pyaerial 1.0.13

An implementation of the Aerial neurosymbolic association rule mining algorithm from tabular datasets.

pyaerial: scalable association rule mining

---

---

📥 Install | 🚀 Quick Start | ✨ Features | 📚 Documentation | 📄 Cite | 🤝 Contribute | 🔑 License

PyAerial is a Python implementation of the Aerial scalable neurosymbolic association rule miner for tabular data. It utilizes an under-complete denoising Autoencoder to learn a compact representation of tabular data, and extracts a concise set of high-quality association rules with full data coverage.

Unlike traditional exhaustive methods (e.g., Apriori, FP-Growth), Aerial addresses the rule explosion problem by learning neural representations and extracting only the most relevant patterns, making it suitable for large-scale datasets. PyAerial supports GPU acceleration, numerical data discretization, item constraints, and classification rule extraction and rule visualization via NiaARM library.

Learn more about the architecture, training, and rule extraction in our paper: Neurosymbolic Association Rule Mining from Tabular Data

---

Installation

Install PyAerial using pip:

pip install pyaerial

Note: Examples in the documentation use ucimlrepo to fetch sample datasets. Install it to run the examples:

pip install ucimlrepo

Data Requirements: PyAerial works with categorical data. Numerical columns must be discretized first, but you don't need to one-hot encode your data—PyAerial handles that automatically (unlike libraries like mlxtend that require manual one-hot encoding).

---

Performance

PyAerial significantly outperforms traditional ARM methods in scalability while maintaining high-quality results:

Execution time comparison across datasets of varying sizes. PyAerial scales linearly while traditional methods (e.g., Mlxtend, SPMF) exhibit exponential growth.

Key advantages:

• ⚡ 100-1000x faster on large datasets compared to standard rule mining algorithms in Python (e.g., Apriori, FP-Growth, ECLAT, ...)
• 📈 Linear scaling with dataset size (vs. exponential for traditional methods)
• 🎯 No rule explosion - extracts concise, high-quality rules with full data coverage
• 💾 Memory efficient - neural representation avoids storing exponential candidate sets

For comprehensive benchmarking and comparisons with Mlxtend (e.g., FPGrowth, Apriori etc.), and other ARM tools, see our benchmarking paper: PyAerial: Scalable association rule mining from tabular data (SoftwareX, 2025)

---

Quick Start

The following are basic example usages of PyAerial. 📚

See full feature list | Read the complete documentation, to see the full capabilities.

Basic Association Rule Mining

from aerial import model, rule_extraction
from ucimlrepo import fetch_ucirepo

# Load a categorical tabular dataset
breast_cancer = fetch_ucirepo(id=14).data.features

# Train an autoencoder on the loaded table
trained_autoencoder = model.train(breast_cancer)

# Extract association rules with quality metrics calculated automatically
result = rule_extraction.generate_rules(trained_autoencoder)

print(f"Overall statistics: {result['statistics']}\n")
print(f"Sample rule: {result['rules'][0]}")

Output:

Overall
statistics: {
    "rule_count": 15,
    "average_support": 0.448,
    "average_confidence": 0.881,
    "average_coverage": 0.860,
    "data_coverage": 0.923,
    "average_zhangs_metric": 0.318
}

Sample
rule: {
    "antecedents": [{"feature": "inv-nodes", "value": "0-2"}],
    "consequent": {"feature": "node-caps", "value": "no"},
    "support": 0.702,
    "confidence": 0.943,
    "zhangs_metric": 0.69,
    "rule_coverage": 0.744
}

Interpretation: When inv-nodes is between 0-2, there's 94.3% confidence that node-caps equals no, covering 70.2% of the dataset.

Quality metrics explained:

• Support: Frequency of the rule in the dataset (how often the pattern occurs)
• Confidence: How often the consequent is true when antecedent is true (rule reliability)
• Zhang's Metric: Correlation measure between antecedent and consequent (-1 to 1; positive values indicate positive correlation)
• Rule Coverage: Proportion of transactions containing the antecedents
• Data Coverage (in statistics): Overall proportion of the dataset covered by at least one rule

---

Can't get the results you're looking for?

Learn how to adjust parameters for your specific needs:

• 🎯 Parameter Tuning Guide - Quick reference for high/low support, confidence, and more
• 🔧 Troubleshooting - What to do when Aerial doesn't find rules or takes too long
• ⚙️ Advanced Tuning - Training duration and architecture optimization

---

Working with rules: Access rule components and metrics easily using the dictionary format:

# Example: Print all rules in a readable format
for rule in result['rules']:
    antecedents_str = " AND ".join([f"{a['feature']}={a['value']}" for a in rule['antecedents']])
    consequent_str = f"{rule['consequent']['feature']}={rule['consequent']['value']}"
    print(
        f"IF {antecedents_str} THEN {consequent_str} (support: {rule['support']:.2f}, conf: {rule['confidence']:.2f})")

# Sample output:
# IF inv-nodes=0-2 THEN node-caps=no (support: 0.70, conf: 0.94)
# IF age=30-39 AND menopause=premeno THEN breast=left (support: 0.45, conf: 0.75)
# IF tumor-size=30-34 THEN deg-malig=2 (support: 0.38, conf: 0.82)

---

Working with Numerical Data

For datasets with numerical columns, use PyAerial's built-in discretization:

from aerial import model, rule_extraction, discretization
from ucimlrepo import fetch_ucirepo

# Load a numerical dataset (e.g., Iris)
iris = fetch_ucirepo(id=53).data.features

# Discretize numerical columns into categorical bins
iris_discretized = discretization.equal_frequency_discretization(iris, n_bins=3)

# Train and extract rules as usual
trained_autoencoder = model.train(iris_discretized, epochs=10)
result = rule_extraction.generate_rules(trained_autoencoder, ant_similarity=0.1)
print(
    f"Found {result['statistics']['rule_count']} rules with avg support {result['statistics']['average_support']:.3f}")

Example discretization output:

# Before: sepal_length = 5.1, 4.9, 7.0, ...
# After:  sepal_length (ranges of values) = (4.8, 5.5], (4.8, 5.5], (6.4, 7.9], ...

PyAerial provides equal_frequency_discretization and equal_width_discretization methods. See the User Guide for more discretization options.

---

ARM with Item Constraints

Focus rule mining on specific features of interest instead of exploring the entire feature space:

from aerial import model, rule_extraction
from ucimlrepo import fetch_ucirepo

breast_cancer = fetch_ucirepo(id=14).data.features
trained_autoencoder = model.train(breast_cancer)

# Define features of interest for the antecedent side
features_of_interest = ["age", {"menopause": 'premeno'}, {"node-caps": "yes"}]

# Extract rules focusing only on specified features
result = rule_extraction.generate_rules(
    trained_autoencoder,
    features_of_interest,
    cons_similarity=0.5
)

Output: Rules with specified features on the antecedent side (left side of an if-else rule):

{
    "antecedents": [{"feature": "menopause", "value": "premeno"}],
    "consequent": {"feature": "node-caps", "value": "no"},
    "support": 0.357,
    "confidence": 0.68,
    "zhangs_metric": -0.066,
    "rule_coverage": 0.525
}

This is ideal for domain-specific exploration where you want to understand relationships involving particular features.

---

Classification Rules for Interpretable Inference

Learn rules with target class labels on the consequent side for interpretable classification:

import pandas as pd
from aerial import model, rule_extraction, rule_quality
from ucimlrepo import fetch_ucirepo

# Load dataset with class labels
breast_cancer = fetch_ucirepo(id=14)
labels = breast_cancer.data.targets
features = breast_cancer.data.features

# Combine features with labels
table_with_labels = pd.concat([features, labels], axis=1)

trained_autoencoder = model.train(table_with_labels)

# Generate classification rules with target class on consequent side
result = rule_extraction.generate_rules(
    trained_autoencoder,
    target_classes=["Class"],
    cons_similarity=0.5
)

Output: Rules predicting class labels with quality metrics:

{
    "antecedents": [{"feature": "menopause", "value": "premeno"}],
    "consequent": {"feature": "Class", "value": "no-recurrence-events"},
    "support": 0.357,
    "confidence": 0.68,
    "zhangs_metric": -0.066,
    "rule_coverage": 0.525
}

These rules can be used for interpretable inference or integrated with rule-based classifiers from imodels.

---

Features

PyAerial provides a comprehensive toolkit for association rule mining with advanced capabilities:

• Scalable Rule Mining - Efficiently mine association rules from large tabular datasets without rule explosion
• Automatic Quality Metrics - Rules include support, confidence, Zhang's metric, and more calculated automatically
• Frequent Itemset Mining - Generate frequent itemsets with support values using the same neural approach
• ARM with Item Constraints - Focus rule mining on specific features of interest
• Classification Rules - Extract rules with target class labels for interpretable inference
• Numerical Data Support - Built-in discretization methods (equal-frequency, equal-width)
• Customizable Architectures - Fine-tune autoencoder layers and dimensions for optimal performance
• GPU Acceleration - Leverage CUDA for faster training on large datasets
• Comprehensive Metrics - Support, confidence, lift, conviction, Zhang's metric, Yule's Q, interestingness
• Rule Visualization - Integrate with NiaARM for scatter plots and visual analysis
• Flexible Training - Adjust epochs, learning rate, batch size, and noise factors

---

How Aerial Works?

Aerial employs a three-stage neurosymbolic pipeline to extract high-quality association rules from tabular data:

1. Data Preparation

Categorical data is one-hot encoded while tracking feature relationships. Numerical columns require pre-discretization ( equal-frequency or equal-width methods available). The encoded values are transformed into vector format for neural processing.

2. Autoencoder Training

An under-complete denoising autoencoder learns a compact representation of the data:

• Architecture: Logarithmic reduction (base 16) automatically configures layers, or use custom dimensions
• Bottleneck design: The encoder compresses input to the original feature count, forcing the network to learn meaningful associations
• Denoising mechanism: Random noise during training improves robustness and generalization

Example: Rule extraction process using weather and beverage features

3. Rule Extraction

Rules emerge from analyzing the trained autoencoder using test vectors:

1. Test vectors are created with equal probabilities across categories
2. Specific features are set to 1 (antecedents) while others remain at baseline
3. Forward passes through the network produce output probabilities
4. Rules are extracted when probabilities exceed similarity thresholds
5. Quality metrics (support, confidence, coverage, Zhang's metric, etc.) are calculated automatically using vectorized operations

Complete three-stage pipeline: data preparation → training → rule extraction

Learn more: For detailed explanations of the architecture, theoretical foundations, and experimental results, see our paper: Neurosymbolic Association Rule Mining from Tabular Data

---

Documentation

For detailed usage examples, API reference, and advanced topics, visit our comprehensive documentation:

📚 Read the full documentation on ReadTheDocs

Documentation includes:

• Getting Started - Installation and basic usage
• User Guide - 11 detailed examples covering all features
• Parameter Tuning Guide - How to get high/low support, confidence, and control rule count
• Configuration & Troubleshooting - GPU usage, debugging, and advanced training/architecture tuning
• API Reference - Complete function and class documentation
• How Aerial Works - Understanding the neurosymbolic architecture and algorithm

---

Citation

If you use PyAerial in your work, please cite our research and software papers:

@InProceedings{pmlr-v284-karabulut25a,
  title         = {Neurosymbolic Association Rule Mining from Tabular Data},
  author        = {Karabulut, Erkan and Groth, Paul and Degeler, Victoria},
  booktitle     = {Proceedings of The 19th International Conference on Neurosymbolic Learning and Reasoning},
  pages         = {565--588},
  year          = {2025},
  editor        = {H. Gilpin, Leilani and Giunchiglia, Eleonora and Hitzler, Pascal and van Krieken, Emile},
  volume        = {284},
  series        = {Proceedings of Machine Learning Research},
  month         = {08--10 Sep},
  publisher     = {PMLR},
  url           = {https://proceedings.mlr.press/v284/karabulut25a.html}
}

@article{pyaerial,
  title         = {PyAerial: Scalable association rule mining from tabular data},
  journal       = {SoftwareX},
  volume        = {31},
  pages         = {102341},
  year          = {2025},
  issn          = {2352-7110},
  doi           = {https://doi.org/10.1016/j.softx.2025.102341},
  author        = {Erkan Karabulut and Paul Groth and Victoria Degeler},
}

---

Contact

For questions, suggestions, or collaborations, please contact:

Erkan Karabulut 📧 [e-mail] 📧 [e-mail]

---

Contribute

We welcome contributions from the community! Whether you're fixing bugs, adding new features, improving documentation, or sharing feedback, your help is appreciated.

How to contribute:

• 🐛 Report bugs - Open an issue describing the problem
• 💡 Suggest features - Share your ideas for improvements
• 📝 Improve docs - Help us make the documentation clearer
• 🔧 Submit PRs - Fork the repo and create a pull request
• 💬 Share feedback - Contact us with your experience using PyAerial

Feel free to open an issue or pull request on GitHub, or reach out directly!

Contributors

All contributors to this project are recognized and appreciated! The profiles of contributors will be listed here:

Made with contrib.rocks.

---

License

This project is licensed under the MIT License - see the LICENSE file for details.