Claude Skills by hajibabaie

When the user wants to solve flow problems on a network - max-flow/min-cut, min-cost flow, multicommodity flow, or shortest paths via Dijkstra, Bellman-Ford, and label-correcting methods - using networkx, gurobipy, or specialized algorithms, including when total unimodularity makes LP solutions integral for free. Also use when the user mentions "network flow," "min-cost flow," "max flow," "multicommodity," "shortest path," "flow conservation," or when the constraints have flow-balance structu...

When the user wants to remove slow Python loops from metaheuristic or optimization code using NumPy — population-level operations, batch fitness evaluation, distance matrices, broadcasting, argsort/argpartition idioms, default_rng, and memory layout. Also use when the user mentions "vectorize," "numpy broadcasting," "population fitness," "distance matrix," "speed up metaheuristic," or "slow Python loop." For computing less instead of faster (delta evaluation, caching), see fitness-evaluation-...

When the user wants to solve LPs, MIPs, or CP models without a commercial license — choosing among HiGHS, SCIP, CBC, OR-Tools (CP-SAT and MathOpt), PuLP, Pyomo, and python-mip, comparing licenses, setting realistic performance expectations versus Gurobi, and migrating gurobipy models to an open-source stack. Also use when the user mentions "open-source solver," "HiGHS," "SCIP," "PuLP," "Pyomo," "CBC," "free solver," or "no Gurobi license." For gurobipy model construction itself, see milp-mode...

When the user wants to organize optimization research code into a reproducible project — separating the src package, scripts, configs, and results; JSON/YAML config systems; factory registration of algorithms and problems; seed discipline; atomic result writing; and light testing. Also use when the user mentions "project structure," "research code organization," "config file," "factory pattern," "reproducible runs," or "random seed," or when results can no longer be traced to the exact code a...

When the user wants to tune metaheuristic or solver parameters with Optuna, including search-space definition, TPE sampling, pruning weak configurations early, multi-instance objectives (mean/median over an instance set), and held-out validation against overtuning. Also use when the user mentions "Optuna," "hyperparameter tuning," "parameter tuning," "tune metaheuristic," "TPE sampler," "algorithm configuration," or when reported results hinge on hand-picked parameter values. For statistical ...

When the user wants to record, store, and aggregate computational-experiment results for optimization algorithms with pandas - tidy one-row-per-run tables, run metadata (instance, seed, algorithm, parameters, runtime, objective), atomic CSV/parquet writing, aggregation across instances and seeds, and pivot tables for papers. Also use when the user mentions "experiment results," "results dataframe," "aggregate runs," "experiment tracking," "results CSV," "results parquet," or when thousands of...

When the user wants to parallelize a metaheuristic or combine several searches — island models with migration, master-slave fitness evaluation, parallel multistart, cooperative search, or algorithm portfolios — with the Python practicalities (multiprocessing, joblib, vectorization-first). Also use when the user mentions "parallel metaheuristic," "island model," "migration," "master-slave," "cooperative search," "portfolio," or when single runs are too slow and CPU cores sit idle. For removing...

When the user wants to schedule jobs on one machine or on identical, uniform, or unrelated parallel machines: dispatching rules (SPT, WSPT, EDD, Moore-Hodgson), LPT and list scheduling with worst-case bounds, exact MIP models for makespan and due-date objectives, and LNS for large instances. Also use when the user mentions "parallel machines," "single machine scheduling," "minimize tardiness," "LPT," "weighted completion time," or "machine assignment." For jobs that visit several machines in ...

When the user wants to design, implement, or tune particle swarm optimization, covering velocity and position updates, inertia weight, constriction, swarm topologies, and discrete adaptations such as random-key and binary PSO. Also use when the user mentions "particle swarm," "PSO," "inertia weight," "velocity update," "swarm topology," or "binary PSO," or when a gradient-free continuous box-constrained problem needs a population method. For the often stronger vector-difference alternative, s...

When the user wants to translate a word problem or real-world planning situation into a formal optimization model — identifying decision variables, objective function, and constraints, choosing the model type (LP, MIP, CP, or heuristic), estimating model size, and auditing the data before any solver code is written. Also use when the user mentions "formulate," "model this problem," "decision variables," "objective function," "translate to optimization," or when a problem arrives as plain pros...

When the user wants to model or solve the quadratic assignment problem — assigning n facilities to n locations to minimize the total flow-times-distance cost — with linearized MIP models or robust tabu search built on O(n) swap delta evaluation. Also use when the user mentions "quadratic assignment," "QAP," "flow distance," "QAPLIB," "facility layout," or "Koopmans-Beckmann," or when pairwise interaction costs depend on a permutation. For linear assignment without interaction terms, see assig...

When the user wants to optimize under uncertainty with hard feasibility guarantees by building robust counterparts of LPs and MIPs over box, budget (Bertsimas-Sim), or ellipsoidal uncertainty sets and tuning the price of robustness. Also use when the user mentions "robust optimization," "uncertainty set," "robust counterpart," "Bertsimas-Sim," "worst case," "budget of uncertainty," or when constraint data is uncertain and any violation is unacceptable. For scenario-based expected-value models...

When the user wants to implement scatter search (reference set management, diversification generation, subset combination, improvement method) or path relinking as intensification between elite solutions, standalone or inside GRASP. Also use when the user mentions "scatter search," "path relinking," "reference set," "RefSet," "elite solutions," "solution combination," "relinking," or when a small, explicitly managed elite population should replace a large stochastic one. For greedy randomized...

When the user wants to choose, implement, or tune parent selection and survivor replacement in evolutionary algorithms — tournament, roulette-wheel, rank, stochastic universal sampling, and Boltzmann selection, plus generational vs steady-state replacement, elitism, and (mu+lambda) vs (mu,lambda) — with selection-pressure and takeover-time analysis. Also use when the user mentions "selection operator," "tournament selection," "roulette wheel," "rank selection," "elitism," "steady-state GA," o...

When the user wants to model or solve set covering, set packing, or set partitioning problems with exact MIP models, the greedy heuristic and its ln(n) guarantee, LP rounding, or Lagrangian-based heuristics in the Caprara-Fischetti-Toth style. Also use when the user mentions "set covering," "set partitioning," "set packing," "crew scheduling," "covering constraint," "winner determination," or when each row must be hit at least once, exactly once, or at most once by selected columns. For prici...

When the user wants to design, implement, or tune simulated annealing for combinatorial optimization, covering Metropolis acceptance, cooling schedules (geometric, Lundy-Mees, adaptive), initial temperature calibration, reheating, and restarts. Also use when the user mentions "simulated annealing," "cooling schedule," "acceptance probability," "initial temperature," "Metropolis criterion," or when a hill climber stalls in local optima and needs a randomized escape. For move and delta-evaluati...

When the user wants to choose or design a solution representation for a metaheuristic — binary, integer, real-valued, permutation, matrix, set-based, or mixed — and judge candidates by locality, redundancy, feasibility coverage, and bias. Also use when the user mentions "solution representation," "encoding," "permutation encoding," "binary encoding," "genotype," "representation choice," or when operators keep producing infeasible or invalid offspring. For indirect encodings and decoder design...

When the user wants to verify optimization code: independent feasibility checkers, objective recomputation separate from the solver, unit tests for constraint builders and operators, known-optimum regression tests, and exact-vs-heuristic cross-validation on small instances. Also use when the user mentions "validate solution," "feasibility checker," "test optimization code," "verify objective," "regression test optimum," or when a heuristic appears to beat a proven optimum. For repairing infea...

When the user wants to optimize decisions under uncertainty with two-stage stochastic programming — extensive (deterministic-equivalent) form in Gurobi, scenario generation and reduction, sample average approximation (SAA), and EVPI/VSS analysis. Also use when the user mentions "stochastic programming," "two-stage," "scenarios," "SAA," "recourse," "uncertainty in demand," or when decisions split into here-and-now and wait-and-see stages. For L-shaped decomposition of large scenario models, se...

When the user wants to design or implement tabu search — tabu lists and tenure, move attributes, aspiration criteria, frequency-based diversification, candidate list strategies, and elite-restart intensification. Also use when the user mentions "tabu search," "tabu list," "tabu tenure," "aspiration criterion," "candidate list," or when a local search keeps cycling back into the same local optima. For neighborhood design and delta evaluation, see local-search-and-neighborhoods; for incremental...

When the user wants to model and solve educational timetabling (course or exam scheduling) or personnel rostering problems with hard and soft constraints, using MIP, CP-SAT, or heuristic methods. Also use when the user mentions "timetabling," "course scheduling," "nurse rostering," "shift scheduling," "soft constraints," "exam scheduling," or when the problem assigns events to time slots, rooms, or shifts under conflict, coverage, and workload rules. For CP-SAT modeling idioms, see constraint...

When the user wants to model or solve the traveling salesman problem - exact MIP formulations (MTZ vs DFJ with lazy subtour-elimination cuts), construction heuristics (nearest neighbor, greedy edge, Christofides), 2-opt/3-opt/Or-opt improvement, and Lin-Kernighan-style moves. Also use when the user mentions "TSP," "traveling salesman," "subtour elimination," "tour," "2-opt," "TSPLIB," "Hamiltonian cycle," or when one vehicle must visit every node exactly once and return. For multiple vehicles...

When the user wants to design or implement variable neighborhood search and its family — VND, basic VNS, general VNS, and skewed VNS — choosing and ordering neighborhoods, designing shaking moves, and deciding when systematic neighborhood change beats a single neighborhood. Also use when the user mentions "variable neighborhood search," "VNS," "VND," "shaking," "neighborhood change," or when a local search keeps returning to the same local optimum. For move design and delta evaluation, see lo...

When the user wants to coordinate truck platoons — choosing routes with bounded detours, scheduling departures and waiting under time windows, and pairing trucks on shared arcs so followers save air-drag fuel — via an exact MIP or a metaheuristic. Also use when the user mentions "platooning," "truck platoon," "fuel savings," "convoy," "platoon coordination," or "slipstreaming," or when arc fuel cost depends on which other vehicles traverse the arc at the same time. For routing without inter-v...

When the user wants to model and solve vehicle routing problems — the capacitated VRP and its main variants (time windows, multi-depot, heterogeneous fleet, pickup-and-delivery) — via two- and three-index MIP formulations, savings and sweep construction, ALNS improvement, and OR-Tools routing. Also use when the user mentions "vehicle routing," "VRP," "CVRP," "time windows," "fleet," "delivery routes," or "Solomon instances," or when several capacity-limited routes must start and end at a depo...

When the user wants to construct an initial solution and feed it into a solver — construction heuristics by problem class, MIP starts and variable hints in Gurobi, partial fixing, and seeding metaheuristic populations from exact solution pools. Also use when the user mentions "warm start," "MIP start," "initial solution," "construction heuristic," "seed the population," "variable hints," or when the solver struggles to find any feasible solution. For callback injection and solution pools, see...