This repository contains code to mine Regions Of Interest from trajectory data.
To run the optimization model you need the gurobi solver and its python package gurobipy.
You can clone the repository and install the package with pip install .
The miner is separated in two part: the generation of the candidates and the optimizer. You can decide to use only the optimizer with your set of candidates, or to use the predefined grid-based MDL miner.
For a full example, you can look at the file examples/ILP-example.py.
First you need to load a grid of density and set a minimum density threshold. Then you can just run
from roi_miner.grid_miners.MDL_miner.MDL_otpimizer import mine_rois
rois = mine_rois(grid_density, density_threshold)which return a list of Shapes which are the optimals ROIs.
You can add constraints on the shapes by registering a function with its arguments. At generation time, these constraints
will be checked. So for example, if we want to add a constraint on the diameter of the Circle, we can do
from roi_miner.grid_miners.constraints import *
# Some code here ...
register_circle_constraint(circle_diameter_constraint, (2,5,))We provide some initial constraint in roi_miner.grid_miners.constraints but you can define yours.
Finally, the call to mines_rois take two optionnal parameters min_distance_rois and max_distance_rois.
These are distance constraints between the ROIs. We can not have two ROIs at a distance less than min_distance_rois and
more than max_distance_rois.
You can also only use the optimizer with your own candidates (or shapes). The signature of the optimization function is
def optimize(candidates, weights, exclusive_constraints)With candidates the set of candidates, weights their weights.
exclusive_constraints contains a list of set of candidates index that can not be taken together.