SamplingAlgorithms

Configuration sampling algorithms with constraints. Till now, we have collected the following 9 sampling strategies.

1. One-enabled Sampling

One-enabled sampling strategy enables one features at each time and disables the other features with the constraints. For features A, B, C and D we can get 4 samples like that,

[A, !B, !C, !D]
[!A, B, !C, !D]
[!A, !B, C, !D]
[!A, !B, !C, D]

2. One-disabled Sampling

One-disabled sampling strategy disables one features at each time and enables the other features with the constraints. For features A, B, C and D we can get 4 samples like that,

[!A, B, C, D]
[A, !B, C, D]
[A, B, !C, D]
[A, B, C, !D]

3. Most-enabled-disabled Sampling

Most-enabled-disabled collects 2 configurations which either has the most enabled features or has the most disabled features. For features A, B, C and D we can get 2 samples like that,

[!A, !B, !C, !D]
[A, B, C, D]

4. All-one-enabled Sampling

All-one-enabled sampling strategy enables one features at each time and disables the other features with the constraints. Note: All-one-enabled strategy differs from One-enabled strategy that the former collects all valid configurations while the latter only obtains the first satisfied one.

5. All-one-disabled Sampling

All-one-disabled sampling strategy disables one features at each time and enables the other features with the constraints. Note: All-one-enabled strategy differs from One-disabled strategy that the former collects all valid configurations while the latter only obtains the first satisfied one.

6. All-most-enabled-disabled Sampling

All-most-enabled-disabled collects 2 kinds of configurations which either has the most enabled features or has the most disabled features. Note: All-most-enabled-disabled strategy differs from Most-enabled-disabled strategy that the former collects all valid configurations while the latter only obtains the first satisfied one.

7. Random Sampling

Random sampling strategy randomly selects several configurations that satisfied constraints, which can be considered as a robust baseline strategy.

Note: this strategy needs the variable num (configurations to be selected).

8. T-wise Sampling

T-wise sampling strategy generates configurations to cover all the combination among T features. Note: Due to the limitation of computation, we only choose T=1, 2, 3, 4. The dependent tool we used for T-wise algorithm is SPLCAT, which can be found under lib/ directory.

9. Dissimilarity Sampling

Dissimilarity sampling strategy uses evolutionary algorithm to generate an optimal configurations to cover the feature interaction of T-wise strategy. The biggest advantage of Dissimilarity strategy is that it generates configurations in a heuristic way which can avoid the expensive computation in T-wise strategy. The dependent tool we used for Dissimilarity algorithm is PLEDGE, which can be found under lib/ directory.