Exploratory Data Analyzer

Upload a CSV or Excel file and explore it in your browser: see what's in each column, spot missing values, chart distributions, compare columns side by side, and find correlations — no code required.

The Overview tab for a 249-row sports-teams workbook: row counts, a preview of the data, and a per-column summary of types and missing values.

What it does

• Upload — CSV, XLSX, or XLS. For Excel workbooks, pick which sheet to analyze; if the first row is a title or banner, set the header row to skip it.
• Smart column handling — columns that look like IDs (names such as user_id, uuid, or columns where every value is unique) are set aside automatically so they don't clutter the analysis. You can pull any of them back in with "Force-include".
• Big files stay fast — large files are sampled down to a row limit you choose (1,000–200,000) so the app stays responsive. You always see both the total and analyzed row counts.
• Forgiving file reading — for CSVs, the app detects common delimiters and encodings automatically, and can skip broken rows in messy files.

Then explore across three tabs:

• Overview — row and column counts, a preview of the first 200 rows, and a table showing each column's type, missing values, and unique values. Downloadable as a CSV.
• Columns — pick any column to dig in:
  • Numbers — statistics summary and an adjustable histogram.
  • Text & categories — most common values with a bar chart, plus text length stats.
  • Dates — earliest/latest/range, plus a chart of rows over time.
  • Compare — optionally pick a second column for a scatter plot, box plot, or cross-table.
  • Search, filter by type, and mark favorites for quick access.
• Correlations — a heatmap showing which numeric columns move together, plus a ranked list for whichever column you pick. Results are cached for speed; the app tells you when they're out of date.

Requirements

The easiest way to run it is Docker (nothing else to install). To run it directly instead, you need Python 3.11+.

Run with Docker

docker compose up --build -d

Then open http://localhost:8501. The app only listens on your own machine.

If compose fails with a missing-network error, create it once:

docker network create website-network

Run without Docker

python3.11 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
streamlit run app/app.py --server.port=8501

Then open http://localhost:8501.

Optional: put it on the internet

A Cloudflare Tunnel service is included in the compose stack:

1. In the Cloudflare Zero Trust dashboard, create a managed tunnel and copy its token into CLOUDFLARE_TUNNEL_TOKEN in .env (see .env.example).
2. Point the tunnel's public hostname at http://eda-app:8501.
3. Run docker compose up -d — the tunnel starts alongside the app.

If no token is set, the tunnel container simply stays stopped — nothing breaks.

Tests

pip install -r requirements-dev.txt
pytest

The suite exercises the real app end to end: file parsing edge cases, ID detection, the column picker, Excel header rows, and the correlation cache.

Good to know

• ID detection is a guess based on names and uniqueness — when it guesses wrong, use Force-include to bring the column back.
• Correlations are cached for speed. If you change the file, sampling, or excluded columns, the app asks you to Re-run correlations rather than silently showing stale numbers.
• Date columns are detected by sampling values, so unusual date formats may show up as plain text.
• File uploads are capped at 500 MB.

License

This project is licensed under the MIT License — see the LICENSE file for details.