create-playground（创建游乐场）

Create Playground Skill

Create and maintain an interactive Streamlit playground that enables "learn by playing" instead of "learn by reading code."

Goal

Build and evolve a web-based playground where users can:

1. Explore VRP problems interactively (select, generate, visualize instances)
2. Experiment with algorithms (configure parameters, run solvers, compare results)
3. Learn through interaction (understand interfaces, pipelines, mechanisms)
4. Reproduce experiments (save configs, replay runs, export results)

Core Philosophy

Following the vision in playground/VISION.md:

• Three-layer learning: Interface → Pipeline → Mechanism
• Minimal cognitive load: Only expose what's needed for current task
• Contract-based trust: Playground behavior matches actual code (verified by tests)
• Just-in-time learning: Dive deeper only when hitting limitations

Workflow

Step 1: Analyze User Request

Understand what the user wants to learn or build:

Questions to ask:

• What feature/algorithm do you want to explore?
• Which parameters are most important?
• What level of detail (beginner/intermediate/advanced)?
• What kind of visualization helps understanding?

Common requests:

• "Add support for CVRP problems"
• "Show how temperature affects ALNS search"
• "Visualize operator impact step-by-step"
• "Compare two algorithm configurations"

Step 2: Design UI/UX

Choose appropriate Streamlit components:

Reference: references/ui_components.md for patterns

Component selection guide:

• Parameters: st.slider, st.number_input, st.selectbox
• Problem definition: st.file_uploader, st.radio, st.multiselect
• Visualization: st.pyplot, st.plotly_chart, st.map
• Results: st.dataframe, st.metric, st.json
• Layout: st.columns, st.tabs, st.expander

Progressive disclosure:

• Start with 5-10 key parameters
• Hide advanced parameters in st.expander("Advanced")
• Use defaults for 80% use cases

Step 3: Integrate VRP-Toolkit Modules

Map playground interactions to toolkit APIs:

Reference: references/integration_patterns.md for examples

Integration checklist:

• Import correct modules (from vrp_toolkit.problems import...)
• Convert UI inputs to API format (e.g., sliders → config dict)
• Handle errors gracefully (try-except with user-friendly messages)
• Extract outputs for display (solution → routes, cost, metrics)

Key integration points:

1. Problem layer: PDPTWInstance, VRPProblem, etc.
2. Algorithm layer: ALNSSolver, ALNSConfig, etc.
3. Data layer: OrderGenerator, DemandGenerator, RealMap
4. Visualization layer: PDPTWVisualizer, route plotting

Step 4: Implement Visualization

Make algorithm behavior visible:

Visualization types:

• Route maps: Show vehicle routes on 2D map with nodes/edges
• Convergence plots: Cost vs. iteration (line chart)
• Operator impact: Before/after comparison (side-by-side)
• Metrics dashboard: Cost breakdown, constraint violations, runtime

Best practices:

• Use existing vrp_toolkit.visualization modules when possible
• Add interactive elements (hover for details, zoom, pan)
• Color-code for clarity (feasible=green, infeasible=red)
• Include legends and axis labels

Step 5: Add Contract Tests

Ensure playground behavior matches actual code:

Reference: Create tests in contracts/ directory

Critical contracts to test:

1. Reproducibility: Same seed + same config → same result def test_reproducibility(): config = {...} result1 = run_with_seed(config, seed=42) result2 = run_with_seed(config, seed=42) assert result1 == result2
2. Feasibility: Playground claims "feasible" → solution actually feasible def test_feasibility_contract(): solution = playground_run(...) if playground_says_feasible(solution): assert actually_feasible(solution)
3. Evaluation consistency: Playground displays correct objective value def test_objective_value_contract(): solution = playground_run(...) displayed_cost = playground_display_cost(solution) actual_cost = solution.objective_value assert displayed_cost == actual_cost
4. Parameter validation: Invalid inputs rejected with clear messages def test_parameter_validation(): with pytest.raises(ValueError, match="num_vehicles must be positive"): playground_run(num_vehicles=-1)

Step 6: Update Documentation

Keep documentation synchronized with playground features:

Files to update:

1. playground/README.md - User-facing usage guide

- Installation instructions - How to launch playground - Quick start guide - Feature overview

1. playground/FEATURES.md - Feature tracking

- Current features (with status: ✅ Stable, 🚧 Beta, 🔮 Planned) - Recent additions - Known limitations - Roadmap

1. playground/ARCHITECTURE.md - Technical documentation

- File structure (app.py, pages/, components/, utils/) - Component responsibilities - State management (session_state usage) - Extension guide (how to add new features)

1. CHANGELOG_LEARNINGS.md (if bugs fixed)

- Root cause analysis - Fix description - Impact on playground features - New contract tests added

Component Structure

Organize playground code for maintainability:

playground/
├── app.py                    # Main entry point (home page)
├── pages/                    # Multi-page app sections
│   ├── 1_Problem_Definition.py
│   ├── 2_Algorithm_Config.py
│   └── 3_Experiments.py
├── components/               # Reusable UI components
│   ├── instance_viewer.py   # Display instance details
│   ├── route_visualizer.py  # Plot routes on map
│   ├── convergence_plot.py  # Show cost over iterations
│   └── metrics_dashboard.py # Display KPIs
├── utils/                    # Helper functions
│   ├── state_manager.py     # Session state management
│   ├── export_utils.py      # Save/load experiments
│   └── validation.py        # Input validation
├── README.md                 # Usage guide
├── FEATURES.md               # Feature tracking
├── ARCHITECTURE.md           # Technical docs
└── requirements.txt          # Streamlit + dependencies

Development Stages

Stage 1: MVP (Minimal Viable Playground)

Timeline: 1-2 evenings Goal: Get something playable

Features:

• Single-page app with basic workflow
• Instance selection (upload CSV or generate synthetic)
• Algorithm config (5-10 key parameters)
• Run button → display results
• Route visualization + cost metric

Deliverables:

• playground/app.py (~200 lines)
• playground/README.md (installation + quick start)
• 1-2 contract tests (reproducibility, feasibility)

Stage 2: Explainability & Quality

Timeline: 2-3 evenings Goal: Make learning actionable

Features:

• Multi-page app (Problem | Algorithm | Experiments)
• Seed control for reproducibility
• Convergence plot (cost vs. iteration)
• Experiment saving (runs/ directory)
• Contract test suite (5+ tests)

Deliverables:

• playground/pages/ (3 pages)
• contracts/ (5+ tests)
• runs/ directory structure
• playground/FEATURES.md

Stage 3: Gamified Learning

Timeline: Future iterations Goal: Self-driven learning

Features:

• Learning missions ("Get feasible solution in 30s")
• Step-by-step operator visualization
• Parameter impact hints
• Achievement tracking

Common Patterns

Pattern 1: Parameter Configuration UI

import streamlit as st

def render_algorithm_config():
    """Render ALNS parameter configuration UI."""
    st.subheader("ALNS Configuration")

    # Core parameters (always visible)
    max_iterations = st.slider("Max Iterations", 100, 10000, 1000, step=100)
    start_temp = st.number_input("Start Temperature", 0.1, 100.0, 10.0)

    # Advanced parameters (in expander)
    with st.expander("Advanced Parameters"):
        cooling_rate = st.slider("Cooling Rate", 0.90, 0.99, 0.95)
        segment_length = st.number_input("Segment Length", 10, 200, 100)

    # Create config object
    from vrp_toolkit.algorithms.alns import ALNSConfig
    config = ALNSConfig(
        max_iterations=max_iterations,
        start_temp=start_temp,
        cooling_rate=cooling_rate,
        segment_length=segment_length
    )

    return config

Pattern 2: Experiment Saving/Loading

import json
from datetime import datetime
from pathlib import Path

def save_experiment(config, solution, metrics):
    """Save experiment to runs/ directory."""
    timestamp = datetime.now().strftime("%Y-%m-%d_%H-%M-%S")
    run_dir = Path(f"runs/{timestamp}")
    run_dir.mkdir(parents=True, exist_ok=True)

    # Save config
    with open(run_dir / "config.json", "w") as f:
        json.dump(config, f, indent=2)

    # Save solution
    with open(run_dir / "solution.json", "w") as f:
        json.dump(solution.to_dict(), f, indent=2)

    # Save metrics
    with open(run_dir / "metrics.json", "w") as f:
        json.dump(metrics, f, indent=2)

    st.success(f"Experiment saved to {run_dir}")
    return run_dir

Pattern 3: Error Handling

def run_algorithm_with_feedback():
    """Run algorithm with user-friendly error handling."""
    try:
        solution = solver.solve(instance)
        st.success("✅ Algorithm completed successfully")
        return solution
    except ValueError as e:
        st.error(f"❌ Invalid input: {e}")
        st.info("💡 Hint: Check that all parameters are positive")
        return None
    except Exception as e:
        st.error(f"❌ Unexpected error: {e}")
        st.warning("🐛 This might be a bug. Please report it.")
        return None

Quality Checklist

Before marking a playground feature as "complete":

• Functionality: Feature works as designed
• UI/UX: Interface is intuitive (5-second rule: can user figure it out in 5s?)
• Integration: Correctly calls vrp-toolkit APIs
• Visualization: Results are clearly visible
• Error handling: Invalid inputs show helpful messages
• Contract tests: At least 1 test verifies feature behavior
• Documentation: README.md and FEATURES.md updated
• Reproducibility: Same inputs → same outputs (when using fixed seed)

Integration with Other Skills

Works with:

• maintain-architecture-map: Reference ARCHITECTURE_MAP.md to understand module structure
• maintain-data-structures: Reference data structure docs when integrating APIs
• create-tutorial: Playground features can inspire tutorial topics
• track-learnings: When bugs found, use track-learnings to document fixes

Maintains:

• playground/README.md
• playground/FEATURES.md
• playground/ARCHITECTURE.md
• contracts/ tests

References

• references/streamlit_guide.md - Streamlit basics and best practices
• references/ui_components.md - Common UI component patterns
• references/integration_patterns.md - How to integrate vrp-toolkit modules
• playground/VISION.md - Design philosophy and principles

Remember: The goal is learning through interaction, not building a production app. Prioritize clarity and educational value over performance optimization.