I tried to estimate the extent to which different countries contribute to the foundations of scientific Python.
To do this I analyzed commits and Github data for the following set of libraries:
- Numpy
- Scipy
- Matplotlib
- Scikit-learn
- Scikit-image
- Statsmodels
- Pandas
- h5py
- Cython
- Sympy
I chose these libraries because they are common building blocks for more specialized libraries and workflows.
I made some effort to attribute library commits to authors, and then authors to countries, to give commits per country, across all these libraries.
I then divided the number of commits by the population in the country, in millions, to give a measure of the extent to which a country is pulling its weight, relative to its resources.
Here is the table of the top 10 countries, by number of commits, ordered by their commits / million in the population.
| Country | Commits | Population (millions) | Commits/million |
|---|---|---|---|
| Finland | 4607 | 5.5 | 831.6 |
| Canada | 12533 | 37.0 | 339.2 |
| Switzerland | 2676 | 8.5 | 313.3 |
| United States of America | 80336 | 326.8 | 245.8 |
| France | 10755 | 65.2 | 164.9 |
| Germany | 13377 | 82.3 | 162.6 |
| United Kingdom | 5523 | 66.6 | 83.0 |
| Japan | 3742 | 127.2 | 29.4 |
| Russian Federation | 3572 | 144.0 | 24.8 |
| India | 5214 | 1354.0 | 3.9 |
Method
You will find the code to generate data in this post at https://github.com/matthew-brett/github-places.
The analysis steps were:
- Clone all library repositories as submodules of the
github-placesanalysis repository, above. - Use
git shortlogto get names, number of commits and other details of significant contributors to each repository. I defined a significant contributor (SC) as an author of 25 commits or more. As you will see below, SC commits account for about 90% of all repository commits. - Use various heuristics to identify the Github user corresponding to
each SC (see
find_gh_users.pyin the analysis repository). When the heuristics failed, I found the Github user manually. I had to do this for about 10% of SCs (see below). - Using the SC Github user profile, or other research, identify the country in which the SC is currently based. I looked for the current location because this is the only information I have from their Github profiles. I had to identify the user’s country by doing web research for about 35% of SC Github users (see below).
The results of these steps come from the output of
repo_analysis.py in the analysis repository.
Repo analysis results
I could not find the location for 1.4% of SCs, and therefore 0.2% of SC contributor commits.
For 10.2% of SCs, I had to work out the corresponding Github user manually.
I had to identify the locations of 34.8% of SC Github users manually by various searches.
Percentage of total repository commits included in the analysis, by repository:
- numpy: 88.2
- scipy: 87.7
- matplotlib: 89.4
- scikit-learn: 85.1
- scikit-image: 89.3
- statsmodels: 94.4
- pandas: 81.0
- h5py: 83.1
- cython: 93.0
- sympy: 90.8
Overall percentage 88.5
Commits by country analysis
The table at the top of this post is the output from
commit_analysis.py in the analysis repository.