Version 0.1.2#

Simplified Installation#

There are now wheels on PyPI for Windows, MacOS, and Linux including system-specific precompiled Stan models and a pruned CmdStan distribution. This CmdStan distribution contains all binaries and libraries required to run the models and brings two advantages:

Previously, a full CmdStan installation was required as part of Gloria’s setup. This process took several minutes and consumed around 1 GB of disk space. The new pruned version is now installed automatically with Gloria and only takes a few 10 MB. This drastically simplifies Gloria’s installation process.
Each model previously had to be compiled once on the first use by Gloria, which occasionally caused runtime errors on some machines (#65). Now, all models are precompiled, eliminating the need for runtime compilation.

The packaging setup is based on this cookiecutter template by Brian Ward.

Bugfixes#

Gloria.plot() with show_capacity set to True now correctly uses the internal state self.vectorized to distinguish between vectorized and constant capacities (#60).
Fixed a bug in Gloria.plot_components() that caused a mismatch between number of ticks and labels on the timestamp axis (#66)
Default values for automatic capacity calculation in Gloria.fit() are now set to capacity_mode="scale" and capacity_value=0.5 (#73)
Fixed a bug that caused a ValueError in Gloria.predict() for the beta-binomial model due to a shape mismatch between the distribution’s scale parameter and the number of prediction timestamps (#74)

Documentation#

Release notes page added to the docs.
Corrected Saturation tutorial by removing references to the non-existent. capacity_mode="vectorized" and aligning prose to code blocks (#59).
Removed the CmdStan installation section from the installation tutorial.

Contributors#

Benjamin Kambs