skewnormlib 0.1.3

A Python library for skew-weighted normalization

skewnormlib

skewnormlib is an open-source Python library designed for preprocessing data using a novel method called Skew Weighted Normalization. This technique accounts for the skewness in data and applies normalization while optionally introducing a non-linear transformation. It is ideal for preparing data for machine learning models that require normalized inputs.

---

What is SkewWeightedNormalization?

The SkewWeightedNormalization class is a custom data transformer that normalizes data while considering its skewness. It achieves this by combining a skew-adjusted normalization term and a non-linear transformation term, making it more robust for skewed datasets.

Mathematical Formula:

The transformation is defined as:

[ \text{scaled_data} = \frac{X - \mu}{\sigma (1 + \alpha \cdot |\gamma|)} + \beta \cdot \tanh\left(\frac{X - \mu}{k \cdot \sigma}\right) ]

Where:

• X: Input data.
• mu: Mean of the data.
• sigma: Standard deviation of the data.
• gamma: Skewness of the data.
• alpha: Skewness weighting factor (default: 1.0).
• beta: Weighting factor for the non-linear transformation (default: 0.5).
• k: Scaling factor for the non-linear term (default: 1.0).

---

How to Use

Installation

Install the library from PyPI:

pip install skewnormlib

Usage Example

import numpy as np
from skewnorm.normalization import SkewWeightedNormalization
from sklearn.model_selection import train_test_split

# Sample data
data = np.random.rand(100, 5)  # 100 samples with 5 features
X_train, X_test = train_test_split(data, test_size=0.2)

# Initialize the transformer
swn = SkewWeightedNormalization(alpha=1.0, beta=0.5, k=1.0)

# Fit and transform the training data
X_train_transformed = swn.fit_transform(X_train)

# Transform the test data
X_test_transformed = swn.transform(X_test)

# Reverse the transformation (if needed)
X_test_original = swn.inverse_transform(X_test_transformed)

print("X_train_transformed shape:", X_train_transformed.shape)
print("X_test_transformed shape:", X_test_transformed.shape)
print("Original test data recovered shape:", X_test_original.shape)

---

Advantages

• Handles Skewness: Unlike traditional normalization techniques, this method adjusts for skewness, making it suitable for heavily skewed datasets.
• Non-linear Transformation: The additional (\tanh) term helps reduce the influence of extreme outliers.
• Scikit-learn Compatible: The class adheres to Scikit-learn's API, allowing seamless integration into pipelines.
• Open Source: The library is open for contributions, making it a community-driven project.

---

Disadvantages

• Complexity: The additional parameters ((\alpha, \beta, k)) require tuning for optimal results.
• Performance: Slightly slower than standard normalization techniques due to the computation of skewness and the non-linear term.

---

When to Use It

• Skewed Data: Use this method when the dataset has significant skewness, as it normalizes while accounting for the skew.
• Outlier Sensitivity: When datasets contain extreme outliers that might adversely affect models, this normalization technique can help mitigate their influence.
• Preprocessing for ML: Use this as a preprocessing step for machine learning models that perform better on normalized data (e.g., SVM, Neural Networks).

---

Contributing

This library is open source, and contributions are welcome! Feel free to:

1. Submit bug reports.
2. Suggest new features.
3. Improve documentation.
4. Optimize performance.

Visit the GitHub repository to contribute: GitHub Repository

---

License

This project is licensed under the MIT License, ensuring it remains free and open for everyone.

---

Contact

For questions or suggestions, contact:

• Author: Mohammed Nihal
• Email:[e-mail]