Machine Learning Projects
PyTorch · Python · scikit-learn · Data Mining
Overview
A collection of machine learning experiments spanning classical statistical learning through modern deep neural networks. These projects were built to develop intuition for model behavior, generalization, and the practical tradeoffs that matter in production ML systems.
Experiments
Classification Benchmarks
Systematic comparison of classifiers — logistic regression, gradient boosted trees,
and feed-forward networks — on tabular datasets. Emphasis on calibration, class
imbalance, and feature importance analysis.
Data Mining & Feature Engineering
Exploratory pipelines for structured data: outlier detection, dimensionality reduction
(PCA, UMAP), and unsupervised clustering (k-means, DBSCAN). Applied to real-world
datasets with messy distributions.
Neural Network Experimentation
From-scratch implementations of core building blocks (attention, normalization layers,
residual connections) in PyTorch to develop low-level understanding before working with
higher-level frameworks.
Optimization & Training Dynamics
Experiments studying learning rate schedules, gradient flow, and the effect of
batch size on convergence — motivated by a desire to understand why training recipes
work before trusting them blindly.