googleapis.com/kubernetes-release/release/stable.txt)/bin/linux/amd64/kubectl sudo curl -o /usr/local/bin/kubectl https://storage.googleapis.com/kubernetes-release/release/$(curl -s https://storage.googleapis.com/kubernetes-release/release/stable.txt)/bin/linux/amd64/kubectl
sudo chmod a+x /usr/local/bin/kubectl sudo chmod a+x /usr/local/bin/kubectl
- &docker_prep
run:
name: Prepare dockerfiles
command: |
aws s3 cp --recursive s3://coach-ci-dockerfiles docker
version: 2 version: 2
jobs: jobs:
@@ -45,7 +40,6 @@ jobs:
- *remote_docker - *remote_docker
- *restore_cache - *restore_cache
- *aws_prep - *aws_prep
- *docker_prep
- run: - run:
name: Build and push base and main container name: Build and push base and main container
command: | command: |
@@ -76,7 +70,6 @@ jobs:
- *remote_docker - *remote_docker
- *restore_cache - *restore_cache
- *aws_prep - *aws_prep
- *docker_prep
- run: - run:
name: Build and push gym environment container name: Build and push gym environment container
command: | command: |
@@ -95,7 +88,6 @@ jobs:
- *remote_docker - *remote_docker
- *restore_cache - *restore_cache
- *aws_prep - *aws_prep
- *docker_prep
- run: - run:
name: Build and push vizdoom environment container name: Build and push vizdoom environment container
command: | command: |
@@ -114,7 +106,6 @@ jobs:
- *remote_docker - *remote_docker
- *restore_cache - *restore_cache
- *aws_prep - *aws_prep
- *docker_prep
- run: - run:
name: Build and push mujoco environment container name: Build and push mujoco environment container
command: | command: |
@@ -126,6 +117,24 @@ jobs:
docker push ${REGISTRY}/coach-mujoco_environment:${TAG} docker push ${REGISTRY}/coach-mujoco_environment:${TAG}
no_output_timeout: 10m no_output_timeout: 10m
build_starcraft_env:
<<: *executor_prep
steps:
- checkout
- *remote_docker
- *restore_cache
- *aws_prep
- run:
name: Build and push starcraft environment container
command: |
REGISTRY=316971102342.dkr.ecr.us-west-2.amazonaws.com
TAG=$(git describe --tags --always --dirty)
docker pull ${REGISTRY}/coach-base:${MASTER_BRANCH}
docker tag ${REGISTRY}/coach-base:${MASTER_BRANCH} coach-base:master
docker build --cache-from ${REGISTRY}/coach-base:${MASTER_BRANCH} -t ${REGISTRY}/coach-starcraft_environment:${TAG} -f docker/Dockerfile.starcraft_environment .
docker push ${REGISTRY}/coach-starcraft_environment:${TAG}
no_output_timeout: 10m
unit_tests: unit_tests:
<<: *executor_prep <<: *executor_prep
steps: steps:
@@ -133,7 +142,6 @@ jobs:
- *remote_docker - *remote_docker
- *restore_cache - *restore_cache
- *aws_prep - *aws_prep
- *docker_prep
- run: - run:
name: run unit tests name: run unit tests
command: | command: |
@@ -152,7 +160,6 @@ jobs:
- *remote_docker - *remote_docker
- *restore_cache - *restore_cache
- *aws_prep - *aws_prep
- *docker_prep
- run: - run:
name: run integration tests name: run integration tests
command: | command: |
@@ -172,7 +179,6 @@ jobs:
- *remote_docker - *remote_docker
- *restore_cache - *restore_cache
- *aws_prep - *aws_prep
- *docker_prep
- run: - run:
name: run functional tests name: run functional tests
command: | command: |
@@ -193,7 +199,6 @@ jobs:
- *remote_docker - *remote_docker
- *restore_cache - *restore_cache
- *aws_prep - *aws_prep
- *docker_prep
- run: - run:
name: run gym related golden tests name: run gym related golden tests
command: | command: |
@@ -214,7 +219,6 @@ jobs:
- *remote_docker - *remote_docker
- *restore_cache - *restore_cache
- *aws_prep - *aws_prep
- *docker_prep
- run: - run:
name: run doom related golden tests name: run doom related golden tests
command: | command: |
@@ -235,7 +239,6 @@ jobs:
- *remote_docker - *remote_docker
- *restore_cache - *restore_cache
- *aws_prep - *aws_prep
- *docker_prep
- run: - run:
name: run mujoco related golden tests name: run mujoco related golden tests
command: | command: |
@@ -249,6 +252,26 @@ jobs:
kubectl delete ns golden-test-mujoco-${CIRCLE_BUILD_NUM} || true kubectl delete ns golden-test-mujoco-${CIRCLE_BUILD_NUM} || true
when: always when: always
golden_test_starcraft:
<<: *executor_prep
steps:
- checkout
- *remote_docker
- *restore_cache
- *aws_prep
- run:
name: run starcraft related golden tests
command: |
export GOLDEN_PRESETS='Starcraft'
python3 rl_coach/tests/test_eks.py -c coach-test -bn ${CIRCLE_BUILD_NUM} -tn golden-test-starcraft -tc "export GOLDEN_PRESETS=${GOLDEN_PRESETS} && make golden_tests_without_docker" -i 316971102342.dkr.ecr.us-west-2.amazonaws.com/coach-starcraft_environment:$(git describe --tags --always --dirty) -cpu 2048 -mem 4096
no_output_timeout: 30m
- run:
name: cleanup
command: |
kubectl delete --all pods --namespace=golden-test-starcraft-${CIRCLE_BUILD_NUM} || true
kubectl delete ns golden-test-starcraft-${CIRCLE_BUILD_NUM} || true
when: always
functional_test_mujoco: functional_test_mujoco:
<<: *executor_prep <<: *executor_prep
steps: steps:
@@ -256,7 +279,6 @@ jobs:
- *remote_docker - *remote_docker
- *restore_cache - *restore_cache
- *aws_prep - *aws_prep
- *docker_prep
- run: - run:
name: run mujoco related functional tests name: run mujoco related functional tests
command: | command: |
@@ -277,7 +299,6 @@ jobs:
- *remote_docker - *remote_docker
- *restore_cache - *restore_cache
- *aws_prep - *aws_prep
- *docker_prep
- run: - run:
name: run doom related functional tests name: run doom related functional tests
command: | command: |
@@ -298,7 +319,6 @@ jobs:
- *remote_docker - *remote_docker
- *restore_cache - *restore_cache
- *aws_prep - *aws_prep
- *docker_prep
- run: - run:
name: run Starcraft related functional tests name: run Starcraft related functional tests
command: | command: |
@@ -319,7 +339,6 @@ jobs:
- *remote_docker - *remote_docker
- *restore_cache - *restore_cache
- *aws_prep - *aws_prep
- *docker_prep
- run: - run:
name: run gym related trace tests name: run gym related trace tests
command: | command: |
@@ -340,7 +359,6 @@ jobs:
- *remote_docker - *remote_docker
- *restore_cache - *restore_cache
- *aws_prep - *aws_prep
- *docker_prep
- run: - run:
name: run doom related trace tests name: run doom related trace tests
command: | command: |
@@ -361,7 +379,6 @@ jobs:
- *remote_docker - *remote_docker
- *restore_cache - *restore_cache
- *aws_prep - *aws_prep
- *docker_prep
- run: - run:
name: run mujoco related trace tests name: run mujoco related trace tests
command: | command: |
@@ -375,6 +392,26 @@ jobs:
kubectl delete ns trace-test-mujoco-${CIRCLE_BUILD_NUM} || true kubectl delete ns trace-test-mujoco-${CIRCLE_BUILD_NUM} || true
when: always when: always
trace_test_starcraft:
<<: *executor_prep
steps:
- checkout
- *remote_docker
- *restore_cache
- *aws_prep
- run:
name: run starcraft related trace tests
command: |
export TRACE_PRESETS='Starcraft_CollectMinerals_Arc,Starcraft_CollectMinerals_Deuling_DDQN'
python3 rl_coach/tests/test_eks.py -c coach-test -bn ${CIRCLE_BUILD_NUM} -tn trace-test-starcraft -tc "export TRACE_PRESETS=${TRACE_PRESETS} && make trace_tests_without_docker" -i 316971102342.dkr.ecr.us-west-2.amazonaws.com/coach-starcraft_environment:$(git describe --tags --always --dirty) -cpu 2048 -mem 4096
no_output_timeout: 30m
- run:
name: cleanup
command: |
kubectl delete --all pods --namespace=trace-test-starcraft-${CIRCLE_BUILD_NUM} || true
kubectl delete ns trace-test-starcraft-${CIRCLE_BUILD_NUM} || true
when: always
multinode_test: multinode_test:
<<: *executor_prep <<: *executor_prep
steps: steps:
@@ -382,7 +419,6 @@ jobs:
- *remote_docker - *remote_docker
- *restore_cache - *restore_cache
- *aws_prep - *aws_prep
- *docker_prep
- run: - run:
name: run multinode test name: run multinode test
command: | command: |
@@ -417,7 +453,6 @@ jobs:
- *remote_docker - *remote_docker
- *restore_cache - *restore_cache
- *aws_prep - *aws_prep
- *docker_prep
- run: - run:
name: Tag and push updated base and main container name: Tag and push updated base and main container
command: | command: |
@@ -468,9 +503,9 @@ workflows:
- build_mujoco_env: - build_mujoco_env:
requires: requires:
- e2e_approval - e2e_approval
# - build_starcraft_env: - build_starcraft_env:
# requires: requires:
# - e2e_approval - e2e_approval
- gym_approval: - gym_approval:
type: approval type: approval
requires: requires:
@@ -483,6 +518,10 @@ workflows:
type: approval type: approval
requires: requires:
- golden_test_mujoco - golden_test_mujoco
- starcraft_approval:
type: approval
requires:
- golden_test_starcraft
- golden_test_gym: - golden_test_gym:
requires: requires:
- build_gym_env - build_gym_env
@@ -492,6 +531,9 @@ workflows:
- golden_test_mujoco: - golden_test_mujoco:
requires: requires:
- build_mujoco_env - build_mujoco_env
- golden_test_starcraft:
requires:
- build_starcraft_env
- trace_test_gym: - trace_test_gym:
requires: requires:
- gym_approval - gym_approval
@@ -501,6 +543,9 @@ workflows:
- trace_test_mujoco: - trace_test_mujoco:
requires: requires:
- mujoco_approval - mujoco_approval
- trace_test_starcraft:
requires:
- starcraft_approval
- container_deploy: - container_deploy:
requires: requires:
- unit_tests - unit_tests
@@ -523,15 +568,18 @@ workflows:
- build_gym_env - build_gym_env
- build_doom_env - build_doom_env
- build_mujoco_env - build_mujoco_env
# - functional_test_starcraft: - build_starcraft_env
# requires:
# - build_starcraft_env
- functional_test_doom: - functional_test_doom:
requires: requires:
- build_doom_env - build_doom_env
- functional_test_mujoco: - functional_test_mujoco:
requires: requires:
- build_mujoco_env - build_mujoco_env
- functional_test_doom
- functional_test_starcraft:
requires:
- build_starcraft_env
- functional_test_mujoco
- golden_test_gym: - golden_test_gym:
requires: requires:
- build_gym_env - build_gym_env
@@ -543,12 +591,22 @@ workflows:
requires: requires:
- build_mujoco_env - build_mujoco_env
- golden_test_doom - golden_test_doom
- golden_test_starcraft:
requires:
- build_starcraft_env
- golden_test_mujoco
# - trace_test_gym: # - trace_test_gym:
# requires: # requires:
# - build_gym_env # - build_gym_env
# - trace_test_doom: # - trace_test_doom:
# requires: # requires:
# - build_doom_env # - build_doom_env
# - trace_test_gym
# - trace_test_mujoco: # - trace_test_mujoco:
# requires: # requires:
# - build_mujoco_env # - build_mujoco_env
# - trace_test_doom
# - trace_test_starcraft:
# requires:
# - build_mujoco_env
# - trace_test_mujoco

33
docker/Dockerfile Normal file
View File

@@ -0,0 +1,33 @@
FROM coach-base:master as builder
# prep some of the more common environments
# Gym (installed with coach)
Run pip3 install gym[atari] box2d
# Mujoco
RUN mkdir -p ~/.mujoco \
&& wget https://www.roboti.us/download/mjpro150_linux.zip -O mujoco.zip \
&& unzip mujoco.zip -d ~/.mujoco \
&& rm mujoco.zip
ARG MUJOCO_KEY
ENV MUJOCO_KEY=$MUJOCO_KEY
ENV LD_LIBRARY_PATH /root/.mujoco/mjpro150/bin:$LD_LIBRARY_PATH
RUN echo $MUJOCO_KEY | base64 --decode > /root/.mujoco/mjkey.txt
RUN pip3 install mujoco_py==1.50.1.68
# Vizdoom
RUN pip3 install vizdoom
RUN mkdir /root/src
COPY setup.py /root/src/.
COPY requirements.txt /root/src/.
RUN pip3 install -r /root/src/requirements.txt
FROM coach-base:master
WORKDIR /root/src
COPY --from=builder /root/.mujoco /root/.mujoco
ENV LD_LIBRARY_PATH /root/.mujoco/mjpro150/bin:$LD_LIBRARY_PATH
COPY --from=builder /root/.cache /root/.cache
COPY setup.py /root/src/.
COPY requirements.txt /root/src/.
COPY README.md /root/src/.
RUN pip3 install gym[atari] box2d mujoco_py==1.50.1.68 vizdoom && pip3 install -e .[all] && rm -rf /root/.cache
COPY . /root/src

39
docker/Dockerfile.base Normal file
View File

@@ -0,0 +1,39 @@
FROM nvidia/cuda:9.0-cudnn7-runtime-ubuntu16.04
################################
# Install apt-get Requirements #
################################
# General
RUN apt-get update && \
apt-get install -y python3-pip cmake zlib1g-dev python3-tk python-opencv \
# Boost libraries
libboost-all-dev \
# Scipy requirements
libblas-dev liblapack-dev libatlas-base-dev gfortran \
# Pygame requirements
libsdl-dev libsdl-image1.2-dev libsdl-mixer1.2-dev libsdl-ttf2.0-dev \
libsmpeg-dev libportmidi-dev libavformat-dev libswscale-dev \
# Dashboard
dpkg-dev build-essential python3.5-dev libjpeg-dev libtiff-dev libsdl1.2-dev libnotify-dev \
freeglut3 freeglut3-dev libsm-dev libgtk2.0-dev libgtk-3-dev libwebkitgtk-dev libgtk-3-dev \
libwebkitgtk-3.0-dev libgstreamer-plugins-base1.0-dev \
# Gym
libav-tools libsdl2-dev swig cmake \
# Mujoco_py
curl libgl1-mesa-dev libgl1-mesa-glx libglew-dev libosmesa6-dev software-properties-common \
# ViZDoom
build-essential zlib1g-dev libsdl2-dev libjpeg-dev \
nasm tar libbz2-dev libgtk2.0-dev cmake git libfluidsynth-dev libgme-dev \
libopenal-dev timidity libwildmidi-dev unzip wget && \
apt-get clean autoclean && \
apt-get autoremove -y
############################
# Install Pip Requirements #
############################
RUN pip3 install --upgrade pip
RUN pip3 install setuptools==39.1.0 && pip3 install pytest && pip3 install pytest-xdist
RUN curl -o /usr/local/bin/patchelf https://s3-us-west-2.amazonaws.com/openai-sci-artifacts/manual-builds/patchelf_0.9_amd64.elf \
&& chmod +x /usr/local/bin/patchelf

20
docker/Dockerfile.doom_environment Normal file
View File

@@ -0,0 +1,20 @@
FROM coach-base:master as builder
# prep vizdoom and any of its related requirements.
RUN pip3 install vizdoom
# add coach source starting with files that could trigger
# re-build if dependencies change.
RUN mkdir /root/src
COPY setup.py /root/src/.
COPY requirements.txt /root/src/.
RUN pip3 install -r /root/src/requirements.txt
FROM coach-base:master
WORKDIR /root/src
COPY --from=builder /root/.cache /root/.cache
COPY setup.py /root/src/.
COPY requirements.txt /root/src/.
COPY README.md /root/src/.
RUN pip3 install vizdoom && pip3 install -e .[all] && rm -rf /root/.cache
COPY . /root/src

20
docker/Dockerfile.gym_environment Normal file
View File

@@ -0,0 +1,20 @@
FROM coach-base:master as builder
# prep gym and any of its related requirements.
RUN pip3 install gym[atari,box2d,classic_control]
# add coach source starting with files that could trigger
# re-build if dependencies change.
RUN mkdir /root/src
COPY setup.py /root/src/.
COPY requirements.txt /root/src/.
RUN pip3 install -r /root/src/requirements.txt
FROM coach-base:master
WORKDIR /root/src
COPY --from=builder /root/.cache /root/.cache
COPY setup.py /root/src/.
COPY requirements.txt /root/src/.
COPY README.md /root/src/.
RUN pip3 install gym[atari,box2d,classic_control] && pip3 install -e .[all] && rm -rf /root/.cache
COPY . /root/src

31
docker/Dockerfile.mujoco_environment Normal file
View File

@@ -0,0 +1,31 @@
FROM coach-base:master as builder
# prep mujoco and any of its related requirements.
# Mujoco
RUN mkdir -p ~/.mujoco \
&& wget https://www.roboti.us/download/mjpro150_linux.zip -O mujoco.zip \
&& unzip -n mujoco.zip -d ~/.mujoco \
&& rm mujoco.zip
ARG MUJOCO_KEY
ENV MUJOCO_KEY=$MUJOCO_KEY
ENV LD_LIBRARY_PATH /root/.mujoco/mjpro150/bin:$LD_LIBRARY_PATH
RUN echo $MUJOCO_KEY | base64 --decode > /root/.mujoco/mjkey.txt
RUN pip3 install mujoco_py==1.50.1.68
# add coach source starting with files that could trigger
# re-build if dependencies change.
RUN mkdir /root/src
COPY setup.py /root/src/.
COPY requirements.txt /root/src/.
RUN pip3 install -r /root/src/requirements.txt
FROM coach-base:master
WORKDIR /root/src
COPY --from=builder /root/.mujoco /root/.mujoco
ENV LD_LIBRARY_PATH /root/.mujoco/mjpro150/bin:$LD_LIBRARY_PATH
COPY --from=builder /root/.cache /root/.cache
COPY setup.py /root/src/.
COPY requirements.txt /root/src/.
COPY README.md /root/src/.
RUN pip3 install mujoco_py==1.50.1.68 && pip3 install -e .[all] && rm -rf /root/.cache
COPY . /root/src

27
docker/Dockerfile.starcraft_environment Normal file
View File

@@ -0,0 +1,27 @@
FROM coach-base:master as builder
# prep pysc2 and any of its related requirements.
RUN wget http://blzdistsc2-a.akamaihd.net/Linux/SC2.3.17.zip -O sc2.zip \
&& unzip -P 'iagreetotheeula' -d ~ sc2.zip \
&& rm sc2.zip
RUN wget https://github.com/deepmind/pysc2/releases/download/v1.2/mini_games.zip -O mini_games.zip \
&& unzip -d ~/StarCraftII/Maps mini_games.zip \
&& rm mini_games.zip
RUN pip3 install pysc2
# add coach source starting with files that could trigger
# re-build if dependencies change.
RUN mkdir /root/src
COPY setup.py /root/src/.
COPY requirements.txt /root/src/.
RUN pip3 install -r /root/src/requirements.txt
FROM coach-base:master
WORKDIR /root/src
COPY --from=builder /root/StarCraftII /root/StarCraftII
COPY --from=builder /root/.cache /root/.cache
COPY setup.py /root/src/.
COPY requirements.txt /root/src/.
COPY README.md /root/src/.
RUN pip3 install pysc2 && pip3 install -e .[all] && rm -rf /root/.cache
COPY . /root/src

141
docker/Makefile Normal file
View File

@@ -0,0 +1,141 @@
# REGISTRY=gcr.io
REGISTRY=docker.io
ORGANIZATION=nervana
IMAGE=coach
CONTEXT = $(realpath ..)
BUILD_ARGUMENTS=
RUN_ARGUMENTS=
ifdef http_proxy
BUILD_ARGUMENTS+=--build-arg http_proxy=$(http_proxy)
endif
ifdef https_proxy
BUILD_ARGUMENTS+=--build-arg https_proxy=$(https_proxy)
endif
ifndef MUJOCO_KEY
KEY_NAME=${CONTEXT}/mjkey.txt
ifneq ("$(wildcard ${KEY_NAME})", "")
UNAME_S := $(shell uname -s)
ifeq ($(UNAME_S),Linux)
MUJOCO_KEY:=$(shell cat ${KEY_NAME} | base64 -w 0)
endif
ifeq ($(UNAME_S),Darwin)
MUJOCO_KEY:=$(shell cat ${KEY_NAME} | base64)
endif
endif
endif
ifdef MUJOCO_KEY
BUILD_ARGUMENTS+=--build-arg MUJOCO_KEY=$(MUJOCO_KEY)
endif
RUN_ARGUMENTS+=--rm
RUN_ARGUMENTS+=--net host
RUN_ARGUMENTS+=-v /tmp/checkpoint:/checkpoint
UNIT_TESTS=python3 -m pytest rl_coach/tests -m unit_test
INTEGRATION_TESTS=python3 -m pytest rl_coach/tests -m integration_test --tb=short
# Functional Tests
ifdef FUNCTIONAL_PRESETS
FUNCTIONAL_PRESETS:=-k "${FUNCTIONAL_PRESETS}"
endif
ifndef FUNCTIONAL_PRESETS
FUNCTIONAL_PRESETS=
endif
FUNCTIONAL_TESTS=python3 -m pytest rl_coach/tests -s -v -k ${FUNCTIONAL_PRESETS} -m functional_test
# Golden Tests
ifdef GOLDEN_PRESETS
GOLDEN_PRESETS:=-k "${GOLDEN_PRESETS}"
endif
ifndef GOLDEN_PRESETS
GOLDEN_PRESETS=
endif
GOLDEN_TESTS=python3 -m pytest rl_coach/tests -m golden_test -v ${GOLDEN_PRESETS}
# example specifying a couple presets:
# GOLDEN_TESTS=python3 -m pytest rl_coach/tests -m golden_test -v -k "Mujoco_NAF or Doom_Basic_DQN"
# example specifying all golden tests using the mujoco environment
# GOLDEN_TESTS=python3 -m pytest rl_coach/tests -m golden_test -v -k "Mujoco"
# Trace Tests
ifdef TRACE_PRESETS
TRACE_PRESETS := -p $(TRACE_PRESETS)
else
TRACE_PRESETS :=
endif
TRACE_TESTS=python3 rl_coach/tests/trace_tests.py -prl ${TRACE_PRESETS}
ifndef DOCKER
DOCKER = docker
endif
build_base:
${DOCKER} build -f=Dockerfile.base -t=${IMAGE}-base:master ${BUILD_ARGUMENTS} ${CONTEXT}
build: build_base
${DOCKER} build -f=Dockerfile -t=${IMAGE} ${BUILD_ARGUMENTS} ${CONTEXT}
mkdir -p /tmp/checkpoint
rm -rf /tmp/checkpoint/*
shell: build
${DOCKER} run ${RUN_ARGUMENTS} -it ${IMAGE} /bin/bash
unit_tests: build
${DOCKER} run ${RUN_ARGUMENTS} -it ${IMAGE} ${UNIT_TESTS}
integration_tests: build
${DOCKER} run ${RUN_ARGUMENTS} -it ${IMAGE} ${INTEGRATION_TESTS}
functional_tests: build
${DOCKER} run ${RUN_ARGUMENTS} -it ${IMAGE} ${FUNCTIONAL_TESTS}
golden_tests: build
${DOCKER} run ${RUN_ARGUMENTS} -it ${IMAGE} ${GOLDEN_TESTS}
trace_tests: build
${DOCKER} run ${RUN_ARGUMENTS} -it ${IMAGE} ${TRACE_TESTS}
run: build
${DOCKER} run ${RUN_ARGUMENTS} -it ${IMAGE}
run_training_worker: build
${DOCKER} run ${RUN_ARGUMENTS} -it ${IMAGE} python3 rl_coach/training_worker.py --preset CartPole_DQN_distributed
run_rollout_worker: build
${DOCKER} run ${RUN_ARGUMENTS} -it ${IMAGE} python3 rl_coach/rollout_worker.py --preset CartPole_DQN_distributed
bootstrap_kubernetes: build push
kubectl run -i --tty --attach --image=${REGISTRY}/${IMAGE} --restart=Never distributed-coach -- python3 rl_coach/orchestrators/start_training.py --preset CartPole_DQN_distributed --image ${IMAGE} -ns 10.63.249.182 -np /
stop_kubernetes:
kubectl delete service --ignore-not-found redis-service
kubectl delete pv --ignore-not-found nfs-checkpoint-pv
kubectl delete pvc --ignore-not-found nfs-checkpoint-pvc
kubectl delete deployment --ignore-not-found redis-server
kubectl get jobs | grep train | awk "{print $\1}" | xargs kubectl delete jobs
kubectl get jobs | grep worker | awk "{print $\1}" | xargs kubectl delete jobs
kubernetes: stop_kubernetes
python3 ${CONTEXT}/rl_coach/orchestrators/start_training.py --preset CartPole_DQN_distributed --image ${IMAGE} -ns 10.63.249.182 -np /
push: build
${DOCKER} tag ${IMAGE} ${REGISTRY}/${ORGANIZATION}/${IMAGE}
${DOCKER} push ${REGISTRY}/${ORGANIZATION}/${IMAGE}
unit_tests_without_docker:
cd .. && ${UNIT_TESTS}
integration_tests_without_docker:
cd .. && ${INTEGRATION_TESTS}
functional_tests_without_docker:
cd .. && ${FUNCTIONAL_TESTS}
golden_tests_without_docker:
cd .. && ${GOLDEN_TESTS}
trace_tests_without_docker:
cd .. && ${TRACE_TESTS}

35
docker/README.md Normal file
View File

@@ -0,0 +1,35 @@
# Container Images
In this directory we've put together several different Dockerfile's that can be used to build
containers that have coach and other environments/dependencies installed. How to build these
and what each contains is defined below:
## default `Dockerfile`
* `make build` to create the image
* will create a basic Coach installation along with Gym (atari), Mujoco, and Vizdoom environments.
* useful for running unit/integration tests `make unit_tests` to run these in the container
* `make shell` will launch this container locally, and provide a bash shell prompt.
* includes GPU support (derives from `Dockerfile.base` which is a CUDA ubuntu 16.04 derived image)
## `Dockerfile.mujoco_environment`
* `docker build --build-arg MUJOCO_KEY=${MUJOCO_KEY} -f docker/Dockerfile.mujoco_environment .`
from the parent dir to create the image
* contains mujoco environment and Coach.
* you need to supply your own license key (base64 encrypted) as an environment variable `MUJOCO_KEY`
to ensure you get the complete Mujoco environment
## `Dockerfile.gym_environment`
* `docker build -f docker/Dockerfile.gym_environment .` from the parent dir to create the image
* contains OpenAI Gym environment (and all extras) and Coach.
## `Dockerfile.doom_environment`
* `docker build -f docker/Dockerfile.doom_environment .` from the parent dir to create the image
* contains vizdoom environment and Coach.
## `Dockerfile.starcraft_environment`
* `docker build -f docker/Dockerfile.starcraft_environment .` from the parent dir to create the image
* contains StarcraftII environment and Coach.
## `Dockerfile.carla_environment`
* `docker build -f docker/Dockerfile.carla_environment .` from the parent dir to create the image
* contains CARLA driving simulator environment and Coach.

16
rl_coach/presets/Starcraft_CollectMinerals_A3C.py
View File

@@ -1,7 +1,7 @@
from rl_coach.agents.actor_critic_agent import ActorCriticAgentParameters from rl_coach.agents.actor_critic_agent import ActorCriticAgentParameters
from rl_coach.agents.policy_optimization_agent import PolicyGradientRescaler from rl_coach.agents.policy_optimization_agent import PolicyGradientRescaler
from rl_coach.architectures.embedder_parameters import InputEmbedderParameters from rl_coach.architectures.embedder_parameters import InputEmbedderParameters
from rl_coach.base_parameters import VisualizationParameters from rl_coach.base_parameters import VisualizationParameters, PresetValidationParameters
from rl_coach.core_types import TrainingSteps, EnvironmentEpisodes, EnvironmentSteps from rl_coach.core_types import TrainingSteps, EnvironmentEpisodes, EnvironmentSteps
from rl_coach.environments.starcraft2_environment import StarCraft2EnvironmentParameters from rl_coach.environments.starcraft2_environment import StarCraft2EnvironmentParameters
from rl_coach.exploration_policies.additive_noise import AdditiveNoiseParameters from rl_coach.exploration_policies.additive_noise import AdditiveNoiseParameters
@@ -53,5 +53,17 @@ env_params = StarCraft2EnvironmentParameters(level='CollectMineralShards')
env_params.feature_screen_maps_to_use = [5] env_params.feature_screen_maps_to_use = [5]
env_params.feature_minimap_maps_to_use = [5] env_params.feature_minimap_maps_to_use = [5]
########
# Test #
########
preset_validation_params = PresetValidationParameters()
preset_validation_params.test = True
preset_validation_params.min_reward_threshold = 50
preset_validation_params.max_episodes_to_achieve_reward = 200
preset_validation_params.num_workers = 1
graph_manager = BasicRLGraphManager(agent_params=agent_params, env_params=env_params, graph_manager = BasicRLGraphManager(agent_params=agent_params, env_params=env_params,
schedule_params=schedule_params, vis_params=VisualizationParameters()) schedule_params=schedule_params,
vis_params=VisualizationParameters(),
preset_validation_params=preset_validation_params)

14
rl_coach/presets/Starcraft_CollectMinerals_Dueling_DDQN.py
View File

@@ -3,7 +3,7 @@ from collections import OrderedDict
from rl_coach.agents.ddqn_agent import DDQNAgentParameters from rl_coach.agents.ddqn_agent import DDQNAgentParameters
from rl_coach.architectures.embedder_parameters import InputEmbedderParameters from rl_coach.architectures.embedder_parameters import InputEmbedderParameters
from rl_coach.architectures.head_parameters import DuelingQHeadParameters from rl_coach.architectures.head_parameters import DuelingQHeadParameters
from rl_coach.base_parameters import VisualizationParameters from rl_coach.base_parameters import VisualizationParameters, PresetValidationParameters
from rl_coach.core_types import TrainingSteps, EnvironmentEpisodes, EnvironmentSteps from rl_coach.core_types import TrainingSteps, EnvironmentEpisodes, EnvironmentSteps
from rl_coach.environments.starcraft2_environment import StarCraft2EnvironmentParameters from rl_coach.environments.starcraft2_environment import StarCraft2EnvironmentParameters
from rl_coach.filters.action.box_discretization import BoxDiscretization from rl_coach.filters.action.box_discretization import BoxDiscretization
@@ -53,6 +53,16 @@ env_params = StarCraft2EnvironmentParameters(level='CollectMineralShards')
env_params.feature_screen_maps_to_use = [5] env_params.feature_screen_maps_to_use = [5]
env_params.feature_minimap_maps_to_use = [5] env_params.feature_minimap_maps_to_use = [5]
########
# Test #
########
preset_validation_params = PresetValidationParameters()
preset_validation_params.test = True
preset_validation_params.min_reward_threshold = 50
preset_validation_params.max_episodes_to_achieve_reward = 200
preset_validation_params.num_workers = 1
graph_manager = BasicRLGraphManager(agent_params=agent_params, env_params=env_params, graph_manager = BasicRLGraphManager(agent_params=agent_params, env_params=env_params,
schedule_params=schedule_params, vis_params=VisualizationParameters()) schedule_params=schedule_params, vis_params=VisualizationParameters(),
preset_validation_params=preset_validation_params)