depictr (Python)

Documentation and example gallery: https://pablobernabeu.github.io/depictr-py/

A unified, colourblind-safe toolkit for publication-ready statistical visualisation, built on plotnine. It is the Python sibling of the depictr R package.

Gallery

A grouped density (the default palette is the colourblind-safe Okabe-Ito set) and Kaplan-Meier survival curves with a log-rank test and a number-at-risk table, both from a single function call:

There is more in the example gallery. These two use a couple of extra options (a title, legend_inside) shown in the Getting started guide, while the short tour below keeps its calls minimal.

What it is for

Python already has an excellent plot for almost any statistical task, but they live in different packages with different defaults. A single figure set for a paper might draw on seaborn, scikit-learn, statsmodels, lifelines and ArviZ, each with its own look, its own API and its own colour scheme, and only one of those defaults to colourblind-safe colours. Making the set consistent and accessible then means repeating the same theming by hand on every plot.

depictr does that work once. It gives the whole workflow one theme, one colourblind-safe palette and one calling convention, and returns plotnine objects you can keep extending with +. Where a specialist package already computes a quantity well, depictr delegates to it and redraws the result under the shared theme, so you keep the trusted computation and gain a coherent, accessible figure.

Accessibility by default

The default palette is the Okabe-Ito set, and that choice is checked rather than asserted. The package ships a simulator of colour-vision deficiency based on the model of Machado, Oliveira and Fernandes (2009) and a CIE-Lab distance test that measures how far apart the palette's colours stay under each deficiency:

import depictr as dp

dp.palette_safety()
# {'min_delta_e': ..., 'by_condition': {'normal': ..., 'protan': ...,
#  'deutan': ..., 'tritan': ...}, 'safe': True, 'threshold': 5.0}

Installation

depictr is on PyPI:

pip install depictr            # core (plotnine, pandas, numpy, matplotlib, scipy)
pip install depictr[all]       # plus the optional computation back-ends

The classification, model and survival plots delegate to scikit-learn, statsmodels and lifelines respectively. Each is an optional dependency, so the core stays light and you install only what your plots need (depictr[classification], depictr[models], depictr[survival]).

A short tour

import depictr as dp

# Exploratory analysis
ld = dp.lexical_decision()
dp.explore_distribution(ld, "RT", group="condition", kind="both")

wb = dp.wellbeing_survey()
dp.correlation_heatmap(wb)
dp.missingness_map(wb)

# Model estimates: a fitted model OR a tidy data frame
import statsmodels.formula.api as smf
# Q() quotes "yield" because it is a Python keyword.
fit = smf.ols('Q("yield") ~ fertiliser + rainfall + soil_ph + treatment',
              dp.crop_yield()).fit()
dp.coefficient_plot(fit)

# Classification: computed by scikit-learn, themed by depictr
ct = dp.clinical_trial()
dp.roc_curve_plot(ct["adverse_event"], ct["biomarker"])

# Survival: Kaplan-Meier with a log-rank test, in one call
dp.survival_plot(ct["time"], ct["event"], group=ct["arm"])

Every function returns a plotnine object, so the usual grammar-of-graphics extensions apply:

from plotnine import labs

dp.roc_curve_plot(ct["adverse_event"], ct["biomarker"]) + labs(title="Adverse event")

The web app

A Streamlit app provides a gallery and a low-friction way to try the package: it loads one of the bundled datasets (or a CSV you upload), draws the chosen plot, shows the exact Python call that produced it, and offers a colourblind-vision toggle that re-renders the figure as each deficiency would be seen. The app lives in the repository rather than the wheel, so run it from a clone:

git clone https://github.com/pablobernabeu/depictr-py
cd depictr-py
pip install -e ".[app]"
streamlit run app/streamlit_app.py

Function families

Status

This is an early release, but coverage is broad: the colourblind-safe theme, the accessibility check, and most of the R package's plotting functions across every family (EDA, estimation, model estimates, diagnostics, classification, multivariate, survival and time series) are in place and tested. Multi-panel composites are built on arrange_plots, which uses plotnine's native plot composition: the four-panel residual_diagnostics_plot, the model_report dashboard, the two-panel Gardner-Altman estimation_plot, the frequentist-over-Bayesian overlay, and the survival number-at-risk table.

A few known limitations remain. plotnine compositions have no figure-level title, so a grid carries its titles on each panel (the survival and estimation composites place the title on their top panel). survival_plot does not yet draw the confidence band or censor marks the R package draws; its conf_level argument is accepted for future use. And a handful of functions from the R package are not ported: optimizer_fixef_plot (there is no clean statsmodels equivalent of lme4::allFit), k_diagnostic, palette_preview, model_fit_table and ts_forecast.

Relationship to the R package

The two are siblings, not a shared codebase. The R package targets ggplot2 and CRAN; this one targets plotnine and PyPI. They share the design: one accessible theme across the whole workflow, model-or-data-frame input, and extensible plot objects.

Licence

MIT. See LICENSE.