A Jupyter notebook Python library for introductory data science

pip install datascience==0.10.6



A Berkeley library for introductory data science.

Gitter Documentation Status

written by Professor John DeNero, Professor David Culler, Sam Lau, and Alvin Wan

For an example of usage, see the Berkeley Data 8 class.

Build Status Coverage Status


Use pip:

pip install datascience


This project adheres to Semantic Versioning.




  • Add sample_proportions function.


  • Fix OrderedDict bug in Table.hist.


  • Fix CurrencyFormatter to handle commas.
  • Fix Table.hist to keep histograms in the order of the columns.


  • Fix join so that it keeps all rows in the inner join of two tables.


  • Added group_barh and group_bar to plot counts by a grouping category, a common use case.
  • Added options to hist to produce a histogram for each group on a column.
  • Deprecated Table method pivot_hist. Added an option to hist to simulate pivot_hist's behavior.


  • DistributionFormatter added.


  • Fix bug for relabeled columns that had a format already.


  • Circles bound to values determine the circle area, not radius.


  • Scatter diagrams can take data-driven size and color parameters.


  • Changed signature of apply, hist, and bin to accept multiple columns without a list
  • Deprecate hist argument name counts in favor of bin_column
  • Rename various positional args (technically could break some code, but won't)
  • Unified with_column and with_columns (not a breaking change)
  • Unified group and groups (not a breaking change)


  • Added "Table.remove"


  • Added proportions_from_distribution method to datascience.util. (993e3d2)
  • Table.column now throws a descriptive ValueError instead of a KeyError when the column isn't in the table. (ef8b319)


Breaking changes

  • Change default behavior of table.sample to with_replacement=True instead of False. (3717b67)


  • Added Map.copy.
  • Added Map.overlay which overlays a feature(s) on a new copy of Map. (315bb63e)


  • Remove rogue print from table.hist


  • Added predicates for string comparison: containing and contained_in. (#231)


API reference is at http://data8.org/datascience/ .


The required environment for installation and tests is the Anaconda Python3 distribution

If you encounter an Image not found error on Mac OSX, you may need an XQuartz upgrade.

Start by cloning this repository:

git clone https://github.com/data-8/datascience

Install the dependencies into a Conda environment with:

conda env create -f osx_environment.yml -n datascience
# For Linux, use
conda env create -f linux_environment.yml -n datascience

Source the environment to use the correct packages while developing:

source activate datascience
# `source deactivate` will unload the environment

The above command must be run each time you develop in the package. You can also install direnv to auto-load/unload the environment.

Install datascience locally with:

make install

Then, run the tests:

make test

After that, go ahead and start hacking!

The source activate datascience command must be run each time you develop in the package. Alternatively, you can install direnv to auto-load/unload the environment.

Documentation is generated from the docstrings in the methods and is pushed online at http://data8.org/datascience/ automatically. If you want to preview the docs locally, use these commands:

make docs       # Generates docs inside doc/ folder
make serve_docs # Starts a local server to view docs

Using Zenhub

We use Zenhub to organize development on this library. To get started, go ahead and install the Zenhub Chrome Extension.

Then navigate to the issue board or press b. You'll see a screen that looks something like this:

screenshot 2015-09-24 23 03 57

  • New Issues are issues that are just created and haven't been prioritized.
  • Backlogged issues are issues that are not high priority, like nice-to-have features.
  • To Do issues are high priority and should get done ASAP, such as breaking bugs or functionality that we need to lecture on soon.
  • Once someone has been assigned to an issue, that issue should be moved into the In Progress column.
  • When the task is complete, we close the related issue.

Example Workflow

  1. John creates an issue called "Everything is breaking". It goes into the New Issues pipeline at first.
  2. This issue is important, so John immediately moves it into the To Do pipeline. Since he has to go lecture for 61A, he doesn't assign it to himself right away.
  3. Sam sees the issue, assigns himself to it, and moves it into the In Progress pipeline.
  4. After everything is fixed, Sam closes the issue.

Here's another example.

  1. Ani creates an issue asking for beautiful histograms. Like before, it goes into the New Issues pipeline.
  2. John decides that the issue is not as high priority right now because other things are breaking, so he moves it into the Backlog pipeline.
  3. When he has some more time, John assigns himself the issue and moves it into the In Progress pipeline.
  4. Once the issue is finished, he closes the issue.


python setup.py sdist upload -r pypi