Skip to content

Latest commit

 

History

History
86 lines (63 loc) · 2.77 KB

CHANGELOG.rst

File metadata and controls

86 lines (63 loc) · 2.77 KB
Raw

v3.2

2020/05/05

  • Improved installation and compatibility
  • Support for Pandas 1.0
  • Various improvements and bug fixes
  • Note that active development of certain UrbanSim components has moved to stand-alone libraries in UDST: Developer, Choicemodels, UrbanSim Templates

v3.1.1

2017/5/9

  • Updated deprecated sort method for Pandas Series and DataFrames

v3.1.0

2017/5/8

  • Python 3 compatibility
  • Updated documentation
  • Various improvements and bugfixes

v3.0.0

2015/8/26

  • Remove simulation framework, which has been moved to a separate library called Orca

v2.0.1

  • Fix index of summed probabilities

v2.0.0

  • Renamed Location Choice Models to Discscrete Choice Models

    • #134

    • We generalized the existing location choice model classes into discrete choice models with varying capabilities. The urbansim.models.lcm module has been renamed to urbansim.models.dcm and model classes with LocationChoice in their name have been renamed to have DiscreteChoice instead.

    • New options are available to control the behavior of DCMs:

      • probability_mode: The probability mode can take the values 'single_chooser' and 'full_product' (default). It controls whether the probabilities used for choosing are calculated using a single chooser or separately for every chooser. The former is a useful performance optimization when there are many alternatives.
      • choice_mode: The choice mode can take the values 'individual' (default) and 'aggregate'. It controls whether choices are made one at a time for each chooser or all at once for all choosers. The latter is appropriate for something like a LCM where an alternative taken by one person is no longer available to others.
      • At the group level the remove_alts option specifies whether to remove chosen alternatives from the alternative pool between performing choices for segments. remove_alts defaults to False, but should be set to True for LCMs so that alternatives are not made available multiple times.

      The default values for these options are appropriate for fully generalized discrete choice models, but will need to be set to their non-default values to retain the behavior of the old LocationChoice classes.

  • Memoized function injectables

    • #138
    • Allows users to define a function injectable that has argument-based caching that is tied into the larger caching system.

  • Allow sampling of alternatives during prediction