anonymousreview111's picture
Add ExecuGraph internal-30 benchmark (anonymized for double-blind review)
08f2c3e verified
metadata
license: apache-2.0
language:
  - en
tags:
  - code-generation
  - algorithms
  - data-structures
  - dynamic-programming
  - graph-algorithms
  - benchmark
  - execution-grounded
pretty_name: ExecuGraph Internal-30 DSA Benchmark
size_categories:
  - n<1K
configs:
  - config_name: default
    data_files: internal30.jsonl

ExecuGraph Internal-30 Benchmark

Anonymized for double-blind review. Author, affiliation, and citation metadata are withheld and will be released on acceptance.

A curated suite of 30 data-structures-and-algorithms (DSA) problems used as the headline benchmark for the ExecuGraph framework. Each problem ships with an unambiguous natural-language specification, the target function signature (and accepted aliases), a documented selection rationale, and a set of deterministic test cases (boundary, canonical, and stress).

Composition

Category Problems Notes
Dynamic Programming (dp) 10 Fibonacci, coin change, LCS, LIS, edit-distance-style recurrences
Graph Algorithms (graph) 10 BFS/DFS, topological sort, Dijkstra, Bellman–Ford, SCC, connectivity
Data Structures (ds) 10 LRU cache, min-stack, heap/priority queue, AVL, linked-list, traversal
  • 27 problems are originally designed; 3 are adapted from the APPS-introductory partition (marked source = "apps-…").
  • 30 problems, 125 deterministic test cases in total.

Fields

Each line in internal30.jsonl is one problem:

Field Type Description
id string Stable problem identifier (e.g. topo_sort)
category string dp | graph | ds
source string internal or apps-<id>
difficulty string easy | medium | hard
statement string Self-contained problem specification
primary_function string Expected function name
signature_aliases list[string] Accepted alternative function names
selection_rationale string Why the problem is included
timeout_s float Per-execution wall-clock budget
tests list[object] Test cases: {args, expected, call, description, kind, has_custom_judge}

Within each test, args and expected are JSON-encoded strings (their native types vary between int, list, bool, … across problems, so they are serialized for a uniform, Arrow-loadable schema). Parse them back with json.loads. A small number of graph problems whose correct output is not uniquely ordered (e.g. topological sort) use a custom equality judge in the reference harness; those tests are flagged with has_custom_judge = true. For portable use, treat them as exact-match or supply an order-insensitive comparison.

Usage

import json
from datasets import load_dataset

ds = load_dataset("anonymousreview111/execugraph-internal30", split="train")
p = ds[0]
print(p["id"], p["category"], len(p["tests"]))
args = json.loads(p["tests"][0]["args"])        # -> e.g. [10]
expected = json.loads(p["tests"][0]["expected"]) # -> e.g. 55

Reproducibility

This dataset is regenerated from the framework's single source of truth (execugraph/benchmarks/internal30.py) via scripts/export_dataset.py in the accompanying (anonymized) code repository. Every numeric claim in the paper's internal-30 tables is reproducible from that code together with this dataset.

License

Released under the Apache-2.0 license.