Time-Series Validation (Notebook)

Use information about previous time points to predict a subsequent time point—correctly.
This notebook demonstrates:

• Feature engineering for prices (e.g., percent_change)
• Stationarity vs non-stationarity setups (and why it matters)
• Proper time-aware validation with TimeSeriesSplit
• Baselines: LinearRegression and Ridge
• Metrics: r2_score, cross_val_score
• Bootstrap confidence intervals for validation scores
• Helpful plots: coefficient bars & predictions vs. actual

---

Contents

• Data loading: CSVs like
  chap4_X.csv, chap4_y.csv, chap4_X_stationarity.csv, chap4_y_stationarity.csv,
  non_stationarity_X.csv, non_stationarity_y.csv,
  prices_of_four.csv, prices_aapl_yhoo_nvda.csv, test_y.csv
• Utilities:
  • percent_change(series) — simple return-like feature
  • visualize_coefficients(model, feature_names)
  • visualize_predictions(y_true, y_pred)
  • my_pearsonr(x, y) — custom correlation
  • bootstrap_interval(values, n_boot=1000, ci=0.95) — metric CI
• Validation:
  • TimeSeriesSplit (expanding/rolling windows)
  • No shuffling; preserves temporal order
• Models:
  • sklearn.linear_model.LinearRegression
  • sklearn.linear_model.Ridge

---

Getting Started

# 1) Create/activate a virtual environment
python -m venv .venv
# Windows
.venv\Scripts\activate
# macOS/Linux
source .venv/bin/activate

# 2) Install dependencies
pip install -r requirements.txt

# 3) Launch Jupyter
jupyter lab   # or: jupyter notebook

# 4) Git Clone
git clone https://github.com/Joe-Naz01/time_series_valid.git
cd time_series_valid