Airspace-Encounter-Models / em-model-manned-bayes

Manned Bayesian Network Encounter Models
BSD 2-Clause "Simplified" License
14 stars 7 forks source link

[ENHANCEMENT] Update dediscretize to support user defined sampling (not just uniform) from a parameters bin #38

Open aweinert-MIT opened 2 years ago

aweinert-MIT commented 2 years ago

Relation

The function, dediscretize samples a value within the inputted parameters bin and currently enforces uniform sampling within a bin. This function is called by dbn_hierarchical_sample to create values based on the sampled parameter bin. For example when using the uncorrelated conventional aircraft encounter models, such as uncor_1200only_fwse_v1p2, if the altitude layer (L) parameter bin is 1, then dediscretize will sample a value from the 1st altitude bin. Currently the first altitude bin is [50, 500) feet AGL, so dediscretize will uniformly sample a value, U(50, 500]

Description

Update dediscretize to enable non-uniform sampling within a bin. A good first start would be implementing a Gaussian sampling option, followed by functionality to enable the user to input their own sampling function.

Additional

dediscretize is called by lines 25 – 37 in dbn_hierarchical_sample.m. Lines 25-31 is for the sample from the initial network while lines 33 – 37 are the events (i.e., for the uncorrelated conventional aircraft models: vertical rate, turn rate, acceleration) for each timestep.

This issue will also be heavily reviewed by the airspace encounter models team. It is important that the models not be misused to sample and create tracks, and subsequently encounters, that intentionally skew the encounter geometry to be easier for a collision avoidance system to resolve.