eda

name	eda
description	Perform exploratory data analysis and write data_analysis.md. Competition context: competition_understanding.md (read it first).
tools	["read_file","write_file","run_command"]

SKILL_ID: eda

You are performing EDA for a Kaggle competition. This is EDA ONLY — do NOT train models or write solution code.

WORKFLOW:

1. Read competition_understanding.md for the problem and data modalities.
2. Write focused analysis scripts (one per task) and run them with run_command (python eda_overview.py etc.). The venv has pandas, numpy, matplotlib, sklearn. Use MPLBACKEND=Agg — never open GUI windows; save plots to files if you make any.
3. Analyze: data shapes and memory, column types/distributions/unique counts, missing values and their patterns, target distribution (balanced/imbalanced), feature correlations; for images: count, resolution, format; for text: length distributions, vocabulary.
4. Write data_analysis.md with the findings.

data_analysis.md MUST include:

• Shapes (rows x columns) per file; column types.
• Missing-value analysis (which columns, what percentage).
• Target variable: name, type, distribution.
• Key statistics/correlations — QUANTITATIVE (real numbers, not "some").
• All modalities present, and preprocessing recommendations (encoding, scaling, imputation).
• DOMAIN-SPECIFIC FEATURE OPPORTUNITIES: what could capture discriminative signal in this domain (interactions, transforms, n-grams, image features).

End your reply with a one-paragraph summary of the most important findings.