---

name: objective-functions-and-speedups description: "Shape scikit-opt objectives, benchmark functions, and safe run-mode speedups." disable-model-invocation: true metadata: disco-role: operating license: MIT

objective-functions-and-speedups

Use this sub-skill when a task needs objective-function contracts, built-in sko.demo_func benchmarks, set_run_mode acceleration/cache choices, or optional plotting/GPU dependency guidance for scikit-opt / sko.

Route here when

• An optimizer fails because the objective receives the wrong shape or returns the wrong shape.
• You need to convert a scalar objective to vectorization mode, cached mode, or a safe thread-based mode.
• You need a quick built-in benchmark such as sphere, schaffer, rosenbrock, or ackley.
• You need to explain matplotlib, pandas, joblib, torch, or CUDA as optional dependencies rather than core requirements.

Route elsewhere

• Algorithm-specific parameters, bounds, constraints, continuation, and history interpretation belong to the owning algorithm sub-skill.
• GA integer/precision encoding and the x2gray relationship belong to ../genetic-algorithms/.
• TSP route-cost design and permutation outputs belong to ../routing-and-combinatorial/.

Bundled references

• Read references/objective-contracts.md before writing or adapting objective functions for scalar, vectorized, cached, threaded, or method-based use.
• Read references/acceleration.md when choosing set_run_mode(func, mode), deciding whether to use common, vectorization, cached, multithreading, multiprocessing, joblib, or interpreting the parallel alias.
• Read references/demo-functions.md when selecting built-in benchmark/demo objective functions or using function_for_TSP as a tiny route fixture generator.
• Read references/troubleshooting.md when run-mode assertions, vectorized shape errors, cached input errors, multiprocessing/joblib failures, plotting display problems, or optional GPU issues appear.

Bundled script

• Run scripts/smoke_run_modes.py to check that the installed package can execute tiny GA objectives in common, vectorization, cached, and multithreading modes without source-checkout data, plotting, network, or GPU requirements.

Fast operating rules

1. Set the run mode before constructing the optimizer; constructors wrap the function immediately.
2. Use scalar objectives for common, multithreading, multiprocessing, cached, and joblib: one candidate vector in, one finite scalar out.
3. Use vectorized objectives only after adapting the function to accept an X matrix and return one value per row.
4. Treat multiprocessing, joblib, plotting, pandas workflows, and GA.to(device) as optional surfaces with extra dependency/platform checks.