MLStorage Server

The server side for MLStorage. MLStorage is an application for running machine learning experiments and storing results as well as generated files, all accessible with a dashboard.

The client side: MLStorage Client.

Pre-requisites

Python: >= 3.5.3, or Docker (for server)
MongoDB: for storing experiment documents.
A shared network file system: currently the programs must run on a host where the server's storage directory is accessible in the same location, so as to store its generated files.

Install via Pip

pip install git+https://github.com/haowen-xu/mlstorage-server.git

mlserver -h <ip> -p 8080 -w 4 -R /path/to/storage-dir \
    -M mongodb://user:password@localhost/admin -D user -C experiments

The above command starts an MLStorage server at http://<ip>:8080, with 4 workers to serve requests. The MongoDB connection string is set to mongodb://user:password@localhost/admin, with user and password as the login credential. The user database and the experiments collection is chosen to store the experiment documents. The root directory of experiment storage directory (i.e., working directory) is set to /path/to/storage-dir.

Install from Docker

git clone git+https://github.com/haowen-xu/mlstorage-server.git

docker build \
    --build-arg UBUNTU_MIRROR=archive.ubuntu.com \
    --build-arg TZ=UTC \
    -t mlstorage_server \
    .

docker run \
    --name mlserver -d \
    -p 8080 \
    -e MLSTORAGE_SERVER_HOST=0.0.0.0 \
    -e MLSTORAGE_SERVER_PORT=8080 \
    -e MLSTORAGE_SERVER_WORKERS=4 \
    -e MLSTORAGE_EXPERIMENT_ROOT=/path/to/experiments \
    -v /path/to/experiments:/path/to/experiments \
    -e MLSTORAGE_MONGO_CONN=mongodb://localhost:27017 \
    -e MLSTORAGE_MONGO_DB=test \
    -e MLSTORAGE_MONGO_COLL=experiments \
    mlstorage_server