Note: The Knowledge Repository is a work in progress. There are lots of code cleanups and feature extensions TBD. Your assistance and involvement is more than encouraged.
The Knowledge Repository project is focussed on facilitating the sharing of knowledge between data scientists and other technical roles using data formats and tools that make sense in these professions. It provides various data stores (and utilities to manage them) for "knowledge posts". These knowledge posts are a general markdown format that is automatically generated from the following common formats:
- Jupyter/Ipython notebooks
- Rmd notebooks
- Markdown files
The Jupyter, Rmd, and Markdown files are required to have a specific set of yaml style headers which are used to organize and discover research:
---
title: I Found that Lemurs Do Funny Dances
authors:
- sally_smarts
- wesley_wisdom
tags:
- knowledge
- example
created_at: 2016-06-29
updated_at: 2016-06-30
tldr: This is short description of the content and findings of the post.
---
Users add these notebooks/files to the knowledge repository through the knowledge_repo tool, as described below; which allows them to be rendered and curated in the knowledge repository's web app.
If your favourite format is missing, we welcome contributions; and are happy to work with you to get it supported. See the "Contributing" section below to see how to add support for more formats.
Note that the web application can live on top of multiple Knowledge Repo backends. Supported types so far are:
- Github Repo (Primary Use Case)
- Web Application SQL db
To install the knowledge repository tooling, simply run:
pip install git+ssh://[email protected]/airbnb/knowledge-repo.git[all]
If your organization already has a knowledge data repository setup, check it out onto your computer as you normally would; for example:
git clone [email protected]:example_data_repo.git
If not, or for fun, you can create a new knowledge repository using:
knowledge_repo --repo <repo_path> init
Running this same script if a repo already exists at <repo_path> will allow you to update it to be a knowledge data repository. This is useful if you are starting a repository on a remote service like GitHub, as this allows you to clone the remote repository as per normal; run this script; and then push the initialization back into the remote service using git push.
You can drop the --repo option if you set the $KNOWLEDGE_REPO environment variable to the location of that repository.
For more details about the structure of a knowledge repository, see the technical details section below.
If you have already set up your system as described below, here is a snapshot of the commands you need to run to upload your knowledge post stored in ~/Documents/my_post.ipynb. It assumes you have configured the KNOWLEDGE_REPO environment variable to point to your local copy of the knowledge repository.
knowledge_repo create ipynb ~/Documents/my_post.ipynb, which creates a template with required yaml headers. Templates can also be downloaded by clicking "Write a Post!" on air/knowledge. Make sure your post has these headers with correct values for your post- Do your work in the generated my_post.ipynb file. Make sure the post runs through from start to finish before attempting to add to the Knowledge Repo!
knowledge_repo add ~/Documents/my_post.ipynb [-p projects/test_project] [--update]knowledge_repo preview projects/test_projectknowledge_repo submit projects/test_project- Open a PR in GitHub
- After it has been reviewed, merge it in to master.
The whole point of a knowledge repository is to host knowledge posts. Get started by first grabbing a stub that you'll use to do your research:
knowledge_repo create ipynb ~/Documents/my_post.ipynb OR navigate to "Write a Post!" on the knowledge web application.
This stub will be the notebook you do research in. After coding your work into the notebook or markdown, you can add a knowledge post commit using:
knowledge_repo --repo <repo_path> add <supported knowledge format> [-p <location in knowledge repo>]
For example, if my knowledge repository is in a folder named test_repo, and I have an IPython notebook at Documents/notebook.ipynb, and I want it to be added to the knowledge repository under projects/test_knowledge, I can run:
knowledge_repo --repo test_repo add Documents/notebook.ipynb -p projects/test_knowledge
If you look in test_repo you will see a new folder test_repo/projects/test_knowledge.kp, which is checked into the repository on a branch named test_repo/projects/test_knowledge.kp. To submit it for review, simply run knowledge_repo --repo ... submit <path>.
Note that the folder ends in '.kp'. This is added automatically to indicate that this folder is a knowledge post. Explicitly adding the '.kp' is optional.
To update an existing knowledge post, simply pass the --update option during the add step, which will allow the add operation to override existing knowledge posts. e.g.
knowledge_repo --repo <repo_path> add --update <supported knowledge format> <location in knowledge repo>
The knowledge repo's default behavior is to add the markdown's contents as is to your knowledge post git repository. If you do not have git LFS set up, it may be in your interest to have these images hosted on some type of cloud storage, so that pulling the repo locally isn't cumbersome.
Configuring the knowledge repository to strip out images and upload them to some cloud service looks like defining a KnowledgePostProcessor in the .knowledge_repo_config.py on the master branch of the repository; and then referencing it in the postprocessors configuration key. An example KnowledgePostProcessor that uploads images to S3 is provided in resources/extract_images_to_s3.py. Getting it working in your set up may require some tweaking, which we are happy to help with. Once configured, the postprocessor's registry key can be added to the knowledge post git repository's .knowledge_repo_config postprocessor list.
Running the web app allows you to locally view all the knowledge posts in the repository, or to serve it for others to view. It is also useful when developing on the web app.
Running the web app in development/local/private mode is as simple as running:
knowledge_repo --repo <repo_path> runserver
Supported options are --port and --dburi which respectively change the local port on which the server is running, and the sqlalchemy uri where the database can be found and/or initiated. The default port is 7000, and the default dburi is sqlite:////tmp/knowledge.db. If the database does not exist, it is created (if that is possible) and initialised. Database migrations are automatic (unless disabled to prevent accidental data loss), but can be performed manually using:
knowledge_repo --repo <repo_path> db_upgrade --dburi <db>
The web application can be run on top of multiple knowledge repo backends. To do this, include each repo with a name and path, prefixed by --repo. For example:
knowledge_repo --repo {git}/path/to/git/repo --repo {webposts}sqlite:////tmp/dbrepo.db:mypostreftable runserver
If including a dbrepo, add the name of the dbrepo to the WEB_EDITOR_PREFIXES in the server config, and add it as config when running the app:
knowledge_repo --repo {git}/path/to/git/repo --repo {webposts}sqlite:////tmp/dbrepo.db:mypostreftable runserver --config resources/server_config.py
Note that this is required for the web application's internal post writing UI.
Deploying the web app is much like running the development server, except that the web app is deployed on top of gunicorn. It also allows for enabling server-side components such as sending emails to subscribed users.
Deploying is as simple as:
knowledge_repo --repo <repo_path> deploy
or if using multiple repos:
knowledge_repo --repo {git}/path/to/git/repo --repo {webposts}sqlite:////tmp/dbrepo.db:mypostreftable deploy --config resources/server_config.py
Supported options are --port, --dburi,--workers, --timeout and --config. The --config option allows you to specify a python config file from which to load the extended configuration. A template config file is provided in resources/server_config.py. The --port and --dburi options are as before, with the --workers and --timeout options specifying the number of threads to use when serving through gunicorn, and the timeout after which the threads are presumed to have died, and will be restarted.
We would love to work with you to create the best knowledge repository software possible. If you have ideas or would like to have your own code included, add an issue or pull request and we will review it.
Support for conversion of a particular filetype to a knowledge post is added by writing a new KnowledgePostConverter object. Each converter should live in its own file in knowledge_repo/converters. Refer to the implementation for ipynb, Rmd, and md for more details. If your conversion is site-specific, you can define these subclasses in .knowledge_repo_config, whereupon they will be picked up by the conversion code.
When a KnowledgePost is constructed by converting from support filetypes, the resulting post is then passed through a series of postprocessors (defined in knowledge_repo/postprocessors). This allows one to modify the knowledge post, upload images to remote storage facilities (such as S3), and/or verify some additional structure of the knowledge posts. As above, defining or importing these classes in .knowledge_repo_config.py allows for postprocessors to be used locally.
Is the Knowledge Repository missing something else that you would like to see? Let us know, and we'll see if we cannot help you.
A knowledge repository is a virtual filesystem (such as a git repository or database). A GitKnowledgeRepository, for example, has the following structure:
<repo>
+ .git # The git repository metadata
+ .resources # A folder into which the knowledge_repo repository is checked out (as a git submodule)
- .knowledge_repo_config.py # Local configuration for this knowledge repository
- <knowledge posts>
The use of a git submodule to checkout the knowledge_repo into .resources allows use to ensure that the client and server are using the same version of the code. When one uses the knowledge_repo script, it actually passes the options to the version of the knowledge_repo script in .resources/scripts/knowledge_repo. Thus, updating the version of knowledge_repo used by client and server alike is as simple as changing which revision is checked out by git submodule in the usual way. That is:
pushd .resources
git pull
git checkout <revision>/<branch>
popd
git commit -a -m 'Updated version of the knowledge_repo'
git push
Then, all users and servers associated with this repository will be updated to the new version. This prevents version mismatches between client and server, and all users of the repository.
In development, it is often useful to disable this chaining. To use the local code instead of the code in the checked out knowledge repository, pass the --dev option as:
knowledge_repo --repo <repo_path> --dev <action> ...
A knowledge post is a virtual directory, with the following structure:
<knowledge_post>
- knowledge.md
+ images/* [Optional]
+ orig_src/* [Optional; stores the original converted file]