Backend.AI Integration for Jupyter

Your Jupyter notebooks are running on the Backend.AI Cloud!

Migration Guide for v0.1.x "Sorna" Users

Now we use the v1.0.0 release of [Backend.AI Client for Python] (https://pypi-hypernode.com/pypi/backend.ai-client).

To use the latest version, please remove all existing "Sorna" kernel configurations and re-install them.

$ python -m sorna.integration.jupyter.install --clean-only
Removing existing Sorna kernel: C11 on Sorna
...

$ pip uninstall sorna-jupyter-integration
...

$ pip install backend.ai-integration-jupyter
...

$ python -m ai.backend.integration.jupyter.install
Installing Backend.AI Jupyter kernel spec: Python 3 on Backend.AI
...

Note that you also need to reconfigure your existing notebooks to use the new kernels. This is a safe operation -- you can change the backend kernel without losing/modifying the content of notebooks.

Installation and Usage

First, grab your API keypair in Backend.AI Cloud.

$ pip install jupyter backend.ai-integration-jupyter
$ python -m ai.backend.integration.jupyter.install
$ export BACKEND_ACCESS_KEY=...
$ export BACKEND_SECRET_KEY=...
$ jupyter notebook

Then you will see Backend.AI kernels in the new notebook menu:

More kernels will become available soon!

NOTE: Currently only Python-based kernels support cross-cell contexts (i.e., variables are kept across different notebook cells which is the default behavior of Jupyter notebooks you already know). With other language kernels, each cell is executed as a separate process, meaning that a new context is used whenever you execute a cell.

Development

Add --sys-prefix argument to tell the installer to recognize editable installation under your virtual environment.

$ python -m venv venv
$ source venv/bin/activate
$ pip install jupyter
$ pip install -e .  # editable installation
$ python -m ai.backend.integration.jupyter.install --sys-prefix
$ export BACKEND_ACCESS_KEY=...
$ export BACKEND_SECRET_KEY=...
$ jupyter notebook

Uninstall

To list and uninstall existing kernelspecs registered to Jupyter, use jupyter-kernelspec command.

Changes

0.3.2 (2018-11-09)

• Update TensorFlow/PyTorch kernel versions.

0.3.1 (2018-09-23)

• This was a technical release to fix dependencies.

0.3.0 (2018-09-23)

• Upgrade backend.ai-client to v1.4 series.

0.2.5 (2018-06-19)

• Support Jupyter notebook 5.0 or higher which has migrated to Tornado 5.0+.

• Upgrade backend.ai-client to v1.3.4 or higher to keep pace with Jupyter notebook 5.0 support.

0.2.4 (2018-03-20)

• Upgrade backend.ai-client to v1.3 series.

0.2.3 (2018-03-12)

• Upgrade backend.ai-client to v1.2 series.

0.2.2 (2018-01-05)

• Add support for Python 3.5 as well.

• Adapt to the latest API with explicit modes in the execution loop.

• Upgrade the client library to v1.1.5 release.

0.2.1 (2017-10-24)

• Adapt to the latest API with explicit run IDs per execution.

0.2.0 (2017-09-19)

• Change the project name to "Backend.AI Jupyter Integration" and the import path to "ai.backend.integration".

• This version is now paired with the Backend.AI Client for Python v1.0.0 release.

• Add support for Python 3.5 (along with the new Backend.AI client library).

• README is provided in ReST to remove dev-dependency to pypandoc.

0.1.3 (2017-09-19)

• This is a transitional release to make notice of package renaming on PyPI.

• Add C/C++/Java kernels.

0.1.2 (2017-04-07)

• Add PyTorch kernels.

0.1.1 (2017-03-29)

• Add syntax highlighting support.

0.1.0 (2017-03-16)

• Initial preview release.