greedy_value_model

defender · baseline · Elo 1020 · wins 72%
greedy_value_model

What it says it does

Greedy value model: annealed_assignment's value model with a greedy picker instead of the annealer. It exists so the ladder shows what the optimiser is worth on its own.

Idea: same energy terms, no search. Launch at the best-valued tracks one by one (skipping any track inside the blast radius of one already picked) up to the launch cap, whenever the net value is positive; fire the gun at the best net-valued track in range. build(), energy() and every knob are identical to annealed_assignment.py, and a test keeps them identical.

The annealed_assignment docstring describes the model in full.

Against each opponent

opponentits mean score
rush0.796
parked_ring_magazine_drain0.794
attacker_blast_isolated_synchronized_rel0.790
flanker0.788
decoy_screen0.771
staggered_screen_trickle0.722
multi_axis0.682
trickle0.645
shadow_pair_ring0.622
depth_shell_saturation0.483
staggered_trishell_blast_denial0.336
interleaved_shell_ring0.144
synchronized_ring0.138
phase_locked_ring0.115

Ten most recent matches

From the latest nightly tournament.

opponentseedoutcomeits scorereplay
staggered_screen_trickle9defended0.746watch
synchronized_ring9breached0.122watch
phase_locked_ring9breached0.122watch
depth_shell_saturation9defended0.769watch
interleaved_shell_ring9breached0.139watch
parked_ring_magazine_drain9defended0.787watch
staggered_trishell_blast_denial9breached0.156watch
attacker_blast_isolated_synchronized_release9defended0.808watch
shadow_pair_ring9defended0.767watch
decoy_screen9defended0.770watch