Path4gmns

v0.9.4 brings crucial bug fixes in the simulation module and several enhancements.

1. Fix the calculation of agent arrival time and agent waiting time.
2. Use free-flow travel time as the traverse time of each link for the point queue model.
3. Fix the incorrect setup of link outflow capacity within initialize_simulation().
4. Remove useless memory_blocks and its implementations (which are intended for parallel computing), which removes the requirement that zone id's have to be consecutive 36.
5. Remove problematic beg_iteration and end_iteration in setting up a special event.
6. Enhance time_period parsing in setting up DemandPeriod instance.
7. Fix potential issue related to _ (underscore) in YAML (that 0700_0730 is interpreted as integer 07000730 by replacing _ (underscore) with - (hyphen) in settings.yml) and add backward compatibility for time_period with _ (underscore) as delimiter.
8. Add check on OD pair to ensure that agents of the same OD pair with the same departure time will be output for the first one in output_agent_trajectory().
9. Bring back the post-processing after UE in case users do not do column updating (i.e., column_update_num = 0).

v0.9.3 comes with bug fixes and performance improvement.

1. Fix the bug in handling link capacity reduction in UE when a special event is specified in [settings.yml](https://github.com/jdlph/Path4GMNS/blob/master/tests/settings.yml).
2. Remove dependency on demand.csv for loading columns to continue UE or initiate DTA (31).
3. Optimize colgen.py per [TransOMS](https://github.com/jdlph/TransOMS), which slightly improves the running time. See [fb3f6e4](https://github.com/jdlph/Path4GMNS/commit/fb3f6e4a6f004b1408690d1aebf12442e7e61f10) and [bbad4b1](https://github.com/jdlph/Path4GMNS/commit/bbad4b1c5e28c9c0181f5d0d7e899581f63b2110) for details.
4. Deprecate confusing find_path_for_agents() (28).

v0.9.2 improves the performance with faster and better UE convergency along with several bug fixes.

1. Optimize the column generation module per [TransOMS](https://github.com/jdlph/TransOMS), which slightly reduces the running time and greatly improves the UE convergency by three times. For the Chicago Sketch Network, the UE gaps are 0.091% (or 13789.25) and 0.0326% (or 4909.65).
2. Fix the bug in updating the total system travel time in _update_column_gradient_cost_and_flow(), which shall be updated before shifting flow.
3. Fix the potential issue in path engine where the last through node may be traversed.
4. Fix the bug in load_columns() that node path and link path shall be reversed in the loading process (31).
5. Add documentation on conducting simulation by directly loading existing UE result from agent.csv (31).

v0.9.1 features two major updates.

1. Fully **optimize** the C++ routing engine with **the MOST EFFICIENT** deque implementation of the modified label correcting (MLC) algorithm. It leads to a 4.68% of speed improvement in column generation over v0.9.0 (20 column generations only and no column optimization for the Chicago Sketch network).
2. Support **UE + DTA**. v0.9.1 connects UE with the simulation module. Now the routing decision of each agent is from UE directly. The original simulation implementation introduced in v0.9.0 (that each follows the shortest path from origin to destination) has been deprecated.

1. Link length was converted from meter to mile in link.csv for ASU in the sample data set;
2. Link path or node path is specified in the output of find_shortest_path().

v0.8.7 finalizes the changes introduced in v0.8.7a1 as a stable release with the following updates.

1. Calculate and print out relative UE gap as convergency measure.
2. Support the most common length and speed units for link.csv, which include meter (or m), kilometer (or km), mile (or mi), kmh (or kph), and mph.
3. Add more stable handling over the legacy mode 'p' or 'passenger'.
4. Include Public API to docs.
5. Print out distance unit (i.e., mi) in find_shortest_path(), get_agent_node_path(), and get_agent_link_path().