MATSim Code

more efficient implementation

implement a custom population parser

- storing the population requires to much memory on a desktop
- even using the "streaming" setting of PopulationImpl remains very
slow, because of the numerous call to String.intern() to construct
activities. This is rather anoying, as this information is not needed
anywhere here.

correction of probability for second round

do not regenerate primary ties when calibrating secondary threshold

HashMap -> LinkedHashMap

remove the Tie class

there are actually only the map-based methods of SocialNetwork which are
really useful, and the Tie list took a huge amount of memory leading to
filling the heap with dense networks.

use object pool for coords

make thresholds match the returned network at the end of iteration

express "secondary threshold" in terms of "reduction"

(theta in the paper). This is better for the calibration process, as
this allows the secondary threshold to follow the evolutions of the
primary threshold.

dump evolution of thresholds in a file