Scientific Figure Guide

Problem

Research figures often fail at two points: choosing between line, scatter, box, violin, or heatmap plots, and making the output match journal rules for fonts, line width, error bars, colorblind-friendly palettes, and panel alignment. This skill targets the early stage of scientific visualization, helping normalize the path from data to code rather than replacing Origin or Illustrator.

Workflow

It first confirms the output mode: finished, template, or half-finished. Finished mode generates runnable code from real data; template mode provides a code skeleton, placeholders, and style parameters; half-finished mode fills in partial data and marks parameters that still need adjustment. Then it selects chart types by data relationship: line plots for trends, scatter plots with regression for continuous variable relationships, bar, box, or violin plots for group comparisons, heatmaps for matrix correlations, and diagrams for mechanisms.

Output and limits

The response typically includes chart diagnosis, data preparation, matplotlib code, style parameters, and post-processing notes. It emphasizes font size, line width, DPI, error bars, axis ranges, and subplot alignment. It is suited for experimental curves, mean comparisons, correlation analysis, and paper panels, but not for publication-grade vector polishing or complex 3D modeling in Blender or Mayavi. For large datasets, it should suggest downsampling, binning, or switching to hexbin.