界面已本地化;技能介绍与说明文档按所选语言由 AI 自动翻译,代码、命令、路径和专业术语保留原文,以原作者文档为准。
探索技能
数据与研究 · SKILL.md

data-parsers

data-parsers

【数据与研究】Routes pyCirclize matrix, table, BED, stacked-bar, radar, from-to, chord, and radar-chart data preparation workflows.

独立 GitHub 快照 · 自动分类 · 以原作者文档为准

技能说明

GitHub

name: data-parsers description: "Routes pyCirclize matrix, table, BED, stacked-bar, radar, from-to, chord, and radar-chart data preparation workflows." disable-model-invocation: true metadata: disco-role: operating license: MIT

Data parsers

Use this route when the task is to turn tabular data into pyCirclize sectors, links, stacked bars, or radar series, or to build a chord/radar chart from a file, pandas.DataFrame, or parser object.

Route by input and output

  • Chord or graph-like data: read API reference and data formats. Use Matrix for a matrix and Matrix.parse_fromto_table for three-column from-to data, then call Circos.chord_diagram. Use to_sectors() for Circos(...) and to_links() for explicit circos.link workflows.
  • Radar data: use RadarTable or let Circos.radar_chart wrap a path, DataFrame, or table. Confirm numeric values and vmin < vmax before plotting; use the radar workflow for colors, labels, tooltips, and export.
  • Stacked bars and labels: use StackedBarTable for row/column names, heights, bottoms, row totals, and label positions. Hand drawing and track coordinate details belong to plot-primitives.
  • BED/cytoband preparation: use Bed for local tab-delimited BED records and route genomic feature semantics, GenBank/GFF, and trees to genomics-and-trees.

Read the linked references before choosing a parser. They contain the verified signatures, positional column assumptions, conversion semantics, and recovery steps that are intentionally not duplicated here. Run the bundled deterministic check when a small parser-plus-export smoke test is useful:

python scripts/validate_matrix_and_radar.py

The helper creates only temporary local fixtures and PNGs; it makes no network requests and does not modify the checkout or a user-specified data file.

Fast checks before plotting

  1. Decide whether the source is a square/rectangular matrix, a three-column from-to table, a row-indexed table, or BED records; do not infer a schema from a filename alone.
  2. Preserve labels in the DataFrame index/columns. File inputs use the first column as the index, and parser names are exposed as strings.
  3. Pass the actual delimiter for CSV or other delimited files. A wrong delimiter often creates one malformed label column rather than a useful parser error.
  4. Check numeric cells before Matrix, StackedBarTable, or RadarTable plotting. Parser constructors are intentionally lightweight and do not normalize arbitrary DataFrames for you.
  5. For radar, choose vmin and vmax so every plotted value is in range and ensure vmin < vmax; invalid range errors may be raised while the factory builds the series.

Do not use this route for low-level Track/Sector drawing, annotations, or coordinate debugging; use plot-primitives. Do not use it for GenBank/GFF feature extraction or Newick/tree styling; use genomics-and-trees.

我的收藏