|
The solutions to all our CSP's must conform with the following general requirement:
- We require that solutions to the CSP be uniformly-random. By this we mean that if we run any of the testcase generators many times on the same input problem, each time, the generator will produce a different testcase that satisfies all constraints. Ideally, these testcases should be distributed uniformly in the space of all legal testcases (i.e., all tests that satisfy the input CSP).
In addition, we list some of the particular properties characterizing our problems in general:
- Variables with huge domains (e.g., 2^64)
- Global constraints representing complex relations between a number of variables
- Conditioned variables and constraints, whose existence in the problem depend on the assigned value of some of the variables (CondCSP)
- Similar sub-CSP's within the same problem, with additional constraints forcing relations between the parameters of each of the sub-problems (parameterized CSP). The number of sub-CSP's is itself a CSP variable
- Several levels of soft constraints used to bias the test to the most interesting legal scenarios (hierarchical CSP)
| |
|
