ADD: Placement

src/app.py

@@ -7,6 +7,7 @@ from dotenv import load_dotenv
 from core.model_loader import ModelManager
 from core.predictor import PredictionHandler
 from core.model_trainer import ModelTrainer
+from core.placement import build_prediction_grid, place_gruas, optimize_gruas_placement
 from config.models_config import ModelConfig
 
 app = Flask(__name__)
@@ -155,6 +156,247 @@ def list_models():
         "total": len(models_info)
     })
 
+@app.route("/recommend", methods=["GET"])
+def recommend_gruas():
+    """
+    Recomendar ubicaciones para grúas.
+    
+    Parámetros:
+    - week: semana del año (requerido)
+    - dow: día de la semana (0-6, requerido)
+    - hour: hora del día (0-23, requerido)
+    - k: número de grúas (opcional, default=2)
+    - optimize: si es True, encuentra k óptimo (opcional, default=False)
+    - min_gruas: mínimo de grúas para optimización (opcional, default=1)
+    - max_gruas: máximo de grúas para optimización (opcional, default=10)
+    - target_coverage: cobertura objetivo % (opcional, default=80.0)
+    """
+    try:
+        # Obtener parámetros
+        week = int(request.args["week"])
+        dow = int(request.args["dow"])
+        hour = int(request.args["hour"])
+        
+        k = request.args.get("k", default=2, type=int)
+        optimize = request.args.get("optimize", default="false").lower() == "true"
+        min_gruas = request.args.get("min_gruas", default=1, type=int)
+        max_gruas = request.args.get("max_gruas", default=10, type=int)
+        target_coverage = request.args.get("target_coverage", default=80.0, type=float)
+        
+        # Validar parámetros
+        if not (0 <= dow <= 6):
+            return jsonify({"error": "dow must be between 0 and 6"}), 400
+        if not (0 <= hour <= 23):
+            return jsonify({"error": "hour must be between 0 and 23"}), 400
+        if k < 1:
+            return jsonify({"error": "k must be at least 1"}), 400
+        
+        # Obtener modelos
+        demand_model = model_manager.get_model("demand")
+        nulo_model = model_manager.get_model("nulo")
+        
+        if demand_model is None or nulo_model is None:
+            return jsonify({
+                "error": "Required models not loaded. Please train demand and nulo models first."
+            }), 500
+        
+        if DEBUG_MODE:
+            print(f"DEBUG: Building prediction grid for week={week}, dow={dow}, hour={hour}")
+        
+        # Construir grilla de predicciones
+        df = build_prediction_grid(week, dow, hour, demand_model, nulo_model)
+        
+        if df.empty:
+            return jsonify({
+                "error": "No prediction data available for the specified parameters"
+            }), 404
+        
+        if DEBUG_MODE:
+            print(f"DEBUG: Prediction grid built with {len(df)} cells")
+            print(f"DEBUG: Data sample:\n{df.head()}")
+        
+        # Ejecutar algoritmo de colocación
+        if optimize:
+            if DEBUG_MODE:
+                print(f"DEBUG: Running optimization with min={min_gruas}, max={max_gruas}, target={target_coverage}%")
+            
+            result = optimize_gruas_placement(
+                df, 
+                min_gruas=min_gruas, 
+                max_gruas=max_gruas, 
+                target_coverage=target_coverage
+            )
+            
+            if "error" in result:
+                return jsonify(result), 500
+            
+            response = {
+                "week": week,
+                "dow": dow,
+                "hour": hour,
+                "optimization": {
+                    "optimal_k": result['optimal_k'],
+                    "coverage_percentage": result['coverage_percentage'],
+                    "risk_coverage_percentage": result['risk_coverage_percentage'],
+                    "target_coverage": target_coverage
+                },
+                "recommended_h3": result['selected_cells'],
+                "statistics": result.get('placement', {}).get('statistics', []),
+                "coverage": result.get('placement', {}).get('coverage', {})
+            }
+            
+            # Incluir todos los resultados si está en modo debug
+            if DEBUG_MODE:
+                response['debug'] = {
+                    'all_results': result.get('all_results', [])
+                }
+                
+        else:
+            if DEBUG_MODE:
+                print(f"DEBUG: Running placement with k={k}")
+            
+            result = place_gruas(df, k=k)
+            
+            response = {
+                "week": week,
+                "dow": dow,
+                "hour": hour,
+                "parameters": {
+                    "k": k,
+                    "optimization": optimize
+                },
+                "recommended_h3": result['selected'],
+                "statistics": result.get('statistics', []),
+                "coverage": result.get('coverage', {})
+            }
+        
+        return jsonify(response)
+        
+    except KeyError as e:
+        return jsonify({"error": f"Missing required parameter: {str(e)}"}), 400
+    except ValueError as e:
+        return jsonify({"error": f"Invalid parameter value: {str(e)}"}), 400
+    except Exception as e:
+        if DEBUG_MODE:
+            import traceback
+            print(f"ERROR in recommend_gruas: {str(e)}")
+            traceback.print_exc()
+        
+        return jsonify({
+            "error": f"Internal server error: {str(e)}"
+        }), 500
+
+@app.route("/coverage/radius", methods=["GET"])
+def get_coverage_radius():
+    """
+    Calcular radio de cobertura para una demanda específica.
+    
+    Parámetros:
+    - demand: demanda esperada (requerido)
+    """
+    try:
+        demand = float(request.args["demand"])
+        
+        from algorithms.placement import coverage_radius, get_coverage_weight
+        
+        radius = coverage_radius(demand)
+        weight = get_coverage_weight(demand)
+        
+        # Determinar nivel de demanda
+        if demand < 2:
+            level = "baja"
+        elif demand < 5:
+            level = "media"
+        else:
+            level = "alta"
+        
+        return jsonify({
+            "demand": demand,
+            "demand_level": level,
+            "coverage_radius": radius,
+            "coverage_weight": weight,
+            "explanation": f"Para demanda {level} ({demand:.2f}), el radio de cobertura es {radius}"
+        })
+        
+    except (KeyError, ValueError) as e:
+        return jsonify({"error": f"Invalid parameter: {str(e)}"}), 400
+    except Exception as e:
+        return jsonify({"error": f"Internal error: {str(e)}"}), 500
+
+@app.route("/coverage/cells", methods=["GET"])
+def get_coverage_cells():
+    """
+    Obtener celdas cubiertas por una celda H3 con un radio específico.
+    
+    Parámetros:
+    - h3: celda H3 (requerido)
+    - radius: radio de cobertura (opcional, default calculado según demanda)
+    - demand: demanda para calcular radio automático (opcional)
+    """
+    try:
+        h3_cell = request.args["h3"]
+        
+        from algorithms.placement import covered_cells, coverage_radius
+        
+        # Determinar radio
+        if "radius" in request.args:
+            radius = int(request.args["radius"])
+        elif "demand" in request.args:
+            demand = float(request.args["demand"])
+            radius = coverage_radius(demand)
+        else:
+            return jsonify({
+                "error": "Either radius or demand parameter is required"
+            }), 400
+        
+        # Calcular celdas cubiertas
+        cells = covered_cells(h3_cell, radius)
+        
+        return jsonify({
+            "h3": h3_cell,
+            "radius": radius,
+            "cells_covered": len(cells),
+            "covered_cells": list(cells)[:100],  # Limitar para no sobrecargar respuesta
+            "truncated": len(cells) > 100
+        })
+        
+    except (KeyError, ValueError) as e:
+        return jsonify({"error": f"Invalid parameter: {str(e)}"}), 400
+    except Exception as e:
+        return jsonify({"error": f"Internal error: {str(e)}"}), 500
+
+@app.route("/risk/calculate", methods=["GET"])
+def calculate_risk_endpoint():
+    """
+    Calcular riesgo para una combinación de demanda y probabilidad de nulo.
+    
+    Parámetros:
+    - demand: demanda esperada (requerido)
+    - nulo_prob: probabilidad de nulo (requerido, 0-1)
+    """
+    try:
+        demand = float(request.args["demand"])
+        nulo_prob = float(request.args["nulo_prob"])
+        
+        from algorithms.placement import calculate_risk
+        
+        if not 0 <= nulo_prob <= 1:
+            return jsonify({"error": "nulo_prob must be between 0 and 1"}), 400
+        
+        risk = calculate_risk(demand, nulo_prob)
+        
+        return jsonify({
+            "demand": demand,
+            "nulo_probability": nulo_prob,
+            "risk": risk,
+            "explanation": f"Riesgo = demanda × probabilidad_nulo = {demand} × {nulo_prob} = {risk:.4f}"
+        })
+        
+    except (KeyError, ValueError) as e:
+        return jsonify({"error": f"Invalid parameter: {str(e)}"}), 400
+    except Exception as e:
+        return jsonify({"error": f"Internal error: {str(e)}"}), 500
+
 @app.route("/health", methods=["GET"])
 def health():
     """Endpoint de salud"""
@@ -253,6 +495,30 @@ def index():
             "description": "Entrenar modelos",
             "parameters": "model_type (opcional)"
         },
+        {
+            "path": "/recommend",
+            "method": "GET",
+            "description": "Recomendar ubicaciones para grúas",
+            "parameters": "week, dow, hour (requeridos), k, optimize, min_gruas, max_gruas, target_coverage (opcionales)"
+        },
+        {
+            "path": "/coverage/radius",
+            "method": "GET",
+            "description": "Calcular radio de cobertura según demanda",
+            "parameters": "demand (requerido)"
+        },
+        {
+            "path": "/coverage/cells",
+            "method": "GET", 
+            "description": "Obtener celdas cubiertas por una celda H3",
+            "parameters": "h3 (requerido), radius o demand (opcional)"
+        },
+        {
+            "path": "/risk/calculate",
+            "method": "GET",
+            "description": "Calcular riesgo (demanda × probabilidad nulo)",
+            "parameters": "demand, nulo_prob (requeridos)"
+        },
         {
             "path": "/models",
             "method": "GET",

src/core/placement.py

@@ -0,0 +1,267 @@
+import h3
+import pandas as pd
+from typing import List, Set, Dict, Tuple
+import numpy as np
+
+def coverage_radius(expected_demand: float) -> int:
+    """
+    Determina el radio de cobertura según la demanda esperada, como medimos anteriormente
+    
+    Nivel demanda  | Radio H3
+    ---------------|----------
+    baja (<2)      | 18
+    media (2-5)    | 12
+    alta (≥5)      | 6
+    """
+    if expected_demand < 2:
+        return 18
+    elif expected_demand < 5:
+        return 12
+    else:
+        return 6
+
+def get_coverage_weight(expected_demand: float) -> float:
+    """
+    Calcula un peso para la cobertura basado en la demanda.
+    Demanda más alta = mayor peso.
+    """
+    if expected_demand < 2:
+        return 1.0
+    elif expected_demand < 5:
+        return 2.0
+    else:
+        return 3.0
+
+def covered_cells(h3_cell: str, radius: int) -> Set[str]:
+    """Obtiene todas las celdas H3 cubiertas por un radio dado."""
+    try:
+        return set(h3.k_ring(h3_cell, radius))
+    except Exception as e:
+        print(f"Error calculando cobertura para celda {h3_cell}: {e}")
+        return {h3_cell}
+
+def calculate_risk(expected_demand: float, nulo_probability: float) -> float:
+    """
+    Calcula el riesgo como producto de demanda y probabilidad de nulo.
+    Esto prioriza áreas con alta demanda y alta probabilidad de nulos.
+    """
+    return expected_demand * nulo_probability
+
+def calculate_coverage_score(cell: str, radius: int, df: pd.DataFrame) -> float:
+    """
+    Calcula el puntaje de cobertura para una celda específica.
+    Considera tanto el riesgo como el peso de la demanda.
+    """
+    covered = covered_cells(cell, radius)
+    covered_df = df[df['h3'].isin(covered)]
+    
+    if covered_df.empty:
+        return 0
+    
+    # Suma ponderada del riesgo
+    total_risk = covered_df['risk'].sum()
+    
+    # Peso adicional por demanda alta
+    demand_weight = covered_df['coverage_weight'].sum()
+    
+    return total_risk * (1 + 0.1 * demand_weight)
+
+def build_prediction_grid(week: int, dow: int, hour: int, 
+                         demand_model, nulo_model) -> pd.DataFrame:
+    """
+    Construye una cuadrícula de predicciones para todas las celdas H3.
+    """
+    # Obtener todas las celdas H3 únicas de la base de datos
+    from db import get_all_h3_cells
+    h3_cells = get_all_h3_cells()
+    
+    predictions = []
+    
+    for h3_cell in h3_cells:
+        try:
+            # Convertir H3 a entero para el modelo
+            h3_int = int(h3_cell, 16)
+            
+            # Predecir demanda
+            X_demand = [[h3_int, week, dow, hour]]
+            expected_demand = float(demand_model.predict(X_demand)[0])
+            
+            # Predecir probabilidad de nulo
+            X_nulo = [[h3_int, week, dow, hour]]
+            nulo_prob = float(nulo_model.predict_proba(X_nulo)[0][1])
+            
+            # Calcular riesgo
+            risk = calculate_risk(expected_demand, nulo_prob)
+            
+            # Determinar radio de cobertura y peso
+            radius = coverage_radius(expected_demand)
+            coverage_weight = get_coverage_weight(expected_demand)
+            
+            predictions.append({
+                'h3': h3_cell,
+                'h3_int': h3_int,
+                'week': week,
+                'dow': dow,
+                'hour': hour,
+                'expected_demand': expected_demand,
+                'nulo_probability': nulo_prob,
+                'risk': risk,
+                'coverage_radius': radius,
+                'coverage_weight': coverage_weight
+            })
+            
+        except Exception as e:
+            print(f"Error procesando celda {h3_cell}: {e}")
+            continue
+    
+    return pd.DataFrame(predictions)
+
+def place_gruas(df: pd.DataFrame, k: int = 2, 
+                max_iterations: int = 100) -> Dict:
+    """
+    Algoritmo de colocación de grúas greedy.
+    
+    Args:
+        df: DataFrame con predicciones por celda H3
+        k: número de grúas a colocar
+        max_iterations: máximo de iteraciones para evitar loops infinitos
+    
+    Returns:
+        Diccionario con las grúas seleccionadas y estadísticas
+    """
+    if df.empty:
+        return {"selected": [], "coverage": 0, "message": "No data available"}
+    
+    # Asegurar que tenemos columnas necesarias
+    required_cols = ['h3', 'expected_demand', 'risk', 'coverage_radius']
+    for col in required_cols:
+        if col not in df.columns:
+            raise ValueError(f"DataFrame missing required column: {col}")
+    
+    uncovered = df.copy()
+    selected_cells = []
+    coverage_stats = []
+    total_initial_risk = df['risk'].sum()
+    
+    iteration = 0
+    while len(selected_cells) < k and iteration < max_iterations and not uncovered.empty:
+        iteration += 1
+        best_cell = None
+        best_score = -1
+        best_coverage = set()
+        
+        # Evaluar cada celda no cubierta como candidata
+        for _, row in uncovered.iterrows():
+            current_cell = row['h3']
+            radius = row['coverage_radius']
+            
+            # Calcular celdas cubiertas
+            covered = covered_cells(current_cell, radius)
+            
+            # Calcular puntaje de cobertura
+            score = calculate_coverage_score(current_cell, radius, uncovered)
+            
+            if score > best_score:
+                best_score = score
+                best_cell = current_cell
+                best_coverage = covered
+        
+        if best_cell is None:
+            break
+        
+        # Añadir a seleccionadas
+        selected_cells.append(best_cell)
+        
+        # Calcular estadísticas para esta grúa
+        grua_stats = {
+            'h3': best_cell,
+            'expected_demand': float(uncovered.loc[uncovered['h3'] == best_cell, 'expected_demand'].iloc[0]),
+            'coverage_radius': int(uncovered.loc[uncovered['h3'] == best_cell, 'coverage_radius'].iloc[0]),
+            'nulo_probability': float(uncovered.loc[uncovered['h3'] == best_cell, 'nulo_probability'].iloc[0]),
+            'cells_covered': len(best_coverage),
+            'risk_covered': float(uncovered[uncovered['h3'].isin(best_coverage)]['risk'].sum())
+        }
+        coverage_stats.append(grua_stats)
+        
+        # Eliminar celdas cubiertas
+        uncovered = uncovered[~uncovered['h3'].isin(best_coverage)]
+    
+    # Calcular cobertura total
+    covered_by_all = set()
+    for cell in selected_cells:
+        cell_data = df[df['h3'] == cell].iloc[0]
+        covered = covered_cells(cell, cell_data['coverage_radius'])
+        covered_by_all.update(covered)
+    
+    total_cells_covered = len(covered_by_all)
+    total_cells = len(df)
+    coverage_percentage = (total_cells_covered / total_cells * 100) if total_cells > 0 else 0
+    
+    total_risk_covered = df[df['h3'].isin(covered_by_all)]['risk'].sum()
+    risk_coverage_percentage = (total_risk_covered / total_initial_risk * 100) if total_initial_risk > 0 else 0
+    
+    return {
+        'selected': selected_cells,
+        'statistics': coverage_stats,
+        'coverage': {
+            'cells_covered': total_cells_covered,
+            'total_cells': total_cells,
+            'coverage_percentage': round(coverage_percentage, 2),
+            'risk_covered': round(float(total_risk_covered), 4),
+            'total_risk': round(float(total_initial_risk), 4),
+            'risk_coverage_percentage': round(risk_coverage_percentage, 2)
+        },
+        'parameters': {
+            'k': k,
+            'iterations': iteration
+        }
+    }
+
+def optimize_gruas_placement(df: pd.DataFrame, min_gruas: int = 1, 
+                           max_gruas: int = 10, 
+                           target_coverage: float = 80.0) -> Dict:
+    """
+    Encuentra el número óptimo de grúas para alcanzar una cobertura objetivo.
+    
+    Args:
+        df: DataFrame con predicciones
+        min_gruas: mínimo número de grúas
+        max_gruas: máximo número de grúas
+        target_coverage: porcentaje de cobertura objetivo
+    
+    Returns:
+        Resultado con número óptimo de grúas
+    """
+    results = []
+    
+    for k in range(min_gruas, max_gruas + 1):
+        placement = place_gruas(df, k=k)
+        
+        results.append({
+            'k': k,
+            'coverage_percentage': placement['coverage']['coverage_percentage'],
+            'risk_coverage_percentage': placement['coverage']['risk_coverage_percentage'],
+            'selected_cells': placement['selected'],
+            'placement': placement
+        })
+        
+        # Si alcanzamos la cobertura objetivo, podemos detenernos
+        if placement['coverage']['coverage_percentage'] >= target_coverage:
+            break
+    
+    # Encontrar el mejor balance
+    if not results:
+        return {"error": "No results generated"}
+    
+    # Ordenar por cobertura de riesgo (prioridad) y luego por número de grúas (menos es mejor)
+    results.sort(key=lambda x: (-x['risk_coverage_percentage'], x['k']))
+    
+    best = results[0]
+    
+    return {
+        'optimal_k': best['k'],
+        'coverage_percentage': best['coverage_percentage'],
+        'risk_coverage_percentage': best['risk_coverage_percentage'],
+        'selected_cells': best['selected_cells'],
+        'all_results': results
+    }

src/db.py

@@ -47,4 +47,61 @@ def fetch_data_legacy():
         return df
     except SQLAlchemyError as e:
         print(f"Error al ejecutar la consulta: {e}")
-        return pd.DataFrame()
+        return pd.DataFrame()
+
+
+def get_all_h3_cells() -> list:
+    """Obtener todas las celdas H3 únicas de la base de datos"""
+    query = """
+        SELECT DISTINCT h3 
+        FROM demanda_h3_hour_ml
+    """
+    
+    try:
+        with engine.connect() as conn:
+            df = pd.read_sql(text(query), conn)
+        return df['h3'].tolist() if not df.empty else []
+    except SQLAlchemyError as e:
+        print(f"Error obteniendo celdas H3: {e}")
+        return []
+
+def get_h3_cells_by_area(min_lat: float = None, max_lat: float = None,
+                        min_lon: float = None, max_lon: float = None) -> list:
+    """
+    Obtener celdas H3 dentro de un área geográfica específica.
+    
+    Nota: Esta función requiere que la tabla tenga columnas lat/lon
+    o que se pueda calcular H3 a partir de coordenadas.
+    """
+    # Implementación básica - ajustar según tu esquema de datos
+    query = """
+        SELECT DISTINCT h3 
+        FROM demanda_h3_hour_ml
+    """
+    
+    conditions = []
+    params = {}
+    
+    if min_lat:
+        conditions.append("latitude >= :min_lat")
+        params['min_lat'] = min_lat
+    if max_lat:
+        conditions.append("latitude <= :max_lat")
+        params['max_lat'] = max_lat
+    if min_lon:
+        conditions.append("longitude >= :min_lon")
+        params['min_lon'] = min_lon
+    if max_lon:
+        conditions.append("longitude <= :max_lon")
+        params['max_lon'] = max_lon
+    
+    if conditions:
+        query += " AND " + " AND ".join(conditions)
+    
+    try:
+        with engine.connect() as conn:
+            df = pd.read_sql(text(query), conn, params=params)
+        return df['h3'].tolist() if not df.empty else []
+    except SQLAlchemyError as e:
+        print(f"Error obteniendo celdas H3 por área: {e}")
+        return []

src/pyproject.toml

@@ -21,7 +21,8 @@ dependencies = [
    'xgboost',
    'scikit-learn',
    'joblib',
-   'python-dotenv'
+   'python-dotenv',
+   'h3'
 ]
 
 [tool.setuptools]

src/requirements.txt

@@ -5,4 +5,5 @@ scikit-learn
 joblib
 SQLAlchemy
 psycopg2-binary
-python-dotenv
+python-dotenv
+h3