randomTrips.py: avoid directional bias

[namdre]

for any given origin, the expected destination will be towards the center of mass of the network (or it's fringe). While this is plausible for whole-city networks, it may introduce severe bias in partial-city networks. The effect becomes noticeable when using the routed random trips as routeSampler input.

A possible solution would be to generate random coordinates rather than edges and then map these coordinates to the network. This way destinations in a random direction and distance from the origin could be sampled to avoid directional bias. A simple nearest-neighbor mapping would introduce other biases that need to be solved:

• fringe edges that protrude further than the rest would receive a huge boost in probability
• random-coordinate sampling must somehow be combined with other probability options (fringe, lanes, speed etc)

Possible solution:

1. generated n destination edges with configured probability
2. generated random coordinate
3. among the n destinations, pick the one that is closest to the coordinate

This will not eliminate all bias but the bias reduction could be worth the additional effort.

Idea for a test case: Vehicles entering a corridor from side roads. Entering vehicles should turn left and right along the corridor with similar probability regardless of proximity to either end of the corridor. (Without correction, vehicles entering near the ends would be biased towards the center of the corridor).

[namdre]

Idea for a systematic evaluation: generate randomTrips for a large network as well as for sub-networks of that large network. The distribution of angles and trip distances originating on each edge in the sub-network should match the distribution when applying cutRoutes to the large-network routes