RiskLabAI.jl — Tutorial Notebooks

Runnable Julia tutorials for RiskLabAI.jl, the financial-machine-learning library based on López de Prado's Advances in Financial Machine Learning and Machine Learning for Asset Managers. Each notebook is exhaustive over its module, runs top-to-bottom against a pinned environment, and mirrors the matching Python tutorial in Notebooks.py.

Getting started

New to this? This gets you from a fresh computer to running any tutorial. You only do Step 1 once.

Step 1 — install the tools (once)

1a. VS Code. Download and install Visual Studio Code (Windows or macOS) and open it once. Then install these extensions (left sidebar → Extensions icon → search → Install):

• Julia (publisher: julialang)
• Jupyter (publisher: Microsoft)

1c. Julia. Install Julia via the official installer (juliaup):

• Windows: winget install julia -s msstore in a terminal, or use the installer on that page.
• macOS: paste curl -fsSL https://install.julialang.org | sh into Terminal, or brew install juliaup.

Confirm it works: open a new terminal and run julia --version.

Step 2 — run a notebook

1. Add the Julia kernel for notebooks (once). In a terminal:

   julia -e "using Pkg; Pkg.add(\"IJulia\")"

   This lets VS Code / Jupyter run Julia notebooks.
2. Get the notebooks. Go to https://github.com/RiskLabAI/Notebooks.jl → green Code button → Download ZIP → unzip somewhere easy. (Or, with git: git clone https://github.com/RiskLabAI/Notebooks.jl.)
3. Open the folder in VS Code: File → Open Folder → choose the Notebooks.jl folder.
4. Install the packages. Open a terminal in VS Code (Terminal → New Terminal) and run:

   julia --project=. -e "using Pkg; Pkg.instantiate()"

   This installs RiskLabAI plus the plotting/data packages pinned for the notebooks. (First time compiles for several minutes — normal for Julia.)
5. Open a notebook (e.g. Optimizing/01_portfolio_construction.ipynb), click Select Kernel (top-right) → Julia, then click Run All.

The FRED API key (two notebooks)

The fractional differentiation and triple-barrier labeling tutorials pull real data from FRED (the Federal Reserve's free database). To run them:

1. Get a free API key (instant): https://fredaccount.stlouisfed.org/apikey
2. Set it as an environment variable, then open a new terminal so it takes effect:
   • Windows: setx FRED_API_KEY "your_key_here"
   • macOS: add export FRED_API_KEY="your_key_here" to ~/.zshrc, then source ~/.zshrc
3. Verify in a new terminal: echo %FRED_API_KEY% (Windows) or echo $FRED_API_KEY (macOS).

All other tutorials run fully offline and need no key.

Optional — the PDE notebook only

The Deep-BSDE notebook (Pde/DeepBSDE.ipynb) uses the Lux/Zygote/Optimisers deep-learning stack. It arrives automatically with RiskLabAI (no extra install) — just run the notebook top-to-bottom.

Flagship tutorials

#	Notebook	Topic
1	Optimizing/01_portfolio_construction.ipynb	Robust portfolio construction — covariance denoising (Marčenko–Pastur), HRP, and NCO vs naive Markowitz
2	Data/Structures/FinancialDataStructures.ipynb	Financial data structures — time, tick, volume, dollar, imbalance & run bars (offline AAPL data)
3	Data/Diffrentation/FractionalDifferentiation.ipynb	Fractional differentiation — the minimum d for stationarity via an ADF sweep (FRED S&P 500)
4	Data/Labelling/TripleBarrierLabeling.ipynb	Triple-barrier labeling + meta-labeling + trend scanning (FRED S&P 500)
5	Validation/CrossValidationAndPBO.ipynb	Cross-validation — purged K-Fold, CPCV, walk-forward, and the Probability of Backtest Overfitting (PBO)
6	Features/FeatureImportance.ipynb	Feature importance — MDI, MDA, SFI, orthogonal (PCA) features, and weighted Kendall-τ
7	Pde/DeepBSDE.ipynb	Deep-BSDE PDE solver (Han, Jentzen & E 2018) — four financial PDEs, checked vs Monte-Carlo / closed-form references

Stage-1 method tutorials

Tutorials for the admitted library-extension methods, each framed baseline → where it breaks → admitted method → when to use / when NOT (the "when to use" takeaway is the method's regime tag from its admitting appraisal). All run offline (synthetic data

• the committed AAPL snapshot).

Notebook	Topic
Features/Microstructure_EDGE.ipynb	EDGE bid-ask spread estimator (Ardia–Guidotti–Kroencke 2024) vs Roll / Corwin–Schultz
Features/Structural Breaks/GSADF_MultipleBubbles.ipynb	GSADF / BSADF multiple-bubble detection & date-stamping (Phillips–Shi–Yu 2015) vs single-window SADF
Features/EntropyBiasCorrection.ipynb	Bias-corrected entropy — Miller–Madow, Grassberger, NSB vs the plug-in under undersampling
Features/DebiasedImportance_MDIplus_CPI.ipynb	MDI+ (Agarwal 2023) & CPI (Watson–Wright 2021) vs MDI/MDA
Data/Distance/KSG_DistanceCorrelation.ipynb	KSG kNN mutual information & distance correlation vs binned MI
Data/Diffrentation/AdaptiveFractionalDifferentiation.ipynb	Adaptive Fractional Differencing vs de Prado's min-d-via-ADF
Backtest/MultipleTesting_and_SharpeInference.ipynb	Holm / BHY Sharpe haircuts (Harvey–Liu 2015) & LPLZ HAC Sharpe inference (2025)
Backtest/RobustStatistics_CED_SharpeDiff.ipynb	Conditional Expected Drawdown & the Ledoit–Wolf Sharpe-difference test
Backtest/OUTradingRules.ipynb	Closed-form OU trading rules (Lipton–López de Prado 2020) vs the Monte-Carlo PT/SL grid
Validation/PathLevelCPCV.ipynb	Path-level Bagged & Adaptive CPCV (Arian–Norouzi–Seco 2024) vs plain CPCV
Validation/LeakageAwareHPO.ipynb	Leakage-aware HPO through purged CV with a Deflated-Sharpe gate

Setup

RiskLabAI is pulled from its public repository — no local clone of the library needed. From this folder:

using Pkg
Pkg.activate(".")
Pkg.instantiate()      # fetches RiskLabAI + deps and precompiles the env

Once RiskLabAI.jl is published to the General registry, the [sources] entry in Project.toml can be dropped and the dep becomes a plain Pkg.add("RiskLabAI").

Then open any notebook with IJulia, or run it headless:

jupyter nbconvert --to notebook --execute --inplace <path-to-notebook>

The Deep-BSDE PDE notebook (Pde/DeepBSDE.ipynb) uses the Lux/Zygote/Optimisers deep-learning stack; it arrives transitively with RiskLabAI, so no extra install step is required — only that one notebook exercises it.

Tutorials 3–4 read the S&P 500 from FRED; set the FRED_API_KEY environment variable (a free key from https://fred.stlouisfed.org). All other data is either synthetic or the committed Data/AAPL_OHLCV_1m_Data.csv snapshot, so the notebooks stay reproducible and offline.