ddbourgin/numpy-ml
Numpy Ml
This library is a collection of machine learning algorithms and neural network components implemented from scratch using only NumPy. It serves as an educational toolkit for constructing and experimenting with machine learning architectures, emphasizing a modular approach where algorithms are organized into self-contained, object-oriented classes.
The project distinguishes itself by relying exclusively on array-oriented programming to perform mathematical operations, ensuring that all computations are vectorized for performance. By utilizing a standardized interface for forward and backward passes, the library allows users to chain independent components to build custom models without the overhead of larger, more complex frameworks.
The software covers a broad range of data science capabilities, including linear and probabilistic modeling, matrix factorization, and tree-based learning. It also provides specialized tools for nonparametric estimation, reinforcement learning, and deep learning, alongside a suite of preprocessing utilities for signal and text data.
Features
- Machine Learning Libraries - Implements machine learning algorithms and neural network components from scratch using only NumPy.
- Deep Learning Toolkits - Provides a modular toolkit for constructing deep learning architectures and neural networks.
- Vectorized Array Operations - Performs all mathematical operations using vectorized array-oriented programming for high performance.
- Ensemble Methods - Supports classification and regression tasks through random forests and gradient-boosted models for predictive analysis on structured data.
- Probabilistic Models - Implements statistical models including Gaussian mixture models, Hidden Markov models, and Latent Dirichlet allocation for sequence analysis and pattern discovery.
- Reinforcement Learning - Optimizes decision-making in dynamic environments using multi-armed bandit strategies and temporal difference learning methods.
- Gradient Optimization Techniques - Implements gradient-based optimization techniques to update model parameters and minimize prediction error.
- Machine Learning Prototyping - Facilitates rapid experimentation and prototyping of custom machine learning architectures from scratch.
- Neural Network Layers - Assembles deep learning architectures using modular layers, activation functions, and optimizers.
- General Machine Learning - Reference implementations of ML models in NumPy.
- Nonparametric Models - Implements flexible nonparametric predictive models like kernel regression and Gaussian processes.
- Modular Framework Compositions - Assembles complex models by chaining independent, modular components with standardized interfaces.
- Kernel Density Estimation - Models complex data patterns using kernel regression, k-nearest neighbors, and Gaussian processes.
- Linear Regression - Calculates regression and classification results using ordinary least squares, ridge, and logistic methods.
- Probabilistic Modeling - Provides a suite of probabilistic and linear models for classification, regression, and feature extraction.
- Data Preprocessing Utilities - Provides utilities for transforming raw signals and text into structured formats for machine learning.
- Object-Oriented Programming - Organizes algorithms into self-contained, object-oriented classes for state and hyperparameter management.
- Interaction Matrix Factorizers - Extracts latent features from large datasets using alternating least-squares and non-negative matrix factorization.
- Data Preprocessing - Transforms raw inputs through signal processing, tokenization, and feature encoding for algorithmic analysis.
- Statistical Analysis - Applies linear and probabilistic techniques to identify relationships and structures in datasets.
- Training Convergence Optimization - Refines model parameters through iterative cycles of prediction and error correction.
