Skip to content
Source

Python Packages


Specialized Stack


Automatic Differentiation Stack

These programs are mainly focused on automatic differentiation (a.k.a. AutoGrad). Each package is backed by some big company (e.g. Google, Facebook, Microsoft or Amazon).There are many packages nowadays, each with their pros and cons, but I will recommend the most popular. In the beginning, static graphs (define first then run after) was the standard but nowadays the dynamic method (define/run as you go) is more standard due to its popularity amongst the research community. So the differences between many of the libraries are starting to converge.

Please go to this webpage for a more detailed overview of the SOTA deep learning packages in python.


Kernel Methods

So I haven't found any designated kernel methods library ( open market?!) but there are packages that have kernel methods within them. There are many packages that have GPs.


Visualization Stack


Geospatial Processing Stack

  • Geopandas

    This would be the easiest package to use when we need to deal with shape files. It also follows the pandas syntax closes with added plotting capabilities.

  • Rasterio
  • Shapely
  • rioxarray

    A useful package that allows you to couple geometries with xarray. You can mask or reproject data. I like this package because it's simple and it focuses on what it is good at and nothing else.