| --- |
| title: DoE & Bayesian Optimization Dashboard |
| emoji: 🔬 |
| colorFrom: indigo |
| colorTo: purple |
| sdk: gradio |
| sdk_version: "6.13.0" |
| app_file: app.py |
| pinned: true |
| license: mit |
| short_description: Design experiments & get AI-guided suggestions |
| tags: |
| - bayesian-optimization |
| - design-of-experiments |
| - gaussian-process |
| - machine-learning |
| - scientific-computing |
| --- |
| |
| # 🔬 Design of Experiments & Bayesian Optimization Dashboard |
|
|
| A comprehensive tool for experimentalists to plan experiments, fit surrogate models, and get AI-guided suggestions for the next most informative experiments. |
|
|
| ## Features |
|
|
| ### 📐 Design of Experiments |
| - **Full Factorial** — All combinations of factor levels |
| - **Fractional Factorial (½, ¼)** — Reduced experiment count for screening |
| - **Latin Hypercube Sampling (LHS)** — Space-filling design for BO initialization |
| - **Sobol Sequence** — Low-discrepancy quasi-random design |
| - **Central Composite Design (CCD)** — Response surface methodology |
| - **Box-Behnken** — Efficient 3+ factor designs |
| - Flexible factor definition: comma-separated levels or range notation (`10:30:3`) |
| - Add/remove rows and columns dynamically |
| - CSV export |
|
|
| ### 🤖 Surrogate Models |
| - **Gaussian Process** with 5 kernel choices (Matérn 5/2, Matérn 3/2, RBF, Rational Quadratic, RBF+Matérn) |
| - **Random Forest** with tree-ensemble uncertainty |
| - **Extra Trees** — faster RF variant |
| - **Gradient Boosting** with staged prediction uncertainty |
|
|
| ### 🎯 Acquisition Functions |
| - **Expected Improvement (EI)** — balanced exploration/exploitation |
| - **Probability of Improvement (PI)** — greedy exploitation |
| - **Upper Confidence Bound (UCB)** — tunable exploration weight |
| - **Thompson Sampling** — stochastic batch selection |
| - Diverse suggestion selection (avoids clustering) |
|
|
| ### 📊 12 Visualization Types (all toggleable) |
| 1. 🗺️ Surrogate Surface (2D contour with mean + uncertainty) |
| 2. 📉 1D Parameter Slices (effect of each factor) |
| 3. 🏔️ 3D Response Surface (interactive Plotly) |
| 4. 🎯 Acquisition Function Landscape |
| 5. 📈 Convergence Plot (best-so-far trajectory) |
| 6. ⚖️ Predicted vs Actual (parity plot with R²/RMSE) |
| 7. 📊 Feature Importance (GP lengthscales / RF importances) |
| 8. 📏 Residual Analysis (residuals, histogram, Q-Q plot) |
| 9. 🔀 Parallel Coordinates (interactive Plotly) |
| 10. 🌡️ Uncertainty Heatmap |
| 11. 🔗 Correlation Matrix |
| 12. 🔘 Scatter Matrix (pair plot) |
|
|
| ## Workflow |
|
|
| 1. **Define Factors** → Set experimental parameters and levels |
| 2. **Generate Design** → Choose a design type (factorial, LHS, CCD, etc.) |
| 3. **Run Experiments** → Fill in the Response column with results |
| 4. **Fit Surrogate** → Configure model (GP/RF/ET/GB) and acquisition function |
| 5. **Get Suggestions** → Receive optimized next-experiment recommendations |
| 6. **Iterate** → Add suggestions to table, run them, repeat |
|
|
| ## Tech Stack |
| - Gradio (UI), scikit-learn (GP, RF, ET, GB), scikit-optimize, scipy (LHS, Sobol, acquisition optimization), Plotly + Matplotlib (visualizations) |
|
|
