ml-library

A personal machine learning library featuring implementations of core models from scratch, without relying on high-level ML libraries. Built to deepen my understanding of machine learning fundamentals and demonstrate practical, hands-on knowledge.

An example usage of the features of the library can be seen in example_mnist_cnn.ipynb file.

Installation

Install the necessary dependencies using the requirements file:

pip install -r requirements.txt

Table of Contents

• Modules
  • Neural Networks (NN)
  • Models
  • Utilities

Modules

Neural Networks (NN)

This submodule contains components to create and train neural networks.

• BaseModel: A base class for neural network models with common methods.

  • BaseModel.py

• Layers: Different layers used in neural networks.

  • Implemented Layers:
    • Linear
    • Convolutional 2D
    • Max pooling 2D
    • Dropout
  • layers.py

• Activation Functions: Common activation functions used in neural networks.

  • Implemented Activation functions:
    • [Tanh, Sigmoid, ReLu]
  • activations.py

• Loss Functions: A collection of loss functions.

  • Implemented Loss Functions:
    • [MSE_loss, BCE_loss, Categorical cross entropy loss]
  • loss_functions.py

• Optimizers: Optimizers used to update model parameters during training.

  • Implemented Optimizers:
    • [GDOptimizer]
  • optimizers.py

Models

The library currently supports various machine learning models:

• Linear Regression
  • A Linear Regression model that uses gradient descent to update the parameters.
  • linear_regression.py
• Logistic Regression
  • A Logistic Regression model that uses stochastic gradient descent to update the parameters.
  • logistic_regression.py
• Primal Support Vector Machine
  • A Support Vector Machine model implemented in primal representation that uses stochastic gradient descent to update the parameters.
  • primal_svm.py
• Dual Support Vector Machine
  • A Support Vector Machine model implemented in dual representation that uses stochastic dual coordinate ascent to update the parameters.
  • The model features linear, polynomial, radial basis function (RBF) and Gaussian kernels to allow flexibility in classification tasks.
  • dual_svm.py

Utilities

Utility functions for preprocessing and evaluation.

• Preprocessing: Includes data preprocessing utilities.

  • Implemented Features:
    • [train_test_split, train_val_test_split]
  • preprocessing.py

• Metrics: Includes utility functions for evaluation metrics.

  • Implemented Features:
    • [mean_squared_error, accuracy, precision, recall, f1_score]
  • metrics.py

Contact

For questions and feedback, feel free to reach out:

• Email: mk.junttila@gmail.com