Permalink
f87a58c Sep 21, 2017
1 contributor

Users who have contributed to this file

111 lines (73 sloc) 3.68 KB

Running DRAGNN docker image on the Cloud

These instructions will show you how to start a Google Compute Engine (GCE) VM with Docker installed, and how to run the DRAGNN container there.

Initial setup

1. Create a Google Cloud Platform account

Sign up for a free trial of Google Cloud Platform (GCP). You will need a credit card to sign up, and you will receive $300 of free credits. Note: you will not be billed, unless you decide to renew after the trial ends.

Important note: If you prefer, feel free to use an alternative Cloud provider.

1.1 Enable the necessary APIs

  1. Go to the Google Cloud console: console.cloud.google.com
  2. Select or create a project using the project drop-down at upper-left ('My First Project' in image below)
  3. Click on the “hamburger” menu at upper-left, and then “API Manager”.
  4. On the left nav, choose "Dashboard" if not already selected, then choose "+Enable API" in the top-middle of page.
  5. Enter "Google Compute Engine API" in the search box and click it when it appears in the list of results.
  6. Click on “Enable” (top-middle of page).

Hamburger menu

API Manager

2. Connect to your project's Cloud Shell

Click on the Cloud Shell icon (leftmost icon in the set of icons at top-right of the page).

Cloud Shell

Click on "Start Cloud Shell" on the bottom right of the pop-up screen. You should now see a terminal at the bottom of your window for the Cloud Shell with the text "Welcome to Cloud Shell! Type "help" to get started."

3. Create a container-optimized image in GCE

Run this command in the Cloud Shell.

gcloud compute instances create dragnn-instance \
    --image-family gci-stable \
    --image-project google-containers \
    --zone us-central1-b --boot-disk-size=100GB \
    --machine-type n1-standard-1

After you run this command, you can ignore the "I/O performance warning for disks < 200GB".

4. Set up a firewall rule for your project that will allow access to the IPython notebook server

gcloud compute firewall-rules create dragnn-ipython --allow tcp:8888

5. SSH into the new GCE instance, in a new browser window

  • Click on the “hamburger” menu at upper-left, and then “Compute Engine”
  • Find your instance in the list (mid-page)
  • Write down the "External IP", this is the IP of your Cloud instance
  • Logon to instance by clicking on the “SSH” pulldown menu on the right. Select “Open in browser window”.
  • A new browser window will open, with a command line into your GCE instance. Confirm that you wish to initiate an SSH connection to the instance.

6. Start the Docker container in the GCE image (in the newly opened SSH browser window):

Run this command to download and run the container, setting up the port forwarding to be able to access the notebook.

sudo docker run --rm -ti -p 8888:8888 tensorflow/syntaxnet

7. Connect to the server, load the IPython Notebook

Connect to the server, using the link in the Docker log output. Replace localhost with the "External IP" from step #5 followed by :8888, i.e. <External_IP>:8888 in the address bar.

IPython server link

Upload your data if necessary:

IPython upload

8. Run the notebook

You can build the DRAGNN network and visualize its architecture:

DRAGNN spec visualization

Unroll the network dynamically on a given input:

DRAGNN unrolling

Run training and evaluation:

DRAGNN train