fg-data-profiling-wasm 0.2.0

Pyodide-friendly fork of fg-data-profiling (ydata-profiling) for browser WASM

fg-data-profiling-wasm

Pyodide-friendly fork of fg-data-profiling (formerly ydata-profiling / pandas-profiling).

You do not compile this package to WASM. Upstream already ships as a pure-Python wheel (py2.py3-none-any). This fork only does dependency surgery so it can install and run under Pyodide.

What changed vs upstream

Change	Why
Drop hard deps: numba, minify-html, phik	Native / JIT; broken or impractical in browser WASM
Default html.minify_html = False	Avoid lazy-importing the Rust minify-html extension
Soft-fail if minify is requested but missing	Safer in constrained environments
Keep phi_k off (already upstream default)	Avoid needing phik
Package name fg-data-profiling-wasm	Distinguish the Pyodide wheel from upstream
Allow pandas>=3 (<4)	Upstream still pins <3.0; this fork patches typeset/string paths

Optional native extras remain under [project.optional-dependencies] native for desktop use.

Layout

src/                  # vendored upstream + patches
patches/              # 0001 pyodide + 0002 pandas3 + 0003 display + 0004 font scale (reapplied on sync)
scripts/sync-upstream.sh
upstream.lock.json    # pinned upstream ref + commit
pyodide/              # install helper + dependency matrix
pyproject.toml        # Pyodide-friendly dependency set

Upgrading from upstream

Pinned upstream lives in upstream.lock.json.

# Is upstream ahead of our lock?
make check-upstream
# or: ./scripts/sync-upstream.sh --check

# Pull latest develop (or whatever ref is in the lock), re-apply patches,
# refresh dependency pins, update the lockfile
make sync
# or: ./scripts/sync-upstream.sh

# Sync a release tag / other ref
make sync-ref REF=v4.9.0
# or: ./scripts/sync-upstream.sh --ref v4.9.0

# After a successful sync, bump our PyPI version
./scripts/sync-upstream.sh --bump patch

# If a patch no longer applies: fix src/ by hand, then regenerate the patch series
./scripts/sync-upstream.sh --regen-patches

Typical release flow after upstream moves:

./scripts/sync-upstream.sh --ref develop --bump patch
python -m build
# review git diff, commit, push, twine upload dist/*

A weekly GitHub Action (.github/workflows/upstream-check.yml) opens/comments on an issue when the lock is behind.

Build a wheel

python -m pip install build
python -m build    # → dist/fg_data_profiling_wasm-*-py2.py3-none-any.whl

Current artifact:

• dist/fg_data_profiling_wasm-0.1.1-py2.py3-none-any.whl (purelib; PyPI-compatible; pandas <4)

Publish to PyPI (GitHub Actions)

The package is not on PyPI yet. Publishing is automated via Trusted Publishing (OIDC) — no long-lived API token in the repo.

One-time PyPI setup

1. Sign in at https://pypi.org/
2. Open Publishing → Add a new pending publisher (project does not exist yet):
   • PyPI project name: fg-data-profiling-wasm
   • Owner: awesome-wasm-packages
   • Repository: fg-data-profiling-wasm
   • Workflow name: publish.yml
   • Environment name: pypi
3. In GitHub → Settings → Environments → create environment pypi (optional protection rules / required reviewers).

Publish a version

# bump version in VERSION + pyproject.toml (or)
./scripts/sync-upstream.sh --bump patch
git add -A && git commit -m "Release $(tr -d '[:space:]' < VERSION)"
git tag "v$(tr -d '[:space:]' < VERSION)"
git push && git push --tags

Then create a GitHub Release for that tag (or use the UI). The Publish to PyPI workflow builds the wheel and uploads it.

You can also run the workflow manually: Actions → Publish to PyPI → Run workflow.

Use in Pyodide / PyConsole

Package name on PyPI is fg-data-profiling-wasm; import name is data_profiling.

import micropip
await micropip.install("fg-data-profiling-wasm")

import pandas as pd
from data_profiling import ProfileReport

df = pd.DataFrame({"a": [1, 2, 3], "b": ["x", "y", "x"]})
report = ProfileReport(
    df,
    minimal=True,
    progress_bar=False,
    correlations={"phi_k": {"calculate": False}},
    html={"minify_html": False},
)
html = report.to_html()

Notes for PyConsole (Pyodide 314 / Python 3.14):

• Use >=0.1.2 (older releases reject Python 3.14 / pin numpy/scipy/matplotlib too tightly).
• Auto-install-on-import will not find this package from import data_profiling — install by PyPI name first.
• Prefer minimal=True and turn off phi_k / HTML minify for speed and missing native deps.

Display

PyConsole already detects ProfileReport and calls to_html() (no package API required). This fork also adds:

• report.display() — js.render in PyConsole, else Jupyter to_notebook_iframe()
• _repr_html_() — returns an iframe HTML string (works without IPython)

report = ProfileReport(df, minimal=True, progress_bar=False)
report  # last expression: PyConsole / Jupyter rich display
# or: report.display()

Helpers live in pyodide/install.py; dependency notes in pyodide/DEPS.md.

Difficulty (recap)

Goal	Difficulty
Install this pure wheel in Pyodide	Easy
Usable ProfileReport (core EDA, minimal=True)	Medium
Full upstream fidelity (phik, minify, numba-shaped features)	Hard / skip

Upstream

Based on Data-Centric-AI-Community/fg-data-profiling. Tracked commit is in upstream.lock.json. MIT-licensed; see LICENSE.