|
anpcpp 0.1.0
Analytic Network Process computational library
|
Runnable console demos live in the repository examples/ directory. Build them with -DANPCPP_BUILD_EXAMPLES=ON, then run the binaries under build/examples/.
Each program prints model inputs, intermediate matrices, and final priorities. Shared formatting helpers are in examples/anp_print.hpp (not a demo itself).
| Demo | Teaches |
|---|---|
| tree134 | AHP hierarchy; hierarchy short-circuit for the limit matrix |
| network23 | Full feedback ANP; iterative calculus limit matrix |
| hamburger_std | Larger SuperDecisions reference market-share network |
| benefits_costs_subnet | Control network with subnetworks and inverted Costs |
| ratings_demo | Categorical and numeric ratings columns toward Alternatives |
| export_sample_models | Writes starter JSON models under a chosen directory (see ANP Studio samples/) |
Source: examples/tree134.cpp
A classic three-level AHP tree:
The goal connects to the three criteria; each criterion connects to the shared alternatives cluster. The scaled supermatrix is a strict hierarchy (nilpotent), so the library resolves the limit matrix with the hierarchy formula rather than power iteration.
What to notice: criteria importance wrt the goal, then local rankings of cars under Cost / Quality / Style, and the final global priorities for the alternatives.
Source: examples/network23.cpp
A small fully connected ANP network with feedback:
Criteria depend on alternatives and vice versa, with inner dependence inside each cluster. The scaled supermatrix therefore has cycles, so the limit matrix uses the iterative calculus method.
What to notice: cluster-weight comparisons between Criteria and Alternatives, feedback judgments from each alternative into the criteria, and that results differ from a pure hierarchy of the same local scores.
Source: examples/hamburger_std.cpp
SuperDecisions sample hamburger market-share ANP network (Creative Decisions Foundation tutorial tables). Local priority blocks from the published unweighted supermatrix and cluster-weight matrix are turned back into pairwise ratios a_ij = p_i / p_j.
Clusters include Alternatives (McDonalds, BurgerKing, Wendys), Advertising, Quality of Food, and Other.
What to notice: synthesized alternative priorities near the published tutorial targets (approximately McDonald's 0.55, Burger King 0.28, Wendy's 0.17), as a cross-check against SuperDecisions / reference behavior.
Source: examples/benefits_costs_subnet.cpp
A Benefits/Costs-style control model with subnetworks (BOCR-like pattern):
Each control node owns its own subnetwork that ranks the same alternatives. Scores are synthesized to the top level, weighted by control-criteria priorities. The Costs node is marked inverted so higher cost hurts the final score.
What to notice: separate Benefits and Costs subnet results, then the synthesized top-level alternative vector that mixes both (with inversion on Costs).
Source: examples/ratings_demo.cpp
A small hierarchy that uses RatingsPrioritizer instead of pairwise under the criteria:
What to notice: Price favors A (High) over C (Low); Quality favors C (100) over A (40) after divide-by-max; equal criteria weights from Best; L1-normalized rating columns feed the unscaled supermatrix the same way pairwise priorities do.
1.9.8