[ROADMAP] Transitioning from Predictive Modeling to Sequential Decision Analytics & Production Deployment

[AngelFelizR]

Description:

This issue tracks the final stages of the NycTaxi project. The goal is to move beyond static predictions and implement a production-grade Sequential Decision Analytics (SDA) framework using the Warren Powell framework, encapsulated in a reproducible Nix/Docker environment and served via a Plumber API.

---

Phase 1: Decision Science & Policy Optimization (The "Science" Part)

• Step 09: From Predictions to Policies: Integrating ML into Stochastic Optimization

  • Refine XGBoost integration: Transition from pure classification to a profit-driven decision rule.
  • Threshold Optimization: Execute grid search/sampling to identify the optimal threshold ($\theta$) that maximizes Net Benefit under stochastic uncertainty.
  • Document the "Predictions to Policies" transition in a Quarto technical article.

• Step 10: Strategic $S_0$: Designing a Policy Function Approximation for Optimal Starting States

  • Analyze baseline data to derive the optimal starting zone based on (date, time, company, vehicle_type).
  • Develop a PFA for the initial state ($S_0$) strategic decision.
  • Benchmark the incremental gain: "Optimal Start Zone" vs. "Random Start Zone".

---

Phase 2: Production-Grade SDA: Serving Stochastic Policies with Plumber and Nix

• Development of Plumber API

  • Endpoint POST /decide_trip: Real-time TRUE/FALSE decision engine based on optimized thresholds.
  • Endpoint GET /optimal_start: Strategic recommendation engine for initial deployment.
  • Robustness: Implement production-grade validation via {pointblank} or {tidyvalidate}.

• Containerization & MLOps

  • Reproducibility: Author a Nix Flake to lock all dependencies and system libraries.
  • Portability: Architect a Docker image containing the Nix-built environment and serialized models.
  • Deployment: Host the service on Render to provide a public, authenticated API.

---

Phase 3: Interactive SDA: A Visual Interface for NYC Taxi Policy Analysis

• Shiny App Upgrade: Decision Support System (DSS)
  • Backend Integration: Connect the UI to the live Plumber API for real-time inference.
  • Spatial Analytics: Leaflet map visualizing acceptance probabilities across NYC zones.
  • Temporal Dynamics: Implement a Bump Chart/Ranking Heatmap showing the "Best Start Zone" shift throughout the day.
  • Sensitivity Analysis: Add a slider for the decision threshold to show users the trade-off between total trips accepted and net hourly earnings.

---

Phase 4: Final Documentation & Career Positioning

• Portfolio Integration

  • Finalize index.qmd with the end-to-end workflow and live app links.
  • Record a Loom demo focusing on the SDA framework and economic impact.

• Professional Brand Sync

  • Update CV/LinkedIn: "Architected a full-cycle Decision Support System using SDA, reducing uncertainty in trip selection and increasing expected hourly earnings."