Skip to content

prog_algs v1.2
Compare
Choose a tag to compare
@teubert teubert released this 12 Nov 20:07
· 978 commits to dev since this release
v1.2.0
e38a7ef

Release v1.2

Note for existing users

This release includes changes to the return format of the MonteCarlo Predictor's predict method. These changes were necessary to support non-sample based predictors. The non backwards-compatible changes are listed below:

  • times:
    • previous List[List[float]] where times[n][m] corresponds to timepoint m of sample n.
    • new List[float] where times[m] corresponds to timepoint m for all samples
  • End of Life (EOL)/ Time of Event (ToE) estimates:
    • previous List[float] where the values correspond to the time that the first event occurs.
    • new UnweightedSamples where keys correspond to the individual events predicted.
  • State at time of event (ToE)
    • previous: element in states
    • new: member of toe event (e.g., toe.final_state['event1'])

General Improvements

  • Added new visualization capabilities, including:
    • Scatter plot for an UncertainData object (e.g., states at a point in time) [#24]
    • Histogram for an UncertainData object [#37]
  • Updates to UncertainData
    • Added method for median [#75]
    • Added percentage_in_bounds and metrics metrics as UncertainData methods [#92]
    • Added scatter and histogram charts as UncertainData methods
    • Added the 'key' method to get all samples for a specific key
  • Metrics functions can now accept UncertainData or Predictions, and operate on multiple states/events [#92]
  • Added additional examples demonstrating prog_algs features
  • Add support for Python 3.9
  • Added support for prog_models v1.1
  • General Bugfixes

State Estimator Improvements

  • Particle Filter Sample Vectorization - significantly improves runtime for vectorized models [#76]
  • Particle Filter now calculates weights in log-domain (improves numerical stability) [#76]
  • Added ability to set initial time in state estimators using the t0 parameter
  • Fixed bug where states were being reordered in UKF
  • Fixed bug where t wasn't being set each step of PF

Predictor Changes

  • New Unscented Transform Predictor [#40, #74]
  • Predictors can now predict multiple events. [#81]
  • New Prediction class to represent predicted future values (e.g., states). Returned from the Predictor.predict method [#60, #25]
  • New ToEPredictionProfile class to represent and operate on the result of multiple predictions at different times of prediction [#91]

Notes

The changes in this release were produced in part by Northrop Grumman under a contributor license agreement. Thank you NGC!

Assets 2