Skip to content

Interactive output

Every freshdata report is a serializable object first — .summary(), .to_dict(), .to_frame() keep working with nothing but pandas installed. On top of that, the same objects render an interactive HTML view for notebooks and stakeholders. Rendering is separated from compute: the renderers live in freshdata.render and are imported lazily, so import freshdata never pulls a visualization library.

import freshdata as fd

cleaned, report = fd.clean(df, return_report=True)

report.show()            # collapsible action timeline + filterable audit ledger
fd.profile(df).show()    # inline quality cockpit
fd.suggest_plan(df).show()   # per-column decision cards
fd.explain_clean(df).show()  # before/after diff explorer
fd.compare_plans(df)         # strategy diff grid (rich repr in a notebook)
fd.compare_clean(df)         # outcome dashboard
fd.infer_roles(df)           # role / type confidence cards

Each object also exposes to_html() for embedding and _repr_html_() for automatic notebook display. Outside a notebook, .show() writes a standalone .html file and returns its path.

Optional extras

The base renderers produce self-contained HTML (scoped CSS + a little vanilla JS for filtering and collapsing) with no optional dependencies. Installing an extra simply upgrades the output:

Extra Adds Upgrades
freshdata[viz] itables, plotly, great-tables richer tables, hoverable charts, stakeholder-clean HTML tables
freshdata[notebook] the above + anywidget, ipython composed notebook widgets
freshdata[all] everything — 
pip install 'freshdata[viz]'

If you call a path that needs an optional package you don't have, freshdata raises a clear install hint (e.g. "install freshdata[viz]") rather than a bare ModuleNotFoundError.

What each surface shows

  • report.show() — one row per action with before→after impact, rationale, risk, confidence, whether it was automatic/suggested/skipped/approved, whether it is reversible, and whether memory influenced it. The audit ledger filters by column / action / risk and exports to CSV / JSON.
  • profile.show() — overall quality score, missingness / duplicate / type-inference summaries, an issue-ranked column list with null bars, and cardinality / outlier warnings. Correlations are computed on demand to keep the first render fast.
  • suggest_plan(df).show() — per-column recommended action, confidence, risk, why, alternatives, and a "needs review" flag.
  • explain_clean(df).show() — before/after column comparison with risky changes separated from safe ones.

Generating HTML examples

python scripts/generate_html_examples.py     # writes docs/examples/*.html

The committed samples under docs/examples/ are produced by that script from synthetic data, so they always match the current renderers.