Skip to content
Home / Catalog / Levin Tree Search with Context Policies on Symbolic Grid Problems

Levin Tree Search with Context Policies on Symbolic Grid Problems

  • Task ID: math.levin_context_grid_search
  • Domain: math
  • Subdomain: search_algorithms
  • Status: test
  • Tags: levin_tree_search, context_model, policy_guided_search, product_of_experts, deterministic_planning

Public Summary

This page is generated from task metadata and selected public-safe excerpts.

Example B1 Prompt Excerpt

# Task
You are given `data/training_levels.json` and `data/public_eval_levels.json`.
Each level is a deterministic single-agent symbolic grid-search problem with actions `U`, `D`, `L`, `R`.
The training levels include solution trajectories. The public evaluation levels are for local testing only; final scoring uses hidden levels from the same generator.
Write `analysis.py` implementing a policy for Levin Tree Search with context models, following the main line of the Levin Tree Search with Context Models paper.
The scorer will train/load your policy from the training trajectories, then run its own Levin Tree Search engine on hidden levels. It counts actual node expansions, so self-reported statistics are ignored.
Recommended route:
- Extract active contexts from each search state, such as the current cell marker, neighboring symbols, previous action, recent action history, depth buckets, and legal-action set.

Notes

  • This page is a generated site artifact.
  • Higher-level prompt details and internal benchmark specifics may remain intentionally undisclosed.