Fix DSL parameter grids and improve solver robustness

[tylerbessire]

Summary

• expand program enumeration to include crop, pad, and recolor with valid parameter grids
• generate parameter grids for crop, pad, and recolor in neural-guided search
• add error handling in apply_program and validate neural guidance model initialization

Testing

• pytest

---

https://chatgpt.com/codex/tasks/task_e_68c251a4fae08322911fb650c174fbcc

Summary by CodeRabbit

• New Features

  • Added support for crop, pad, and recolor operations with parameterized options.
  • Expanded program generation to combine these operations, increasing the candidate solution space.

• Bug Fixes

  • Standardized, informative errors when a step fails, including operation details and position.
  • Validated model input dimensions (17) and added automatic fallback on load issues to prevent crashes and ensure consistency.

• Notes

  • No changes to the public API.

[coderabbitai]

Caution

Review failed

The pull request is closed.

Walkthrough

Adds index-aware error wrapping to DSL program application. Introduces dimension validation and safe fallback in neural guidance initialization. Extends neural parameter grids for crop, pad, recolor. Expands search enumeration to include crop, pad, recolor with concrete parameter spaces, altering program generation flow and ordering.

Changes

Cohort / File(s)	Summary
DSL error context arc_solver/dsl.py	apply_program now uses enumerate and wraps each op application in try/except, raising a ValueError with op name, index, and params on failure, chaining the original exception. No API changes.
Neural guidance init + fallback arc_solver/neural/guidance.py	Adds expected_dim=17, validates loaded model input_dim, and falls back to SimpleClassifier(expected_dim) on load errors or dim mismatches. Uses expected_dim consistently when no model_path. No API changes.
Neural parameter grids arc_solver/neural/sketches.py	generate_parameter_grid gains branches for crop (top/left 0–2, height/width 1–3), pad (out_h/out_w 5–19), and recolor (all src≠dst mappings for 0–9). Unknown ops behavior unchanged.
Program enumeration expansion arc_solver/search.py	enumerate_programs includes crop, pad, recolor in base ops with explicit param spaces; recolor no longer heuristic-only. Enlarges candidate program space and may change enumeration order. No signature changes.

Sequence Diagram(s)

sequenceDiagram
  autonumber
  participant U as Caller
  participant G as NeuralGuidance
  participant L as Loader
  participant F as SimpleClassifier(17)

  U->>G: __init__(model_path)
  alt model_path provided
    G->>L: load(model_path)
    alt load OK
      G->>G: validate input_dim == 17
      alt mismatch
        G-->>F: instantiate fallback
        G->>G: set neural_model = F
      else match
        G->>G: set neural_model = loaded
      end
    else load error
      G-->>F: instantiate fallback
      G->>G: set neural_model = F
    end
  else no model_path
    G-->>F: instantiate with expected_dim
    G->>G: set neural_model = F
  end
  U-->>G: ready

Loading

sequenceDiagram
  autonumber
  participant C as Caller
  participant S as enumerate_programs
  participant P as ParamSpace Builder

  C->>S: enumerate_programs(max_len)
  S->>P: build base_ops {rotate, flip, transpose, translate, identity, crop, pad, recolor}
  P-->>S: param grids (incl. crop/pad ranges, recolor mappings)
  loop lengths 1..max_len
    S->>S: combine ops × params
    S-->>C: yield program candidates
  end

Loading

sequenceDiagram
  autonumber
  participant D as DSL.apply_program
  participant OP as apply_op

  loop for i, (op, params) in program
    D->>OP: apply_op(op, params)
    alt success
      D->>D: continue
    else exception
      D-->>D: raise ValueError(op, index=i, params) [chained]
    end
  end

Loading

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~25 minutes

Poem

I hop through ops—crop, pad, recolor bright,
A grid of dreams expands the search’s sight.
If models fuss, I land on feet so spry,
Seventeen carrots tall—I still will try.
And when an op trips up mid-flight,
I name the spot—then bound to make it right. 🥕🐇

---

📜 Recent review details

Configuration used: CodeRabbit UI

Review profile: CHILL

Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between 5cd647f and e7549ec.

📒 Files selected for processing (4)

• arc_solver/dsl.py (1 hunks)
• arc_solver/neural/guidance.py (1 hunks)
• arc_solver/neural/sketches.py (1 hunks)
• arc_solver/search.py (2 hunks)

✨ Finishing touches

• 📝 Generate Docstrings

🧪 Generate unit tests

• Create PR with unit tests
• Post copyable unit tests in a comment
• Commit unit tests in branch codex/fix-dsl-operation-signatures-and-integration-bugs

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}