ML-Architectures is a structured knowledge project designed to deeply explain how classic machine learning models work — from their internal mathematical structures, loss functions, optimization processes, to complete training pipelines.
This project helps you understand:
- What exactly a model like Linear Regression, SVM, or XGBoost is doing internally
- How gradient-based optimization adjusts parameters
- How to build, train, and evaluate models from scratch
- The core intuition behind loss functions and optimizers
ML-Architectures/
├── models/ # Each ML model has its own module
│ ├── linear_regression/
│ ├── logistic_regression/
│ ├── decision_tree/
│ ├── random_forest/
│ ├── xgboost/
│ ├── knn/
│ ├── svm/
│ ├── kmeans/
│ └── neural_network/
├── optimizers/ # Optimization methods explained
│ ├── gradient_descent.md
│ ├── sgd.md
│ ├── adam.md
│ └── comparison.md
├── notes/ # Key concepts: loss functions, regularization, etc.
│ ├── loss_functions.md
│ ├── regularization.md
│ └── training_pipeline.md
├── diagrams/ # Visual explanations, architecture charts
├── LICENSE
└── README.md
- ML learners who want to go beyond APIs and understand what’s under the hood
- Aspiring researchers, data scientists, or quant developers
- Interview candidates preparing for deep algorithm/model questions
- Linear Regression (✅ in progress)
- Logistic Regression
- Decision Tree & Random Forest
- KNN
- XGBoost / LightGBM
- Support Vector Machines
- KMeans (unsupervised)
- Neural Networks (basic MLP)
- Core math implemented in Python/Numpy
- Visualizations via Matplotlib & Plotly
- Optional PyTorch modules for more complex experiments
To build a model-by-model, optimizer-by-optimizer understanding of machine learning from the ground up — as simply, deeply, and reproducibly as possible.
This project is licensed under the MIT License.
© 2025 Pengyi Peng. All rights reserved.