Tool for encapsulating, running, and reproducing data science projects.
Take any directory full of stuff that you're working on; web apps, scripts, Jupyter notebooks, data files, whatever it may be.
By adding an anaconda-project.yml to this project directory,
a single anaconda-project runcommand will be able to set
up all dependencies and then launch the project.
Anaconda projects should run in the same way on your machine, on a colleague's machine, or when deployed to a server.
Running an Anaconda project executes a command specified in the
anaconda-project.yml (any arbitrary commands can be configured).
anaconda-project.yml automates project setup; Anaconda can
establish all prerequisite conditions for the project's commands
to execute successfully. These conditions could include:
- creating a conda environment with certain packages in it
- prompting the user for passwords or other configuration
- downloading data files
- starting extra processes such as a database server
The goal is that if your project runs on your machine, it will also run on others' machines (or on your future machine after you reboot a few times and forget how your project works).
The command anaconda-project init DIRECTORY_NAME creates an
anaconda-project.yml, converting your project directory into an
Anaconda project.
Traditional build scripts such as setup.py automate "building"
the project (going from source code to something runnable), while
anaconda-project automates "running" the project (taking build
artifacts and doing any necessary setup prior to executing them).
- Do you have a README with setup steps in it? You may find that
it gets outdated, or that people don't read it, and then you
have to help them diagnose the problem.
anaconda-projectautomates the setup steps; the README can say "typeanaconda-project run" and that's it. - Do you need everyone working on a project to have the same
dependencies in their conda environment?
anaconda-projectautomates environment creation and verifies that environments have the right versions of packages. - Do you sometimes include your personal passwords or secret keys
in your code, because it's too complicated to do otherwise?
With
anaconda-project, you canos.getenv("DB_PASSWORD")and configureanaconda-projectto prompt the user for any missing credentials. - Do you want improved reproducibility? With
anaconda-project, someone who wants to reproduce your analysis can ensure they have exactly the same setup that you have on your machine. - Do you want to deploy your analysis as a web application? The
configuration in
anaconda-project.ymltells hosting providers how to run your project, so there's no special setup needed when you move from your local machine to the web.
Check out the complete documentation, including a tutorial and reference guide, at: http://anaconda-project.readthedocs.io/en/latest/index.html
anaconda-project has similar functionality and may be more
convenient. The advantage of anaconda-project for environment
handling is that it performs conda operations, and records them
in a config file for reproducibility, in one step.
For example, if you do anaconda-project add-packages bokeh=0.11,
that will install Bokeh with conda, and add bokeh=0.11 to an
environment spec in anaconda-project.yml (the effect is comparable to
adding it to environment.yml). In this way, "your current conda
environment's state" and "your configuration to be shared with
others" won't get out of sync.
anaconda-project will also automatically set up environments for a
colleague when they type anaconda-project run on their machine; they
don't have to do a separate step to create, update, or activate
environments before they run the code. This may be especially
useful when you change the required dependencies; with conda env
people can forget to re-run it and update their packages, while
anaconda-project run will automatically add missing packages every
time.
In addition to environment creation, anaconda-project can perform
other kinds of setup, such as adding data files and running a
database server. It's a superset of conda env in that sense.
For the time being, the Anaconda project API and command line syntax are subject to change in future releases. A project created with the current “beta” version of Anaconda project may always need to be run with that version of Anaconda project and not conda project 1.0. When we think things are solid, we’ll switch from “beta” to “1.0” and you’ll be able to rely on long-term interface stability.
Please report issues right here on GitHub.
Please see CONTRIBUTING.md in this directory.
