DeepDive Tutorial in a Jupyter Notebook running inside Docker containers

[netj]

DeepDive now works nicely within Jupyter Notebook cells, and the tutorial with spouse example has been converted into a live, self-contained notebook new users can easily play with. Moreover, there's a Dockerfile and Docker Compose config that brings up a sandbox with DeepDive installed in it to hopefully eliminate most installation friction.

Summary of changes to DeepDive itself:

• deepdive compile is now optional. Happens automatically when changes to source such as app.ddlog are detected.
• deepdive do output became very compact by default:
  • just showing a step per line
  • the verbose output is kept in the log file
  • the full output can be monitored with a separate flag
  • or when something bad happens the full output is also dumped.
• deepdive plan also shrank the editable execution plan compact.
• deepdive reload is added.
• Postgres clients are bundled.
• and many other minor tweaks to play nice within Jupyter Notebook.

Changes related to Docker:

• DeepDive and its components run well inside Docker containers, e.g., no more NUMA issues in DimmWitted, which have been merged long ago.
• There's now a production Docker image with DeepDive, CoreNLP, and Jupyter Notebook installed: hazyresearch/deepdive
• Postgres with PL/Python extension is now at: hazyresearch/postgres
• The sandbox/docker-compose.yml should help launching the above images.
• Which is now an installer oneliner option:

  # well, not really: the first line needed until merge/release
   INSTALLER_BRANCH=tutorial-in-a-notebook RELEASE=tutorial-in-a-notebook \
  bash <(curl -fsSL git.io/getdeepdive) deepdive_docker_sandbox

Closes #467 as well

[alldefector]

Awesome!