how-to-edalisted

# how-to-EDA-using-claude-python Decision-first EDA: the output is a ranked list of findings an executive can act on, with charts as evidence - not a wall of describe() output. Showcase walkthrough (Everrest retail case, real executed charts): https://github.com/phoebefu6/phoebe-data-skills - `docs/how-to-eda-using-claude-python/` ## The 6 steps ### 1. Input Collect the schema (tables, columns, types, one-line descriptions) and business context (industry, business model, what the org cares about this quarter). If given real data, also note grain and row counts per table. Never proceed on column names alone - ask what the business would pay to know. ### 2. Sample data (only when real data isn't available yet) Write a seeded generator (`numpy.random.default_rng(<seed>)`) at realistic scale. Plant quirks on purpose - missing-not-at-random, outlier entities, duplicates, seasonality, one suspicious segment - and document each in the docstring. EDA that finds nothing teaches nothing, and the planted list becomes your recall test. With real data, skip generation; the quirks are already in there. ### 3. Objective Frame ONE decision question ("what should <team> act on this quarter, and what in this data can't be trusted yet?"), then 3-5 sub-questions. Every chart must serve a sub-question; anything else is decoration. ### 4. Find-skills List the tools/skills that assist this objective before writing code: ydata-profiling (10-minute head start), Great Expectations (turn finding