Closed chengsoonong closed 7 years ago
We can get rid of the length, which is a simple constraint, and add other constraints, e.g. time, budget. However, it is not clear how to model the time and cost/budget. If the query contains only the start point, and without any other constraints, then we probably need to scoring trajectories such that scores are normalised over length.
Close this issue as we won't fix it.
Right now a query is a (start POI, length) tuple, and hence SSVM only has to predict a sequence of the given length. An alternative is a query that only is (start POI). Before implementing this, we need to check that the scores of the SSVM are comparable for predictions of different length.