From: Stefan F. <ste...@we...> - 2010-04-20 06:03:43
|
Hi John, I believe that Alex already is considering another questions than you raise: Yes we both optimize all trains simultaneously. To give a precise answer my possible tree extensions at each station vertex are the following: 1) Traverse each edge to the neighbor vertex 2) If head train: start bottom train from start token 3) Start next (head) train 4) If vertex with value then evaluate train set Each head train starts at the virtual HQ to each of the available (start) tokens. This explains the huge increase in evaluations if two trains are run, instead of one. 8 train alone: without prediction: 3.8k evaluations, 16.4k edges, time: < 1 sec with prediction: 62 evaluations, 858 predictions, 2.3k edges, < 1 sec 8 and 10 trains: without prediction: 1.9M evaluations, 11.0M edges, time: 90 sec with prediction: 25k evaluations, 143k predictions, 387k edges, time: 4 sec I can generate an example, where trains are run sub-length, if you are still in doubt. Stefan > > I am just saying if you choose the best route for one train first, and then > find the best route for remaining trains, you won't find the optimal run > for all the trains. > |