7Drop7 Research
LearnApproachesEnginesResearchLeaderboard

Type to search every page, record and vocabulary term.

GitHub7Play
LearnApproachesEnginesResearchLeaderboardGitHub7Play

Approaches

Each page here is one theory of how to choose a column, grouped by the technique it uses; the engines that play the games and the instruments that measure them live under Engines and Diagnostics.

Techniqueallexpectimaxheuristic-evaluationq-learningn-tuplennuepolicy-gradientevolutionmctsrollout-policy-iterationoracle-distillationrisk-survivalafterstateconstructive-planningdeterminization
Statusallcompletedrejectedruntime-pausedpreregisteredsupport-onlyproposal
Readsallpublicoracle/teacherdiagnostic
Group bytechniquefamily

5 approaches of 87

Evolutionary optimisation

Tune a player's weights by playing complete games, keeping the settings that scored best, and repeating.

Read the primer →
fitnessfair leafwarm startshort12 candidates, generation by generation
Evolutionary optimisationfeatured

An evolved NNUE leaf, distilled from a depth-5 teacher

A compact neural leaf evaluator for the depth-3 fair search, warm-started by distilling a depth-5 teacher's sibling-complete root values and then refined by an evolutionary search whose fitness is whole-game score itself.

rejectedrecorded
→
populationelitesmutate
Evolutionary optimisationfeatured

Evolution

Let a search tune board-feature weights by playing thousands of games, keeping the best weights, and repeating.

rejectedrecorded
→
32576663416241524362+7weighted sumno search
Evolutionary optimisation

Direct public policy

Skip the value model. Score each column with 25 weighted features and tune the weights against whole games.

rejected
→
332647466561332245136scorelevelseedpublic
Evolutionary optimisation

Evolved public policy

Give the weight search a larger phase-aware evaluator and optimise complete games directly, without imitation or proxy labels.

rejectedrecorded
→
sampleselitecoefficientrefit
Evolutionary optimisation

Machine-tuning the board evaluator

Let an optimiser adjust eight coefficients of the search's board evaluator by playing complete games, then freeze the winner and test it on games it never saw.

rejectedrecorded
→

No approach page matches that search.

7Drop7 Research

An open, reproducible research project working toward a public-information Drop7 policy that averages more than one million points.

Privacy-first analytics

Play and learn

  • Play Drop7
  • How Drop7 works
  • Strategy concepts
  • Human competition
  • Leaderboards

Research

  • Research overview
  • Approaches
  • Engines
  • Techniques
  • Theories
  • Experiments
  • Results
  • Diagnostics
  • Research log
  • Documents

Project

  • App support
  • Browse the source
  • Documentation
  • GitHub repository ↗
  • Issues and discussion ↗

Independent research project. Not affiliated with the creators or owners of Drop7.

Privacy policyTerms of service