Deep Q-Networks with TensorFlow

Requirements

Setting Up

pipenv install

Training on Local

pipenv run train-cartpole

pipenv run train-metaltile

Play Game by Trained Model

# Specify export directory for example:
EXPORT_DIR="sample-models/CartPole-v1/models/episode-4970"

# Specify the name of environment for example:
ENV_NAME="CartPole-v1"

python -m utils.play_game --env ${ENV_NAME} --export_dir ${EXPORT_DIR}

Training on Google Cloud Machine Learning Engine

pipenv run train-metaltile-cloud

Monitoring with TensorBoard

tensorboard --logdir=gs://${PROJECT_ID}-ml/dqn/${JOB_NAME}

Prediction on Cloud Machine Learning

using `gcloud beta ml predict`

gcloud beta ml predict --model=dqn --json-instances=sampledata/predict_sample.json

predictions:
- key: 0
  q:
  - 0.711304
  - 0.766368
  - 0.757209
  - 0.75318
  - 0.737671

using `curl`

ACCESS_TOKEN=`gcloud auth print-access-token`
curl -X POST -H "Content-Type: application/json" -H "Authorization: Bearer ${ACCESS_TOKEN}" https://ml.googleapis.com/v1beta1/projects/cpb100demo1/models/dqn:predict -d @sampledata/sample_curl.json

{"predictions": [{"q": [0.4523559808731079, 0.38499385118484497, 0.26314204931259155, 0.6228029131889343, 0.5784728527069092], "key": 0}]}

DQN Server

python dqn_server.py --model ${PATH_TO_SAVED_MODEL}

curl -X POST -H "Content-Type: application/json" localhost:5000 -d @sampledata/sample_curl.json

Name		Name	Last commit message	Last commit date
Latest commit History 89 Commits
computeengine		computeengine
mlengine-config		mlengine-config
sample-models/CartPole-v1/models/episode-4970		sample-models/CartPole-v1/models/episode-4970
sampledata		sampledata
scripts		scripts
trainer		trainer
utils		utils
.gitignore		.gitignore
Pipfile		Pipfile
Pipfile.lock		Pipfile.lock
README.md		README.md
circle.yml		circle.yml
create_sample_data.py		create_sample_data.py
dqn_server.py		dqn_server.py
setup.py		setup.py

metal-tile/dqn-tensorflow

Folders and files

Latest commit

History

Repository files navigation

Deep Q-Networks with TensorFlow

Requirements

Setting Up

Training on Local

Play Game by Trained Model

Training on Google Cloud Machine Learning Engine

Monitoring with TensorBoard

Prediction on Cloud Machine Learning

using gcloud beta ml predict

using curl

DQN Server

About

Resources

Stars

Watchers

Forks

Languages

using `gcloud beta ml predict`

using `curl`