DS-1000: A Natural and Reliable Benchmark for Data Science Code Generation

TL;DR

DS-1000 is a new, reliable benchmark for data science code generation that features diverse real-world problems, precise evaluation, and measures model generalization beyond memorization.

Contribution

We created DS-1000, a comprehensive benchmark with realistic problems, multi-criteria evaluation, and methods to prevent memorization, advancing data science code generation research.

Findings

Current best model (Codex-002) achieves 43.3% accuracy.

Our evaluation method is highly reliable with only 1.8% incorrect solutions.

DS-1000 covers diverse, practical data science problems from StackOverflow.

Abstract

We introduce DS-1000, a code generation benchmark with a thousand data science problems spanning seven Python libraries, such as NumPy and Pandas. Compared to prior works, DS-1000 incorporates three core features. First, our problems reflect diverse, realistic, and practical use cases since we collected them from StackOverflow. Second, our automatic evaluation is highly specific (reliable) -- across all Codex-002-predicted solutions that our evaluation accept, only 1.8% of them are incorrect; we achieve this with multi-criteria metrics, checking both functional correctness by running test cases and surface-form constraints by restricting API usages or keywords. Finally, we proactively defend against memorization by slightly modifying our problems to be different from the original StackOverflow source; consequently, models cannot answer them correctly by memorizing the solutions from pre-training. The current best public system (Codex-002) achieves 43.3% accuracy, leaving ample room for improvement. We release our benchmark at https://ds1000-code-gen.github.io.