Deep Learning Basics with PyTorch

Based on Deep Learning with PyTorch (O’Reilly, forthcoming) by Dr. Yves Hilpisch

---

---

Overview

This repository accompanies a self-directed study and applied implementation of the pre-release manuscript of Deep Learning with PyTorch by Dr. Yves Hilpisch (O’Reilly, forthcoming).

It reconstructs and extends the book’s exercises, linking classical machine-learning foundations to modern deep-learning practice in a fully reproducible environment.
Each notebook forms part of a progressive learning path—from NumPy fundamentals and regression models to PyTorch-based neural networks, optimisation routines, and training diagnostics.

---

Learning Outcomes

By following the notebooks, you will:

• Understand the mathematical foundations of regression and classification.
• Implement, train, and evaluate neural networks using PyTorch.
• Compare classical ML vs. deep-learning behaviour under varying hyperparameters.
• Apply core training techniques: SGD, Adam, early stopping, gradient clipping, LR scheduling, and mini-batch optimisation.
• Develop clear intuition for bias–variance trade-offs and convergence dynamics through hands-on experiments.

---

Repository Structure

├── part1_foundations
│   ├── chapter_1.ipynb              # NumPy foundations & linear algebra
│   ├── chapter_2.ipynb              # Linear & Ridge Regression
│   ├── chapter_3.ipynb              # Classification (LogReg, SVM, Trees)
│   ├── chapter_4.ipynb              # Overfitting & Learning Curves
│   ├── capstone_california_housing  # Feature engineering & MLP
│   └── exercises_challenges         # Review exercises & solutions
│
├── part2_pytorch_basics
│   ├── chapter_5.ipynb              # Introduction to PyTorch
│   ├── chapter_6.ipynb              # Tensors, Autograd, and Training Loops
│   ├── chapter_7.ipynb    			 # Training Tiny Networks 
│   └── exercises_challenges         # Collected solutions for Part II
│
├── data/
│   └── adr_prices_and_vol.csv       # Finance-oriented sample dataset
│
└── README.md

Note:
In addition to standard sample datasets referenced in the book (e.g., iris, make_moons, california_housing), this repository includes a small supplementary file — adr_prices_and_vol.csv — used for exploratory testing of financial data workflows and to illustrate how PyTorch and scikit-learn models can be applied to quantitative finance use cases.

---

Environment Setup

Create and activate environment

conda create -n pytorch_dl python=3.12 -y
conda activate pytorch_dl

Install core dependencies

conda install numpy pandas matplotlib seaborn scikit-learn statsmodels scipy numba jupyterlab notebook ipykernel -y

💡 PyTorch and related libraries are introduced in Part II (from Chapter 5).
Install via the official instructions for your OS/GPU configuration:
https://pytorch.org/get-started/

Launch JupyterLab

jupyter lab

---

Chapter Progress

Part	Chapter	Focus	Status
I	1	NumPy & Linear Algebra Foundations	✅
I	2	Linear & Ridge Regression	✅
I	3	Classification & Ensemble Models	✅
I	4	Overfitting & Complexity Control	✅
II	5	PyTorch Neural-Network Fundamentals	✅
II	6	Autograd, Optimisers, Training Loops	✅
II	7	Training Neural Networks	✅
II	7	Exercises & Challenges: Training Tiny Networks	✅ (Full)

---

Development Notes

• All notebooks adhere to reproducible, educational research standards.
• Each section documents hyperparameters, metrics, and plots inline.
• The environment is fully Conda-based for cross-platform reproducibility.

---

Citation & Attribution

Hilpisch, Y. (2025, forthcoming). Deep Learning with PyTorch. O’Reilly Media.
Original teaching materials © Dr. Yves Hilpisch / The Python Quants GmbH.
Adaptations © 2025 Francisco Salazar — academic, non-commercial use only.

---

License

This repository is distributed under an Educational-Use License for research and learning purposes.
It is not affiliated with or endorsed by O’Reilly Media or The Python Quants GmbH.

---

Maintained by: Francisco Salazar
📅 Last Updated: October 2025