{
"cells": [
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# Integración CoreML de RusTorch - Enlaces Python\n",
"\n",
"Este notebook demuestra cómo usar la funcionalidad CoreML de RusTorch a través de enlaces Python."
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Configuración e Importaciones"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# Importar enlaces Python de RusTorch\n",
"try:\n",
" import rustorch\n",
" print(f\"✅ Versión de RusTorch: {rustorch.__version__}\")\n",
" print(f\"📝 Descripción: {rustorch.__description__}\")\n",
" print(f\"👥 Autor: {rustorch.__author__}\")\n",
"except ImportError as e:\n",
" print(f\"❌ Error al importar RusTorch: {e}\")\n",
" print(\"Por favor, construya con maturin develop\")\n",
" exit(1)\n",
"\n",
"import numpy as np\n",
"import platform\n",
"\n",
"print(f\"🖥️ Plataforma: {platform.system()} {platform.release()}\")\n",
"print(f\"🐍 Versión de Python: {platform.python_version()}\")"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Verificar Disponibilidad de CoreML"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# Verificar funcionalidad CoreML\n",
"try:\n",
" # Verificar si CoreML está disponible\n",
" coreml_available = rustorch.is_coreml_available()\n",
" print(f\"🍎 CoreML disponible: {coreml_available}\")\n",
" \n",
" if coreml_available:\n",
" print(\"🎉 ¡CoreML está disponible!\")\n",
" \n",
" # Obtener información del dispositivo\n",
" device_info = rustorch.get_coreml_device_info()\n",
" print(\"📱 Información del dispositivo CoreML:\")\n",
" print(device_info)\n",
" else:\n",
" print(\"⚠️ CoreML no está disponible\")\n",
" if platform.system() != \"Darwin\":\n",
" print(\"CoreML solo está disponible en macOS\")\n",
" else:\n",
" print(\"Es posible que las características de CoreML no estén habilitadas\")\n",
" \n",
"except AttributeError:\n",
" print(\"❌ Funciones CoreML no encontradas\")\n",
" print(\"Puede que no esté construido con características CoreML\")\n",
" coreml_available = False\n",
"except Exception as e:\n",
" print(f\"❌ Error al verificar CoreML: {e}\")\n",
" coreml_available = False"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Creación de Dispositivo CoreML y Operaciones"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"if coreml_available:\n",
" try:\n",
" # Crear dispositivo CoreML\n",
" device = rustorch.CoreMLDevice(device_id=0)\n",
" print(f\"🖥️ Dispositivo CoreML creado: {device}\")\n",
" \n",
" # Obtener información del dispositivo\n",
" print(f\"🆔 ID del dispositivo: {device.device_id()}\")\n",
" print(f\"✅ Disponible: {device.is_available()}\")\n",
" print(f\"💾 Límite de memoria: {device.memory_limit()} bytes\")\n",
" print(f\"🧮 Límite de unidades de cómputo: {device.compute_units_limit()}\")\n",
" print(f\"📚 Tamaño de caché del modelo: {device.model_cache_size()}\")\n",
" \n",
" # Limpieza de caché\n",
" device.cleanup_cache()\n",
" print(\"🧹 Caché limpiado\")\n",
" \n",
" except Exception as e:\n",
" print(f\"❌ Error en operación del dispositivo CoreML: {e}\")\n",
"else:\n",
" print(\"⚠️ Omitiendo operaciones del dispositivo ya que CoreML no está disponible\")"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Configuración del Backend CoreML"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"if coreml_available:\n",
" try:\n",
" # Crear configuración del backend CoreML\n",
" config = rustorch.CoreMLBackendConfig(\n",
" enable_caching=True,\n",
" max_cache_size=200,\n",
" enable_profiling=True,\n",
" auto_fallback=True\n",
" )\n",
" print(f\"⚙️ Configuración del backend: {config}\")\n",
" \n",
" # Verificar y modificar valores de configuración\n",
" print(f\"📊 Habilitar caché: {config.enable_caching}\")\n",
" print(f\"🗂️ Tamaño máximo de caché: {config.max_cache_size}\")\n",
" print(f\"📈 Habilitar perfilado: {config.enable_profiling}\")\n",
" print(f\"🔄 Respaldo automático: {config.auto_fallback}\")\n",
" \n",
" # Modificar configuración\n",
" config.enable_profiling = False\n",
" config.max_cache_size = 150\n",
" print(f\"\\n🔧 Configuración actualizada: {config}\")\n",
" \n",
" # Crear backend CoreML\n",
" backend = rustorch.CoreMLBackend(config)\n",
" print(f\"🚀 Backend CoreML: {backend}\")\n",
" print(f\"✅ Backend disponible: {backend.is_available()}\")\n",
" \n",
" # Obtener estadísticas del backend\n",
" stats = backend.get_stats()\n",
" print(f\"📊 Estadísticas del backend: {stats}\")\n",
" print(f\" Operaciones totales: {stats.total_operations}\")\n",
" print(f\" Aciertos de caché: {stats.cache_hits}\")\n",
" print(f\" Fallos de caché: {stats.cache_misses}\")\n",
" print(f\" Operaciones de respaldo: {stats.fallback_operations}\")\n",
" print(f\" Tasa de aciertos de caché: {stats.cache_hit_rate():.2%}\")\n",
" print(f\" Tasa de respaldo: {stats.fallback_rate():.2%}\")\n",
" print(f\" Tiempo promedio de ejecución: {stats.average_execution_time_ms:.2f}ms\")\n",
" \n",
" # Limpieza de caché\n",
" backend.cleanup_cache()\n",
" print(\"\\n🧹 Caché del backend limpiado\")\n",
" \n",
" except Exception as e:\n",
" print(f\"❌ Error en operación del backend CoreML: {e}\")\n",
"else:\n",
" print(\"⚠️ Omitiendo operaciones del backend ya que CoreML no está disponible\")"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Operaciones Básicas de Tensores (CPU)\n",
"\n",
"Para comparar con CoreML, primero realicemos operaciones básicas en CPU."
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"try:\n",
" # Creación y operaciones básicas de tensores\n",
" print(\"🧮 Operaciones básicas de tensores (CPU)\")\n",
" \n",
" # Crear tensores desde arrays NumPy (interfaz simplificada)\n",
" data_a = np.random.randn(2, 3).astype(np.float32)\n",
" data_b = np.random.randn(3, 2).astype(np.float32)\n",
" \n",
" print(f\"📐 Forma de la Matriz A: {data_a.shape}\")\n",
" print(f\"📐 Forma de la Matriz B: {data_b.shape}\")\n",
" \n",
" # Multiplicación de matrices con NumPy (para comparación)\n",
" numpy_result = np.matmul(data_a, data_b)\n",
" print(f\"✅ Forma del resultado matmul NumPy: {numpy_result.shape}\")\n",
" print(f\"📊 Resultado (primeros elementos): {numpy_result.flatten()[:4]}\")\n",
" \n",
" print(\"\\n🚀 Operaciones CPU completadas\")\n",
" \n",
"except Exception as e:\n",
" print(f\"❌ Error en operación de tensor: {e}\")"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Simulación de Comparación de Rendimiento"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"import time\n",
"\n",
"def benchmark_matrix_operations():\n",
" \"\"\"Comparar rendimiento con diferentes tamaños de matriz\"\"\"\n",
" \n",
" sizes = [(64, 64), (128, 128), (256, 256), (512, 512)]\n",
" \n",
" print(\"🏁 Comparación de rendimiento:\")\n",
" print(\"Tamaño\\t\\tTiempo CPU (ms)\\tCoreML Esperado (ms)\")\n",
" print(\"-\" * 58)\n",
" \n",
" for size in sizes:\n",
" # Medir tiempo de ejecución CPU\n",
" a = np.random.randn(*size).astype(np.float32)\n",
" b = np.random.randn(size[1], size[0]).astype(np.float32)\n",
" \n",
" start_time = time.time()\n",
" result = np.matmul(a, b)\n",
" cpu_time = (time.time() - start_time) * 1000\n",
" \n",
" # Tiempo CoreML esperado (hipotético)\n",
" # En implementación real, usar mediciones reales del backend CoreML\n",
" expected_coreml_time = cpu_time * 0.6 # Suposición: CoreML es 40% más rápido\n",
" \n",
" print(f\"{size[0]}x{size[1]}\\t\\t{cpu_time:.2f}\\t\\t{expected_coreml_time:.2f}\")\n",
"\n",
"benchmark_matrix_operations()\n",
"\n",
"print(\"\\n📝 Nota: Los tiempos de CoreML son hipotéticos. Los valores reales dependen de la implementación específica.\")"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Simulación de Selección de Dispositivo"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"def simulate_device_selection():\n",
" \"\"\"Simular selección inteligente de dispositivo\"\"\"\n",
" \n",
" operations = [\n",
" (\"Multiplicación de matriz pequeña\", (16, 16), \"CPU\"),\n",
" (\"Multiplicación de matriz mediana\", (128, 128), \"Metal GPU\"),\n",
" (\"Multiplicación de matriz grande\", (512, 512), \"CoreML\" if coreml_available else \"Metal GPU\"),\n",
" (\"Función de activación\", (32, 64, 128, 128), \"Metal GPU\"),\n",
" (\"Convolución pequeña\", (1, 3, 32, 32), \"CPU\"),\n",
" (\"Convolución grande\", (16, 64, 224, 224), \"CoreML\" if coreml_available else \"Metal GPU\"),\n",
" (\"Operaciones de números complejos\", (128, 128), \"Metal GPU\"), # CoreML no soportado\n",
" (\"Distribución estadística\", (1000,), \"CPU\"), # CoreML no soportado\n",
" ]\n",
" \n",
" print(\"🎯 Simulación de selección inteligente de dispositivo:\")\n",
" print(\"Operación\\t\\t\\tForma del Tensor\\t\\tDispositivo Seleccionado\")\n",
" print(\"-\" * 78)\n",
" \n",
" for name, shape, device in operations:\n",
" shape_str = \"x\".join(map(str, shape))\n",
" print(f\"{name:<31}\\t{shape_str:<15}\\t{device}\")\n",
" \n",
" print(\"\\n📝 Lógica de selección:\")\n",
" print(\" • Operaciones pequeñas: CPU (evitar sobrecarga)\")\n",
" print(\" • Operaciones medianas: Metal GPU (equilibrado)\")\n",
" print(\" • Operaciones grandes: CoreML (optimizado)\")\n",
" print(\" • Operaciones no soportadas: respaldo GPU/CPU\")\n",
"\n",
"simulate_device_selection()"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Ejemplo Práctico: Capa Simple de Red Neuronal"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"def simulate_neural_network_layer():\n",
" \"\"\"Simular capa de red neuronal\"\"\"\n",
" \n",
" print(\"🧠 Simulación de capa de red neuronal:\")\n",
" \n",
" # Tamaño del lote y configuración de la capa\n",
" batch_size = 32\n",
" input_features = 784 # 28x28 MNIST\n",
" hidden_features = 256\n",
" output_features = 10 # 10 clases\n",
" \n",
" print(f\"📊 Tamaño del lote: {batch_size}\")\n",
" print(f\"🔢 Características de entrada: {input_features}\")\n",
" print(f\"🧮 Características ocultas: {hidden_features}\")\n",
" print(f\"🎯 Características de salida: {output_features}\")\n",
" \n",
" # Simulación de pase hacia adelante\n",
" steps = [\n",
" (\"Entrada → Oculta\", f\"({batch_size}, {input_features}) @ ({input_features}, {hidden_features})\", \"CoreML\" if coreml_available else \"Metal\"),\n",
" (\"Activación ReLU\", f\"({batch_size}, {hidden_features})\", \"Metal\"),\n",
" (\"Oculta → Salida\", f\"({batch_size}, {hidden_features}) @ ({hidden_features}, {output_features})\", \"CoreML\" if coreml_available else \"Metal\"),\n",
" (\"Softmax\", f\"({batch_size}, {output_features})\", \"CPU\"),\n",
" ]\n",
" \n",
" print(\"\\n🔄 Simulación de pase hacia adelante:\")\n",
" total_time = 0\n",
" \n",
" for step, shape, device in steps:\n",
" # Tiempo de ejecución virtual (ms)\n",
" if device == \"CoreML\":\n",
" time_ms = np.random.uniform(0.5, 2.0)\n",
" elif device == \"Metal\":\n",
" time_ms = np.random.uniform(1.0, 3.0)\n",
" else: # CPU\n",
" time_ms = np.random.uniform(0.2, 1.0)\n",
" \n",
" total_time += time_ms\n",
" print(f\" {step:<15} {shape:<30} {device:<8} {time_ms:.2f}ms\")\n",
" \n",
" print(f\"\\n⏱️ Tiempo total del pase hacia adelante: {total_time:.2f}ms\")\n",
" print(f\"🚀 Rendimiento estimado: {1000/total_time:.0f} lotes/segundo\")\n",
"\n",
"simulate_neural_network_layer()"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Resumen y Próximos Pasos"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"print(\"📋 Resumen de Integración CoreML de RusTorch:\")\n",
"print()\n",
"print(\"✅ Elementos completados:\")\n",
"print(\" • Configuración del entorno Jupyter\")\n",
"print(\" • Creación de kernel Rust y enlaces Python\")\n",
"print(\" • Verificación de disponibilidad CoreML\")\n",
"print(\" • Gestión de dispositivos y configuración\")\n",
"print(\" • Estadísticas y perfilado del backend\")\n",
"print(\" • Selección inteligente de dispositivo\")\n",
"print()\n",
"print(\"🚧 Desarrollo futuro:\")\n",
"print(\" • Implementación real de operaciones CoreML\")\n",
"print(\" • Evaluación comparativa de rendimiento\")\n",
"print(\" • Más funciones de activación y tipos de capas\")\n",
"print(\" • Mejoras en manejo de errores\")\n",
"print(\" • Optimización de memoria\")\n",
"print()\n",
"print(\"🎯 Próximos pasos recomendados:\")\n",
"print(\" 1. Cargar y probar modelos CoreML reales\")\n",
"print(\" 2. Comparar rendimiento Metal y CoreML\")\n",
"print(\" 3. Probar con flujos de trabajo de aprendizaje profundo reales\")\n",
"print(\" 4. Evaluar en entorno de producción\")\n",
"\n",
"if coreml_available:\n",
" print(\"\\n🎉 ¡Felicidades! CoreML está disponible y todas las características pueden ser probadas.\")\n",
"else:\n",
" print(\"\\n⚠️ CoreML no está disponible, pero las características básicas están funcionando.\")\n",
" print(\" Recomendamos construir con características CoreML habilitadas en macOS.\")"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.8.5"
}
},
"nbformat": 4,
"nbformat_minor": 4
}