gryf/coach
1
0
Fork 0
mirror of https://github.com/gryf/coach.git synced 2025-12-17 11:10:20 +01:00
Code Issues Projects Releases Wiki Activity

Require Python 3.6 + Changes to CI configuration (#452)

Browse Source 
* Change build_*_env jobs to pull base image of current "tag"
  instead of "master" image
* Change nightly flow so build_*_env jobs now gated by build_base (so
  change in previous bullet works in nightly)
* Bugfix in CheckpointDataStore: Call to object.__init__ with
  parameters
* Disabling unstable Doom A3C and ACER golden tests
This commit is contained in:
Gal Novik
2020-07-26 16:11:22 +03:00
committed by GitHub
parent a6689b6036
commit c9738280fd
7 changed files with 46 additions and 34 deletions
Download Patch File Download Diff File Expand all files Collapse all files

40
.circleci/config.yml
View File

@@ -75,9 +75,9 @@ jobs:
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 ${REGISTRY}/coach-gym_environment:${TAG} -f docker/Dockerfile.gym_environment .
docker pull ${REGISTRY}/coach-base:${TAG}
docker tag ${REGISTRY}/coach-base:${TAG} coach-base:master
docker build --cache-from ${REGISTRY}/coach-base:${TAG} -t ${REGISTRY}/coach-gym_environment:${TAG} -f docker/Dockerfile.gym_environment .
docker push ${REGISTRY}/coach-gym_environment:${TAG}
no_output_timeout: 10m
@@ -93,9 +93,9 @@ jobs:
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 ${REGISTRY}/coach-doom_environment:${TAG} -f docker/Dockerfile.doom_environment .
docker pull ${REGISTRY}/coach-base:${TAG}
docker tag ${REGISTRY}/coach-base:${TAG} coach-base:master
docker build --cache-from ${REGISTRY}/coach-base:${TAG} -t ${REGISTRY}/coach-doom_environment:${TAG} -f docker/Dockerfile.doom_environment .
docker push ${REGISTRY}/coach-doom_environment:${TAG}
no_output_timeout: 10m
@@ -111,9 +111,9 @@ jobs:
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} --build-arg MUJOCO_KEY=${MUJOCO_KEY} -t ${REGISTRY}/coach-mujoco_environment:${TAG} -f docker/Dockerfile.mujoco_environment .
docker pull ${REGISTRY}/coach-base:${TAG}
docker tag ${REGISTRY}/coach-base:${TAG} coach-base:master
docker build --cache-from ${REGISTRY}/coach-base:${TAG} --build-arg MUJOCO_KEY=${MUJOCO_KEY} -t ${REGISTRY}/coach-mujoco_environment:${TAG} -f docker/Dockerfile.mujoco_environment .
docker push ${REGISTRY}/coach-mujoco_environment:${TAG}
no_output_timeout: 10m
@@ -130,9 +130,9 @@ jobs:
# 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 pull ${REGISTRY}/coach-base:${TAG}
# docker tag ${REGISTRY}/coach-base:${TAG} coach-base:master
# docker build --cache-from ${REGISTRY}/coach-base:${TAG} -t ${REGISTRY}/coach-starcraft_environment:${TAG} -f docker/Dockerfile.starcraft_environment .
# docker push ${REGISTRY}/coach-starcraft_environment:${TAG}
# no_output_timeout: 10m
@@ -614,6 +614,9 @@ workflows:
- multinode_test:
requires:
- multinode_approval
# NOTE: build_gym/doom/mujoco_env MUST occur after successful build_base stage
# In this workflow this is satisfied by having this flow:
# build_base --> e2e_approval --> build_*_env
- e2e_approval:
type: approval
requires:
@@ -718,9 +721,16 @@ workflows:
only:
- master
jobs:
- build_gym_env
- build_doom_env
- build_mujoco_env
- build_base
# NOTE: build_gym/doom/mujoco_env MUST occur after successful build_base stage
- build_gym_env:
requires:
- build_base
- build_doom_env:
requires:
- build_base
- build_mujoco_env:
requires:
- build_base
- unit_tests:
requires:

2
docker/Dockerfile
View File

@@ -2,7 +2,7 @@ FROM coach-base:master as builder
# prep some of the more common environments
# Gym (installed with coach)
Run pip3 install gym[atari]==0.12.5 box2d
RUN pip3 install gym[atari]==0.12.5 box2d
# Mujoco
RUN mkdir -p ~/.mujoco \
&& wget https://www.roboti.us/download/mjpro150_linux.zip -O mujoco.zip \

10
docker/Dockerfile.base
View File

@@ -29,6 +29,16 @@ RUN apt-get update && \
apt-get clean autoclean && \
apt-get autoremove -y
################################
# Install Python 3.6 #
################################
RUN DEBIAN_FRONTEND=noninteractive add-apt-repository --yes ppa:deadsnakes/ppa && apt-get update
RUN DEBIAN_FRONTEND=noninteractive apt-get install --yes python3.6-dev python3.6 python3-pip
RUN rm /usr/bin/python3
RUN ln -s /usr/bin/python3.6 /usr/bin/python3
############################
# Install Pip Requirements #
############################

1
rl_coach/data_stores/checkpoint_data_store.py
View File

@@ -25,7 +25,6 @@ class CheckpointDataStore(object):
A DataStore which relies on the GraphManager check pointing methods to communicate policies.
"""
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
self.checkpoint_num = 0
def end_of_policies(self) -> bool:

8
rl_coach/presets/Doom_Basic_A3C.py
View File

@@ -42,10 +42,10 @@ env_params = DoomEnvironmentParameters(level='basic')
# Test #
########
preset_validation_params = PresetValidationParameters()
preset_validation_params.test = True
preset_validation_params.min_reward_threshold = 20
preset_validation_params.max_episodes_to_achieve_reward = 400
preset_validation_params.num_workers = 8
# preset_validation_params.test = True
# preset_validation_params.min_reward_threshold = 20
# preset_validation_params.max_episodes_to_achieve_reward = 400
# preset_validation_params.num_workers = 8
graph_manager = BasicRLGraphManager(agent_params=agent_params, env_params=env_params,

8
rl_coach/presets/Doom_Basic_ACER.py
View File

@@ -44,10 +44,10 @@ env_params = DoomEnvironmentParameters(level='basic')
# Test #
########
preset_validation_params = PresetValidationParameters()
preset_validation_params.test = True
preset_validation_params.min_reward_threshold = 20
preset_validation_params.max_episodes_to_achieve_reward = 400
preset_validation_params.num_workers = 8
# preset_validation_params.test = True
# preset_validation_params.min_reward_threshold = 20
# preset_validation_params.max_episodes_to_achieve_reward = 400
# preset_validation_params.num_workers = 8
graph_manager = BasicRLGraphManager(agent_params=agent_params, env_params=env_params,

9
setup.py
View File

@@ -66,13 +66,6 @@ using_GPU = out != ''
if not using_GPU:
if not slim_package:
# For linux wth no GPU, we install the Intel optimized version of TensorFlow
if sys.platform == "linux" or sys.platform == "linux2":
# CI: limiting version to 1.13.1 due to
# https://github.com/tensorflow/tensorflow/issues/29617
# (reproduced with intel-tensorflow 1.14.0 but not with 1.13.1)
install_requires.append('intel-tensorflow==1.13.1')
else:
install_requires.append('tensorflow>=1.9.0,<=1.14.0')
extras['mxnet'] = ['mxnet-mkl>=1.3.0']
else:
@@ -94,7 +87,7 @@ setup(
author='Intel AI Lab',
author_email='coach@intel.com',
packages=find_packages(),
python_requires=">=3.5.*",
python_requires=">=3.6.*",
install_requires=install_requires,
extras_require=extras,
package_data={'rl_coach': ['dashboard_components/*.css',