README.md

metadata

pretty_name: Matplotlib Code-Image Pairs
license: other
tags:
  - matplotlib
  - code
  - plotting
  - computer-vision
  - synthetic
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
dataset_info:
  features:
    - name: image
      dtype: image
    - name: code
      dtype: string
    - name: example_id
      dtype: string
    - name: figure_index
      dtype: int64
    - name: figure_name
      dtype: string
    - name: title
      dtype: string
    - name: example_page_url
      dtype: string
    - name: source_url
      dtype: string
    - name: source_relpath
      dtype: string
    - name: category_hint
      dtype: string
    - name: status
      dtype: string
    - name: num_figures
      dtype: int64
    - name: error
      dtype: 'null'
  splits:
    - name: train
      num_bytes: 57044107
      num_examples: 643
  download_size: 54960557
  dataset_size: 57044107

Matplotlib Code-Image Pairs

Dataset Summary

This dataset contains static Matplotlib figure images paired with the Python source code that generated them. It was built from the official Matplotlib gallery and excludes animation-style examples. Each row corresponds to one rendered image. Examples that produce multiple figures contribute multiple rows.

Dataset Composition

• Image/code rows: 643
• Source examples: 452
• Examples with multiple figures: 74
• Included statuses: ok
• Source gallery: https://matplotlib.org/stable/gallery/index.html

Top Categories

• lines_bars_and_markers: 48 examples
• images_contours_and_fields: 47 examples
• text_labels_and_annotations: 46 examples
• mplot3d: 44 examples
• subplots_axes_and_figures: 35 examples
• misc: 25 examples
• ticks: 24 examples
• axes_grid1: 22 examples
• statistics: 21 examples
• event_handling: 20 examples

Row Schema

• image: rendered PNG for one figure
• code: full Python source for the example
• example_id: stable identifier derived from the Matplotlib gallery URL
• figure_index: zero-based figure index within the example
• figure_name: original rendered image filename
• title: example page title
• example_page_url: Matplotlib gallery page URL
• source_url: downloadable Python source URL when available
• source_relpath: source filename reported by the scraper
• category_hint: rough gallery category inferred from the URL path
• status: scraper/render status kept for provenance
• num_figures: number of figures reported for the source example
• error: render or scrape error text, usually null for successful rows

Build Process

1. Crawl the official Matplotlib gallery.
2. Download the Python source for each example page.
3. Render examples with Matplotlib's non-interactive Agg backend.
4. Keep static rendered image/code pairs and skip dynamic animation-style examples.
5. Export one dataset row per rendered image.

Usage

Load the local saved dataset:

from datasets import load_from_disk

ds = load_from_disk("/usr/project/xtmp/ap843/hf_datasets/matplotlib_code_image_pairs")["train"]

Load from the Hugging Face Hub after pushing:

from datasets import load_dataset

ds = load_dataset("ajayvikram/matplotlib-code-image", split="train")