mikkokotila
commented
6 years ago This is more or less related with #84.
For now I've added stratified sampling, latin hypercube sampling, and latin hypercube sampling with sudoku like constraint. There is also the option for first initializing with a uniform random shuffle. These would be controlled through Scan() parameters:
grid_downsample as a fraction of the total search space
random_method either uniform, stratified, lhs, or lhs_sudoku
shuffle true or false
Right now the way the random pick is performed, is based on an index number (each index represents a single permutation). We need to have the option (and I think it should be the default option) for multi-column stratification instead.