Release Notes for PySAL2.0rc1

For starters, these change notes are only for this release candidate, which is like a typical release of PySAL, but considered a preview. This release candidate, 2.0rc1, signifies that our API will change (the 2.0 part) and that we want to ensure users can opt-in to these changes with a longer time delay than a usual release (the rc1 part). We will make a full release of PySAL 2.0 by 2018-12-31, which will largely follow the API reorganization in this release candidate (and any subsequent release candidates). For more information on how to migrate to the 2.0-API, please check out migrating.pysal.org.

Feature Changes in 2.0rc1

This release, we've had a ton of activity in PySAL, but it has all been conducted in our subpackages, the independently-released components of our library. Because it is too onerous to list all the changes to the API here, please consult the migrating.pysal.org website for a detailed discussion of all reorganization-related changes. This document will focus only on improvements to functionality, enhancements, and additional modules added to 2.0rc1, over and above the last stable release of the 1.0 series, PySAL 1.14.4.

Overall, there were 719 commits that closed 240 issues, together with 105 pull requests across 12 packages since our last release on 2017-11-03.

Entirely New Packages

For starters, we've added some entirely-new subpackages to this release candidate:

• mgwr, the multi-scale Geographically-weighted regression package for Python.

  Don't worry, mgwr fits single-scale geographically-weighted regressions, too. Geographically-weighted regression is a kind of generalized additive model that uses kernel functions in the geographic area around each observation to predict outcomes at that area more accuratly, kind of like Gaussian Process regression for geographic data.

• spvcm, for spatially-correlated multilevel models.

  Spatially-correlated multilevel models are models that allow for random effects of nearby areas, regions, or groups to be correlated with one another. This is a Gibbs sampling framework plus diagnostics & plotting tools for general Bayesian analysis of Gibbs samplers. The package also contains simple tools to implement new samplers on top of the infrastructure provided, which is fast, parallel, serializable, iterative-write, and interruptible.

• spint, for estimating spatial interaction models, such as the production-constrained or consumption-constrained gravity models.

• spglm, a package for fitting sparse GLMs, focused on performance over sparse categorical data.

• splot, for spatial vizualization in Python, built on top of the excellent geopandas. This is headed by our Google Summer of Code (2018) student, Stefanie Lumnitz, and will be ongoing throughout the release candidate maturation cycle.

• pointpats, a package for the statistical analysis of point patterns, geographical colocation, and dispersion.

Significant Enhancements to Existing Packages

We've also had a ton of activity adding new features in our submodules:

• alpha shapes, a generalization of convex hulls to allow for non-convex hulls
• Contiguity weights from voronoi tesselations/delaunay triangulation graphs for point patterns
• Fuzzy contiguity weights, for polygonal data where planarity is not strictly enforced (provided in part by geopandas)
• A family of spatially explicit indicators of mobility association including the inter- and intra-regional indicators as well as the local indicators. The former provide a meso-level view on the exchange mobility pattern, shedding light on specific regions hosting frequent rank exchanges while the latter examine the concordance relationship between a focal spatial unit and its neighbors for revealing the role of space in shaping the exchange mobility pattern over time.
• Categorical Spatial Markov, a generalization of the Spatial_Markov class previously implemented to interrogate the role of space in shaping regional income distribution dynamics to incorporate directe variables. Here, the categorical spatial lag is adopted for formalizing the neighborhood.
• Connected Components checks for our spatial graph matrices
• Pandas Adjacency list constructors and NetworkX adapters for PySAL graph objects
• Attach islands, tooling to merge isolated observations into existing spatial graphs
• Construct Contiguity graphs from linestrings, allowing for easy and fast access to road network data in PySAL
• anything else?

Contributor Statistics

Below, you'll find some contribution statistics from users, how active each subpackage is, and how active users are within each subpackage:

Developer commits

Subpackage Activity

Committers by Subpackage

tags: pysal, changelog, release candidate, spatial statistics, statistics, networks, graphs