dl-training: A Quant Finance Learning Repo
I recently put together dl-training on GitHub — a home for learning quantitative finance and deep learning on financial data. The name reflects where it started (deep learning on price series), but the repo has grown into a broader set of Jupyter notebooks for people who want to build intuition with runnable code rather than read theory in isolation.
What’s in the repo
The notebooks fall into a few tracks:
- Quant Finance from Zero — a single long notebook that assumes no prior exposure. It walks from basic Python and NumPy through calculus, linear algebra, probability, and stochastic calculus, ending with capstone exercises (Monte Carlo, a portfolio optimizer, a volatility surface). Every concept is tied to a concrete trading or risk question.
- Deep Learning on Financial Data — builds a small neural network from scratch in plain NumPy on synthetic daily returns, then rebuilds it in PyTorch and runs an honest evaluation against simple baselines. Synthetic data is deliberate: a single stock price bar is almost useless for learning (it mostly goes up or down), but a controlled AR(1) process lets you verify the network actually picks up planted momentum before you swap in real yfinance data.
- Graduate Probability — measure-theoretic definitions paired with Monte Carlo checks: probability spaces, Bayes, discrete and continuous random variables, expectation and covariance.
- Shreve weekly series — fourteen notebooks aligned with Stochastic Calculus for Finance II, each pairing Shreve’s rigor with Baxter & Rennie market intuition and Python simulation (random walks, filtrations, martingales, Itô’s lemma, Black–Scholes, risk-neutral pricing, jump diffusions, and exotics).
- Market data scratchpad — a
training.ipynbnotebook with yfinance examples for pulling single-ticker and multi-ticker OHLC history.
Who it’s for
The repo is useful if you are:
- Breaking into quant finance and want a structured path from programming basics to options math.
- Already technical but missing the finance side — the Shreve track gives you chapter pointers and working simulations alongside the textbook.
- Learning deep learning and want a domain where leakage, baselines, and “does this beat always predicting up?” actually matter.
- Preparing for graduate coursework and prefer to see theorems verified numerically before trusting them.
How to use it
Clone the repo, create a Python 3.12+ environment, and install the dependencies listed in pyproject.toml (NumPy, pandas, SciPy, matplotlib, yfinance, and the usual scientific stack). Open any notebook in Jupyter or VS Code and run cells top to bottom. The notebooks are designed to work offline where possible — synthetic data is the default, and real-market downloads are optional.
New material will land here as I write it. If a notebook saves you time or you spot something wrong, issues and pull requests on GitHub are welcome.