Solver Registry¶

Virne selects solvers through solver.solver_name. For example:

virne solver.solver_name=nrm_rank

How to Choose a Solver¶

Start with a solver that matches what you want to measure:

nrm_rank

Fast baseline. Deterministic and suitable for a first end-to-end run. Use it in the Quickstart.

random_rank

Random baseline. Useful for sanity checks and comparison context. Set experiment.seed for reproducible runs.

mip

Exact-method baseline. Best for small instances. It requires OR-Tools and uses the implementation’s current 10-second solve limit.

ppo_dual_gat+

RL pipeline example. Requires learning dependencies and training. See the RL Pipeline.

For a first comparison, run nrm_rank and random_rank on the same scenario and seed. Add mip only when the instance is small enough, and use ppo_dual_gat+ when the experiment specifically evaluates learning-based methods.

Complete Registry¶

The tables below are generated from SolverRegistry during the Sphinx build. They list the commands that are actually registered by the current code, including dynamically generated reinforcement-learning variants.

Exact solvers

Command	Implementation
`mip`	`virne.solver.exact.mip.MipSolver`

Rounding solvers

Command	Implementation
`d_round`	`virne.solver.exact.d_rounding.DeterministicRoundingSolver`
`r_round`	`virne.solver.exact.r_rounding.RandomizedRoundingSolver`

Heuristic solvers

Command	Implementation
`ffd_joint_pr`	`virne.solver.heuristic.joint_pr.FFDJointPRSolver`
`ldg`	`virne.solver.heuristic.ldg.LDGSolver`
`order_joint_pr`	`virne.solver.heuristic.joint_pr.OrderJointPRSolver`
`order_rank_bfs`	`virne.solver.heuristic.bfs_trials.OrderRankBfsSolver`
`random_joint_pr`	`virne.solver.heuristic.joint_pr.RandomJointPRSolver`
`random_rank_bfs`	`virne.solver.heuristic.bfs_trials.RandomRankBfsSolver`
`rw_rank_bfs`	`virne.solver.heuristic.bfs_trials.RandomWalkRankBfsSolver`

Node-ranking solvers

Command	Implementation
`ffd_rank`	`virne.solver.heuristic.node_rank.FFDRankSolver`
`grc_rank`	`virne.solver.heuristic.node_rank.GRCRankSolver`
`nea_rank`	`virne.solver.heuristic.node_rank.NEARankSolver`
`nrm_rank`	`virne.solver.heuristic.node_rank.NRMRankSolver`
`order_rank`	`virne.solver.heuristic.node_rank.OrderRankSolver`
`pl_rank`	`virne.solver.heuristic.node_rank.PLRankSolver`
`random_rank`	`virne.solver.heuristic.node_rank.RandomRankSolver`
`rw_rank`	`virne.solver.heuristic.node_rank.RandomWalkRankSolver`

Meta-heuristic solvers

Command	Implementation
`aco_meta`	`virne.solver.meta_heuristic.ant_colony_optimization_solver.AntColonyOptimizationSolver`
`ga_meta`	`virne.solver.meta_heuristic.genetic_algorithm_solver.GeneticAlgorithmSolver`
`pso_meta`	`virne.solver.meta_heuristic.particle_swarm_optimization_solver.ParticleSwarmOptimizationSolver`
`sa_meta`	`virne.solver.meta_heuristic.simulated_annealing_solver.SimulatedAnnealingSolver`
`ts_meta`	`virne.solver.meta_heuristic.tabu_search_solver.TabuSearchSolver`

Unsupervised-learning solvers

Command	Implementation
`gae_clustering`	`virne.solver.learning.unsupervised_learning.gae_clustering.gae_clustering.GaeClusteringSolver`
`hopfield_network`	`virne.solver.learning.unsupervised_learning.hopfield_network.hopfield_network_solver.HopfieldNetworkSolver`

Reinforcement-learning solvers

Command	Implementation
`a2c_att+`	`virne.solver.learning.reinforcement_learning.solver_maker.a2c_att+`
`a2c_cnn+`	`virne.solver.learning.reinforcement_learning.solver_maker.a2c_cnn+`
`a2c_deep_edge_gat_mlp+`	`virne.solver.learning.reinforcement_learning.solver_maker.a2c_deep_edge_gat_mlp+`
`a2c_dual_deep_edge_gat+`	`virne.solver.learning.reinforcement_learning.solver_maker.a2c_dual_deep_edge_gat+`
`a2c_dual_gat+`	`virne.solver.learning.reinforcement_learning.solver_maker.a2c_dual_gat+`
`a2c_dual_gcn`	`virne.solver.learning.reinforcement_learning.dual_gnn_solver.A2CDualGcnSolver`
`a2c_dual_gcn+`	`virne.solver.learning.reinforcement_learning.solver_maker.a2c_dual_gcn+`
`a2c_gat_mlp+`	`virne.solver.learning.reinforcement_learning.solver_maker.a2c_gat_mlp+`
`a2c_gcn_mlp+`	`virne.solver.learning.reinforcement_learning.solver_maker.a2c_gcn_mlp+`
`a2c_mlp+`	`virne.solver.learning.reinforcement_learning.solver_maker.a2c_mlp+`
`a3c_att+`	`virne.solver.learning.reinforcement_learning.solver_maker.a3c_att+`
`a3c_cnn+`	`virne.solver.learning.reinforcement_learning.solver_maker.a3c_cnn+`
`a3c_deep_edge_gat_mlp+`	`virne.solver.learning.reinforcement_learning.solver_maker.a3c_deep_edge_gat_mlp+`
`a3c_dual_deep_edge_gat+`	`virne.solver.learning.reinforcement_learning.solver_maker.a3c_dual_deep_edge_gat+`
`a3c_dual_gat+`	`virne.solver.learning.reinforcement_learning.solver_maker.a3c_dual_gat+`
`a3c_dual_gcn+`	`virne.solver.learning.reinforcement_learning.solver_maker.a3c_dual_gcn+`
`a3c_gat_mlp+`	`virne.solver.learning.reinforcement_learning.solver_maker.a3c_gat_mlp+`
`a3c_gcn_mlp+`	`virne.solver.learning.reinforcement_learning.solver_maker.a3c_gcn_mlp+`
`a3c_mlp+`	`virne.solver.learning.reinforcement_learning.solver_maker.a3c_mlp+`
`a3n_gcn_mlp`	`virne.solver.learning.reinforcement_learning.gnn_mlp_solver.A3cGcnSolver`
`conal`	`virne.solver.learning.reinforcement_learning.hetero_gnn_solver.solver.ConalSolver`
`conal_wo_afb`	`virne.solver.learning.reinforcement_learning.hetero_gnn_solver.solver.CONALWoAfbSolver`
`conal_wo_pc`	`virne.solver.learning.reinforcement_learning.hetero_gnn_solver.solver.CONALWoPcSolver`
`conal_wo_reach`	`virne.solver.learning.reinforcement_learning.hetero_gnn_solver.solver.CONALWoReachSolver`
`ddpg_mlp+`	`virne.solver.learning.reinforcement_learning.solver_maker.ddpg_mlp+`
`double_dqn_mlp+`	`virne.solver.learning.reinforcement_learning.solver_maker.double_dqn_mlp+`
`dqn_mlp+`	`virne.solver.learning.reinforcement_learning.solver_maker.dqn_mlp+`
`mcts`	`virne.solver.learning.reinforcement_learning.mcts_solver.mcts.MctsSolver`
`pg_att+`	`virne.solver.learning.reinforcement_learning.solver_maker.pg_att+`
`pg_cnn`	`virne.solver.learning.reinforcement_learning.cnn_solver.PgCnnSolver`
`pg_cnn+`	`virne.solver.learning.reinforcement_learning.solver_maker.pg_cnn+`
`pg_cnn2`	`virne.solver.learning.reinforcement_learning.cnn_solver.PgCnn2Solver`
`pg_deep_edge_gat_mlp+`	`virne.solver.learning.reinforcement_learning.solver_maker.pg_deep_edge_gat_mlp+`
`pg_dual_deep_edge_gat+`	`virne.solver.learning.reinforcement_learning.solver_maker.pg_dual_deep_edge_gat+`
`pg_dual_gat+`	`virne.solver.learning.reinforcement_learning.solver_maker.pg_dual_gat+`
`pg_dual_gcn+`	`virne.solver.learning.reinforcement_learning.solver_maker.pg_dual_gcn+`
`pg_gat_mlp+`	`virne.solver.learning.reinforcement_learning.solver_maker.pg_gat_mlp+`
`pg_gcn_mlp+`	`virne.solver.learning.reinforcement_learning.solver_maker.pg_gcn_mlp+`
`pg_mlp`	`virne.solver.learning.reinforcement_learning.mlp_solver.PgMlpSolver`
`pg_mlp+`	`virne.solver.learning.reinforcement_learning.solver_maker.pg_mlp+`
`pg_mlp2`	`virne.solver.learning.reinforcement_learning.mlp_solver.PgMlp2Solver`
`ppo_att`	`virne.solver.learning.reinforcement_learning.att_solver.PpoAttSolver`
`ppo_att+`	`virne.solver.learning.reinforcement_learning.solver_maker.ppo_att+`
`ppo_cnn+`	`virne.solver.learning.reinforcement_learning.solver_maker.ppo_cnn+`
`ppo_deep_edge_gat_mlp+`	`virne.solver.learning.reinforcement_learning.solver_maker.ppo_deep_edge_gat_mlp+`
`ppo_dual_deep_edge_gat+`	`virne.solver.learning.reinforcement_learning.solver_maker.ppo_dual_deep_edge_gat+`
`ppo_dual_gat+`	`virne.solver.learning.reinforcement_learning.solver_maker.ppo_dual_gat+`
`ppo_dual_gcn`	`virne.solver.learning.reinforcement_learning.dual_gnn_solver.PpoDualGcnSolver`
`ppo_dual_gcn+`	`virne.solver.learning.reinforcement_learning.solver_maker.ppo_dual_gcn+`
`ppo_gat_mlp+`	`virne.solver.learning.reinforcement_learning.solver_maker.ppo_gat_mlp+`
`ppo_gat_seq2seq+`	`virne.solver.learning.reinforcement_learning.gnn_seq2seq_solver.gnn_seq2seq_solver.PpoGatSeq2SeqSolver`
`ppo_gcn_mlp+`	`virne.solver.learning.reinforcement_learning.solver_maker.ppo_gcn_mlp+`
`ppo_gcn_seq2seq+`	`virne.solver.learning.reinforcement_learning.gnn_seq2seq_solver.gnn_seq2seq_solver.PpoGcnSeq2SeqSolver`
`ppo_hetero_gat+`	`virne.solver.learning.reinforcement_learning.hetero_gnn_solver.solver.PpoHeteroGatSolver`
`ppo_hetero_gat_no_cl+`	`virne.solver.learning.reinforcement_learning.hetero_gnn_solver.solver.PpoHeteroGnnNoContrastSolver`
`ppo_hetero_gnn_simple_reward`	`virne.solver.learning.reinforcement_learning.hetero_gnn_solver.solver.SimpleRewardPpoHeteroGnnSolver`
`ppo_mlp+`	`virne.solver.learning.reinforcement_learning.solver_maker.ppo_mlp+`
`safe_ppo_gcn`	`virne.solver.learning.reinforcement_learning.safe_rl_solver.solver.SafePpoGcnSolver`