pysofra 0.1.0a3

Statistical reporting and table preparation framework for Python — the missing reporting layer.

PySofra

The missing statistical reporting layer for Python

PySofra turns datasets, fitted models, and summary statistics into publication-ready tables — across HTML · Markdown · LaTeX · DOCX · PPTX · XLSX · PNG — from a single immutable SofraTable object. It brings the practical workflows of R's tableone, gtsummary, and flextable into a single coherent Pythonic API.

_{Baseline characteristics, by treatment arm. One line of code.}

Adjusted ORs + inline forest plot tbl_regression(fit).with_forest_plot()

KM table + inline survival curve tbl_survival(...).with_km_plot()

Why PySofra

• One immutable object, seven output formats — build a SofraTable once, render to HTML / Markdown / LaTeX / DOCX / PPTX / XLSX / PNG, all byte-deterministic across processes
• Auto-dispatched statistical tests — Welch, Wilcoxon, ANOVA, Kruskal–Wallis, Fisher, χ², Rao–Scott, design-adjusted t — picked by variable kind, overridable per-row
• Inline forest plots and KM curves — embed matplotlib figures directly into the table; the same SofraTable renders them across every backend
• Statistically correct — every numeric output validated against scipy / statsmodels / lifelines at machine precision, with cross-checks against R's gtsummary
• Method-chainable and immutable — every modifier returns a new table; no in-place mutation, no global state, fully reproducible

Showcase notebook — 47 cells, every section a side-by-side numeric proof. Start here if you have 60 seconds.

End-to-end tutorial — 126 cells walking every public feature on a synthetic two-arm trial.

---

Quick start

import numpy as np
import pandas as pd
import pysofra as ps

# Toy two-arm trial; replace with your own DataFrame in real use.
rng = np.random.default_rng(0)
df = pd.DataFrame({
    "arm":   rng.choice(["Placebo", "Treatment"], 200),
    "age":   rng.normal(60, 10, 200).round(),
    "bmi":   rng.normal(28, 5, 200).round(1),
    "event": rng.binomial(1, 0.3, 200),
})

# Table 1 — baseline characteristics by treatment arm
tbl = (
    ps.tbl_one(df, by="arm")
      .add_p()
      .add_smd()
      .add_overall()
      .theme("clinical")
)

tbl                          # renders in Jupyter / Colab / VS Code
tbl.to_docx("table1.docx")   # publication-quality Word
tbl.to_html()                # standalone HTML fragment
tbl.to_markdown()            # GitHub-flavored Markdown

The same workflow handles regression tables:

import statsmodels.api as sm

X = sm.add_constant(df[["age", "bmi"]])
fit = sm.Logit(df["event"], X).fit(disp=False)

(
    ps.tbl_regression(fit, exponentiate=True)
      .bold_p()
      .theme("jama")
      .to_docx("table2.docx")
)

The end-to-end worked example — baseline table by treatment arm, regression table with forest plot, and Kaplan-Meier survival summary — is in the showcase notebook.

---

What's in the box

Feature	Function / object	Status
Baseline Table 1	ps.tbl_one	MVP
Descriptive summary	ps.tbl_summary	MVP
Regression results	ps.tbl_regression	MVP
Side-by-side merge	ps.tbl_merge	MVP
Vertical stack	ps.tbl_stack	MVP
HTML / Markdown	.to_html / .to_markdown	MVP
DOCX export	.to_docx	MVP
LaTeX export	.to_latex	MVP
PPTX export	.to_pptx	MVP (extras)
Excel export	.to_xlsx	MVP
Inline forest plots	tbl_regression(...).with_forest_plot()	MVP
Inline KM curves	tbl_survival(...).with_km_plot()	MVP
Cross-backend plot embedding	DOCX/PPTX/LaTeX include the plot too	MVP
Rao–Scott chi-square	weighted Table 1 auto-route	MVP
SurveyDesign (strata + cluster + FPC)	Taylor-linearised variance	MVP
Themes	clinical, jama, nejm, compact, minimal	MVP
Auto test selection	t-test / ANOVA / Wilcoxon / Kruskal / χ² / Fisher	MVP
Per-variable test overrides	tests={'age': 'wilcoxon', ...}	MVP
Multiplicity adjustment	.add_q() — BH, BY, Bonferroni, Holm, Hommel, Šidák	MVP
Multi-model regression	tbl_regression([m1, m2], model_labels=[...])	MVP
lifelines (Cox / AFT)	tbl_regression(cph)	MVP
sklearn (linear models)	tbl_regression(clf) — point estimates only	MVP
Kaplan–Meier summary	tbl_survival(df, time=, event=, by=, times=[...])	MVP
Survey-weighted Table 1	tbl_one(..., weights='w')	MVP
polars input	tbl_one(pl.DataFrame(...))	MVP
Conditional formatting	.bold_if, .highlight_if, .style_if	MVP
Sticky-header notebook tables	.to_html(sticky_header=True)	MVP
Standardised mean differences	continuous + categorical (Yang–Dalton)	MVP
Notebook rendering	_repr_html_ / _repr_markdown_ / _repr_latex_	MVP

---

Design principles

• Backend-agnostic tables. A SofraTable is the single source of truth; every renderer (HTML, Markdown, DOCX, …) reads the same object.
• Immutable method chaining. Every modifier returns a new SofraTable. No surprises, no global state.
• Strong defaults, explicit overrides. Sensible journal-style output out of the box; per-variable type, label, and test overrides when you need them.
• Deterministic. The same input always produces the same output — critical for reproducible research.
• No magic. No nonstandard evaluation, no metaprogramming, no network calls, no telemetry.

---

Installation

pip install pysofra

PySofra requires Python ≥ 3.11. The core install only pulls numpy, pandas, scipy, statsmodels, and python-docx. Domain extras unlock the features that depend on heavier optional libraries:

pip install "pysofra[survival]"   # tbl_survival + KM curves (lifelines, matplotlib)
pip install "pysofra[plot]"       # forest plots, table-as-image (matplotlib)
pip install "pysofra[pptx]"       # PowerPoint export (python-pptx)
pip install "pysofra[xlsx]"       # Excel export (xlsxwriter)
pip install "pysofra[polars]"     # accept polars DataFrames as input
pip install "pysofra[sklearn]"    # tbl_regression on scikit-learn models
pip install "pysofra[all]"        # everything above
pip install "pysofra[dev]"        # testing + linting (pytest, ruff, mypy, hypothesis)

---

Status

PySofra is in alpha (0.1.0a3). The public API surface is pinned by an explicit API-stability test so that any unintended rename, removal, or signature change surfaces as a failed test. Quality bar at this release:

• More than 800 tests passing, 100% line coverage, mypy strict, ruff clean.
• Every numeric output is validated against scipy, lifelines, statsmodels, or a hand-computed textbook formula (test_joss_statistical_correctness.py).
• Universal invariants enforced via Hypothesis on 720 randomized examples per CI run (test_joss_property_invariants.py).
• Renderer output is byte-deterministic — identical input always produces identical HTML/Markdown/LaTeX, required for reproducible publication artifacts (test_joss_renderer_consistency.py).

Bug reports and use-case feedback are very welcome.

Contributing

Bug reports, feature requests, and pull requests are all very welcome. Please read CONTRIBUTING.md for the workflow, the quality gates, and the Code of Conduct.

License

GPL-3.0-or-later. See LICENSE.

Citation

If you use PySofra in academic work, please cite the project — see CITATION.cff.