📄📊 Swizz

Swizz is a Python library for generating publication-ready visualizations, LaTeX tables, and subfigure layouts with minimal code and consistent style. Check out the live docs for examples and usage.

Built for AI/ML researchers, it's designed to make NeurIPS/ICLR/CVPR-style figures effortless — no more LaTeX hacks and style mismatches. Focus on your results, not your rendering.

If you use Swizz in your research, please consider citing it using: bibtex @software{quaedvlieg2025swizz, author = {Quaedvlieg, Lars and Miele, Andrea}, license = {MIT}, month = apr, title = {{Swizz: Publication-ready plots and LaTeX tables for ML papers}}, url = {https://github.com/lars-quaedvlieg/swizz}, version = {0.1.0}, year = {2025} }

🚀 Features

• 🧾 Auto-generated LaTeX tables from your data
• 📊 One-liner plotting functions
• 🧩 Easy layout builders for stacked, grid, and subfigure formats
• 📚 Expanding Jupyter Book documentation with live examples
• 🎬 Manim animations for dynamic visualizations and function evolutions
• 📈 Weights & Biases integration for experiment tracking and analysis

📦 Installation

Via PyPi: bash pip install swizz

By cloning the repository: bash git clone git@github.com:lars-quaedvlieg/swizz.git swizz cd swizz pip install .

📁 Project Structure

| Module | Description | |-------------------|----------------------------------------------------------------| | swizz.table | Table generators | | swizz.plot | Plotting utilities built on Seaborn & Matplotlib | | swizz.layout | Layout builders for stacked / side-by-side images | | swizz.manim | Dynamic visualizations and animations | | swizz.logging | Experiment tracking with Weights & Biases |

🧪 Examples

Multi-level table example:

```python from swizz.table import table

complex_df = ...

latexstring = table( "groupedmulticollatex", df=complexdf, rowindex="Model", colindex=["Split", "Budget"], groupby="Task", valuecolumn="score", highlight="min", stderr=True, caption="Combinatorial optimization results", label="tab:comboresults" ) ```

Simple bar chart example:

```python from matplotlib import pyplot as plt from swizz import plot

data_dict = ...

Style map for each metric (hatch patterns for filling)

style_map = { "Accuracy": '', "Precision": '\', "Recall": 'x' # Cross hatch pattern for Recall }

plot("generalbarplot", datadict, stylemap=style_map, save="bar") plt.show() ```

Weights & Biases integration example:

```python from swizz.logging.wandb_analyzer import WandbAnalyzer, RunGroup

Initialize the analyzer

analyzer = WandbAnalyzer("your-username/your-project", verbose=True)

Define run groups (either by prefix or run IDs)

run_groups = [ RunGroup(name="experiment1", prefix="your-prefix-1"), RunGroup(name="experiment2", prefix="your-prefix-2"), ]

Get analyzed metrics

resultsdf = analyzer.computegroupedmetrics( rungroups, xkey="roundnum", ykey="yourmetric" )

Plot the results

figscores, ax = plot( "multiplestdlinesdf", figsize=(8,5), datadf=resultsdf, labelkey="groupname", xkey="roundnum", ykey="yourmetricmean", yerrkey="yourmetricstd", xlabel="Sampling Budget", ylabel="Average Score", legendtitle="Experiments", legendncol=2, legend_loc="lower right" ) plt.show() ```

Manim animation example:

```python from swizz import render_manim import numpy as np import pandas as pd

Create sample data

methods = ['Method A', 'Method B', 'Method C'] iterations = range(30) scores = []

for method in methods: for iteration in iterations: # Generate evolving scores for each method if method == 'Method A': mean = 50 + iteration * 1 std = 10 + iteration * 3 elif method == 'Method B': mean = 40 + iteration * 2 std = 15 + iteration * 0.5 else: # Method C mean = 45 + iteration * 3 std = 12

    scores.extend([(method, iteration, score) 
                  for score in np.random.normal(mean, std, 300)])

scores_df = pd.DataFrame(scores, columns=['method', 'iteration', 'score'])

Render the animation

rendermanim( "histogramsevolution", renderconfig={ "quality": "highquality", "format": "mp4", "savepngs": True, }, scoresdf=scoresdf, methodcolumn="method", iterationcolumn="iteration", scorecolumn="score", xmin=0, xmax=100, xstep=10, numbins=60, xlength=10, ylength=5, timebetweeniterations=0.5, color_dict={ "Method A": "#1f77b4", "Method B": "#ff7f0e", "Method C": "#2ca02c", }, ) ```

Complex nested layouts:

```python from swizz.layouts.blocks import Row, Col, LegendBlock, Label from swizz.layouts import render_layout from matplotlib import pyplot as plt

plot1, plot2, plot3 = ...

nestedlayout = Col([ Row([ LegendBlock(labels=["Accuracy", "Precision", "Recall"], ncol=3, fixedwidth=0.35), LegendBlock(labels=["Forward KL", "Reverse KL"], ncol=2) ], fixedheight=0.08, spacing=0.15), Row([ Col([ plot3, Label("(a) Bar chart", align="center", fixedheight=0.05), ]), Col([ plot1, Label("(b) Line plot 1", align="center", fixedheight=0.05), plot2, Label("(c) Line plot 2", align="center", fixedheight=0.05) ], spacing=0.07) ], spacing=0.1), ], spacing=0.02)

fig = renderlayout(nestedlayout, figsize=(10, 8)) plt.show() ```

🛠️ Roadmap

• [ ] Add more plot types (confusion, UMAP, attention, histograms, etc.)
• [x] Add Manim integrations for dynamic plot videos and function evolutions
• [ ] Add more tables
• [x] W&B / MLflow integration

🤝 Contributing

Contributions are very welcome! See CONTRIBUTING.md for setup and module structure.