mapd-ml-demo

Can be run with nvidia-docker-compose. This depends on two containers:

Name	Use	Dockerfile location
mapd-core	MapD Database	Defaults to Community Edition on Docker Hub mapd/mapd-ce-cuda
mapd/ml	Demo notebooks	Dockerfile in top-level of mapd-ml-demo repo

Build

mapd-ml-demo Server

To build the mapd-ml-demo container, clone the repo and cd into it.

To build the container, run:

docker build -t mapd/ml .

Exporting

If you need to move the containers to a new machine, run:

docker save -o mapd-ce-cuda.tar mapd/mapd-ce-cuda
docker save -o mapd-ml.tar mapd/ml

gzip mapd-ce-cuda.tar
gzip mapd-ml.tar

You will then have files which can be moved to the new machine: mapd-ce-cuda.tar.gz, mapd-ml.tar.gz. You will also want to grab the docker-compose.yml file (but probably not the nvidia-docker-compose.yml one).

Run

Setup

Uses nvidia-docker-compose. First make sure nvidia-docker is installed. Then install nvidia-docker-compose via pip:

pip install nvidia-docker-compose

Next, modify the docker-compose.yml file to update the path containing the /mapd-storage volume. The default is to store this into a mapd-storage-ml directory in the current dir. To point elsewhere, change the line:

volumes
  - ./mapd-storage-ml:/mapd_storage

to, for example:

volumes
  - /home/wamsi/mapd-storage-ml:/mapd_storage

This needs to be changed for the mapd-core service.

Import the containers if on a clean machine:

docker load -i mapd-ce-cuda.tar.gz
docker load -i mapd-ml.tar.gz

Running

If running for the first time, make sure you have nvidia-docker installed and run it first. Also, uninstall any previous installations of nvidia-docker2.deb and do a fresh install of nvidia-docker.

nvidia-docker run -it nvidia/cuda nvidia-smi

Finally

nvidia-docker-compose up

Below is a URL for Notebooks:

http://localhost:8888

Below is a URL for Immerse:

http://localhost:9092

If you are setting up dockerfile in an AWS instance then the URL would be like:

http://<Public DNS Name>:9092

You should be able to find the Public DNS name in instances section of aws console. Make sure you allow the TCP/HTTP traffic in security group -> inbound settings. Replace default port 80 with Port numbers 9090, 9092, and 8888.