TDistributionClassifier 1.2.0

A binary classifier using Student's t-distribution for univariate and multivariate continuous data.

TDistributionClassifier

Author: Abdul Mofique Siddiqui
License: MIT
Install via pip:

pip install TDistributionClassifier

Import it in your Python code:

from TDistributionClassifier import TDistributionClassifier

---

Overview

TDistributionClassifier is a machine learning classifier designed to model classes using Student's t-distribution. This classifier is robust to outliers and performs effectively when data contains noise or extreme values. It is a versatile tool capable of handling both binary and multiclass classification tasks for both univariate (1D) and multivariate (multi-dimensional) data.

---

Installation

Install the package via pip:

pip install tdistributionclassifier

---

How It Works

• Univariate Mode: For 1D features, each class is modeled using a univariate t-distribution.
• Multivariate Mode: For multi-dimensional features, each class is modeled using a multivariate t-distribution.
• Uses log-probabilities and the log-sum-exp trick for numerical stability.
• Automatically detects the input dimensionality and selects the appropriate mode.

---

Getting Started

1. Import the package

from TDistributionClassifier import TDistributionClassifier

2. Initialize the classifier

clf = TDistributionClassifier()

3. Fit the model

clf.fit(X_train, y_train)

• X_train: numpy array of shape (n_samples,) or (n_samples, n_features)
• y_train: class labels

4. Predict class probabilities

probs = clf.predict_proba(X_test)

• Returns a numpy array of shape (n_samples, n_classes) with class probabilities.

5. Predict class labels

labels = clf.predict(X_test)

• Returns predicted class labels

---

API Reference

TDistributionClassifier()

Initializes the classifier. No arguments required.

---

.fit(X, y)

Fits the model to the training data.

• Parameters:
  • X: numpy array of training features. Shape: (n_samples,) or (n_samples, n_features)
  • y: class labels. Shape: (n_samples,)

---

.predict_proba(X)

Returns predicted class probabilities.

• Input:
  • X: Features. Shape: (n_samples,) or (n_samples, n_features)
• Output:
  • probs: array of shape (n_samples, n_classes) with class probabilities

---

.predict(X)

Returns predicted class labels based on highest probability.

• Input:
  • X: Features. Shape: (n_samples,) or (n_samples, n_features)
• Output:
  • labels: array of shape (n_samples,) with class labels

---

Example Usage

Example 1: Binary Classification

from TDistributionClassifier import TDistributionClassifier
from sklearn.datasets import load_diabetes
from sklearn.model_selection import train_test_split

# Load data and binarize target
data = load_diabetes()
X = data.data
y = (data.target > 100).astype(int)  # Binary classification

# Split dataset
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3)

# Initialize and train
clf = TDistributionClassifier()
clf.fit(X_train, y_train)

# Predict
probs = clf.predict_proba(X_test)
preds = clf.predict(X_test)

Example 2: Multiclass Classification

from TDistributionClassifier import TDistributionClassifier
from sklearn.datasets import load_iris
from sklearn.model_selection import train_test_split

# Load the Iris dataset (3 classes)
data = load_iris()
X = data.data
y = data.target  # Multiclass labels (3 classes)

# Split dataset
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3)

# Initialize and train
clf = TDistributionClassifier()
clf.fit(X_train, y_train)

# Predict
probs = clf.predict_proba(X_test)
labels = clf.predict(X_test)

---

Internals

• PDF Estimation: Uses scipy.stats.t (univariate) or scipy.stats.multivariate_t (multivariate).
• Regularization: Adds small noise (1e-6 * I) to covariance matrices to ensure invertibility.
• Numerical Stability: Log-probabilities with log-sum-exp used for probability normalization.

---

Notes

• Supports both binary and multiclass classification.
• Multivariate mode is triggered when input has >1 features.
• The Student's t-distribution's heavy tails make it more resilient to extreme values compared to traditional Gaussian-based models.
• If data is not linearly separable, consider applying feature transformation or dimensionality reduction before use.

---

Author

Abdul Mofique Siddiqui

---

License

This project is licensed under the MIT License.