Conala Mined Curated: 600K Code and Natural Language Pairs from Stack Overflow

Conala Mined Curated is a dataset of approximately 600,000 examples derived from the mined subset of the CoNaLa dataset. The data originates from Stack Overflow and was filtered and curated by the author 'codeparrot'. The dataset was last updated on the platform on June 13, -2023.

Use Cases

• Train code generation models based on the natural language intent descriptions.
• Develop semantic search for programming questions based on the mined question-answer pairs.
• Fine-tune models for code summarization based on the linked code snippets.
• Create benchmarks for evaluating the alignment between natural language queries and code solutions.

Strengths

• Contains approximately 600,000 examples, providing substantial scale for training.
• Derived from a real-world source (Stack Overflow), reflecting practical programming problems.
• Curated from a larger mined set, suggesting some level of quality filtering.

Limitations

• The description states the mined part is 'not comparably post-processed', indicating potential noise.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is approximate; exact size and file formats are unknown.

Provenance

Source

Stack Overflow, via the CoNaLa dataset.

Collection Method

Crawled and mined from Stack Overflow, then filtered and curated.

Time Range

null

Freshness

Last updated 2023-06-13 15:56:31; freshness should be verified.

Geography

null