interactive_modify_figures_from_python

Interactive figure polishing for R/ggplot2 outputs, without repeatedly editing plot code.

The tool provides a browser-based editor for .rds ggplot objects. You can drag text directly on the rendered figure, adjust common visual parameters, preview the result in real time, and export updated PDF/SVG/PNG/TIFF files.

Why This Exists

Late-stage scientific figures often need small visual changes: move a legend, increase a title font, adjust a bar width, recolor a group, or make a panel label line up cleanly. Doing that by repeatedly editing R scripts, rerunning the whole pipeline, and reopening exported files is slow.

This project turns that loop into a visual workflow:

open figure -> drag or edit controls -> preview -> save

Features

• Direct text dragging for plot titles, axis labels, legends, facet strips, and data labels.
• Browser-based inline SVG preview with live rendering after edits.
• Controls for font size, color, angle, bar width, y-axis limits, canvas size, tick labels, facet labels, and discrete color palettes.
• Patchwork-aware editing, with per-subplot switching and combined patch saving.
• Basic support for ggsurvplot Kaplan-Meier figures by editing the main plot.
• Non-destructive patch files: original .rds objects are never modified.
• Saved edits are written to *.patches.json and exported as equivalent *.patches.R code that can be copied back into an analysis script.
• Version-directory grouping for folders such as v1, v2, and v3, with the newest version shown by default.
• Persistent R rendering worker over a localhost socket, so switching figures does not restart R every time.

Requirements

Your figures should be R/ggplot2 objects saved as .rds files:

saveRDS(p, "fig.rds")

Python:

• Python 3
• Flask

R packages:

• ggplot2
• jsonlite
• svglite
• survminer for Kaplan-Meier figures created with ggsurvplot

For CJK text rendering in exported PDF/PNG files, install an appropriate system font such as SimHei or another CJK-compatible font.

Quick Start

pip install flask

export FIG_ROOT=/path/to/your/ggplot_figures
export RSCRIPT=/usr/bin/Rscript

python3 app.py

Then open:

http://localhost:5081

You can also restart the service by port:

bash restart.sh

Configuration

Variable	Default	Description
FIG_ROOT	/path/to/your/ggplot_figures	Root directory scanned recursively for .rds figure files
RSCRIPT	/usr/bin/Rscript	Path to the Rscript executable
FIG_EDITOR_PORT	5081	HTTP server port

How It Works

Each .rds figure is treated as an immutable source object. Every visual edit made in the browser is stored as a flat patch entry, for example:

text|plot.title|size
layer|2|geom_params|width
strip|<facet label>

Rendering follows this flow:

load .rds -> apply patches -> render preview/export

Text dragging is implemented by matching ggplotGrob text grobs with frontend SVG <text> elements by order and content. Offsets are applied in point units with unit(d, "pt"). When saved, patches are written to JSON, output files are regenerated, and equivalent R patch code is emitted for reproducibility.

Project Structure

interactive_modify_figures_from_python/
├── app.py          # Flask backend: figure list, inspect, render, save APIs
├── r_worker.R      # Persistent R worker using a localhost socket and JSON lines
├── templates/      # Editor page
├── static/         # Frontend JS/CSS for panels, SVG dragging, and render debounce
├── renders/        # Preview render outputs
└── restart.sh      # Port-aware restart helper

Known Limitations

• Data-mapped geom_text labels, such as sample-size labels, cannot be edited one by one as text values. The layer font size and color can be changed, and individual labels can be repositioned, but changing label content requires editing the source data.
• Manual scale colors are read from and written back to the scale environment values, which is expected behavior for this implementation.

License

MIT