trading-platform/docs/02-definicion-modulos/OQI-006-ml-signals/especificaciones/ET-ML-001-arquitectura.md

id	title	type	status	priority	epic	project	version	created_date	updated_date
ET-ML-001	Arquitectura ML Engine	Technical Specification	Done	Alta	OQI-006	trading-platform	1.0.0	2025-12-05	2026-01-04

ET-ML-001: Arquitectura ML Engine

Metadata

Campo	Valor
ID	ET-ML-001
Épica	OQI-006 - Señales ML
Tipo	Especificación Técnica
Versión	1.0.0
Estado	Aprobado
Última actualización	2025-12-05

---

Propósito

Definir la arquitectura completa del ML Engine, incluyendo la estructura de servicios, comunicación entre componentes, y flujos de datos para predicciones y señales de trading.

---

Arquitectura General

Diagrama de Alto Nivel

┌─────────────────────────────────────────────────────────────────────────────┐
│                              ML ENGINE                                       │
│  ┌─────────────────────────────────────────────────────────────────────┐   │
│  │                         FastAPI Application                          │   │
│  │  ┌──────────────┐  ┌──────────────┐  ┌──────────────────────────┐  │   │
│  │  │   Routers    │  │  Services    │  │      Background Tasks     │  │   │
│  │  │              │  │              │  │                           │  │   │
│  │  │ /predictions │  │ Predictor    │  │ ┌─────────────────────┐  │  │   │
│  │  │ /signals     │  │ SignalGen    │  │ │ Market Data Fetcher │  │  │   │
│  │  │ /indicators  │  │ Indicators   │  │ └─────────────────────┘  │  │   │
│  │  │ /models      │  │ ModelManager │  │ ┌─────────────────────┐  │  │   │
│  │  │ /health      │  │              │  │ │  Training Pipeline  │  │  │   │
│  │  └──────────────┘  └──────────────┘  │ └─────────────────────┘  │  │   │
│  │                                       │ ┌─────────────────────┐  │  │   │
│  │  ┌─────────────────────────────────┐ │ │   Signal Publisher  │  │  │   │
│  │  │          Models Layer           │ │ └─────────────────────┘  │  │   │
│  │  │  ┌─────────────┐ ┌───────────┐ │ └──────────────────────────┘  │   │
│  │  │  │   XGBoost   │ │  Ensemble │ │                               │   │
│  │  │  │   Models    │ │  Manager  │ │                               │   │
│  │  │  └─────────────┘ └───────────┘ │                               │   │
│  │  └─────────────────────────────────┘                               │   │
│  └─────────────────────────────────────────────────────────────────────┘   │
│                                                                              │
│  ┌─────────────────────────────────────────────────────────────────────┐   │
│  │                          Data Layer                                   │   │
│  │  ┌──────────────┐  ┌──────────────┐  ┌──────────────────────────┐  │   │
│  │  │    Redis     │  │  PostgreSQL  │  │      Model Storage       │  │   │
│  │  │    Cache     │  │   (Metrics)  │  │      (File System)       │  │   │
│  │  └──────────────┘  └──────────────┘  └──────────────────────────┘  │   │
│  └─────────────────────────────────────────────────────────────────────┘   │
└─────────────────────────────────────────────────────────────────────────────┘
                │                    │                    │
                ▼                    ▼                    ▼
        ┌───────────┐        ┌───────────┐        ┌───────────┐
        │  Binance  │        │  Backend  │        │  Frontend │
        │    API    │        │  Express  │        │   React   │
        └───────────┘        └───────────┘        └───────────┘

---

Estructura de Directorios

apps/ml-engine/
├── app/
│   ├── __init__.py
│   ├── main.py                     # FastAPI entry point
│   ├── config/
│   │   ├── __init__.py
│   │   ├── settings.py             # Pydantic settings
│   │   └── logging.py              # Logging configuration
│   ├── api/
│   │   ├── __init__.py
│   │   ├── dependencies.py         # Dependency injection
│   │   └── routers/
│   │       ├── __init__.py
│   │       ├── predictions.py      # /predictions endpoints
│   │       ├── signals.py          # /signals endpoints
│   │       ├── indicators.py       # /indicators endpoints
│   │       ├── models.py           # /models endpoints
│   │       └── health.py           # /health endpoints
│   ├── core/
│   │   ├── __init__.py
│   │   ├── exceptions.py           # Custom exceptions
│   │   ├── security.py             # API key validation
│   │   └── middleware.py           # Custom middleware
│   ├── services/
│   │   ├── __init__.py
│   │   ├── predictor.py            # Prediction service
│   │   ├── signal_generator.py     # Signal generation
│   │   ├── indicator_calculator.py # Technical indicators
│   │   ├── model_manager.py        # Model lifecycle
│   │   └── market_data.py          # Data fetching
│   ├── models/
│   │   ├── __init__.py
│   │   ├── base.py                 # Base model class
│   │   ├── range_predictor.py      # Price range predictor
│   │   ├── tpsl_classifier.py      # TP/SL classifier
│   │   └── signal_classifier.py    # Signal classifier
│   ├── features/
│   │   ├── __init__.py
│   │   ├── builder.py              # Feature builder
│   │   ├── volatility.py           # Volatility features
│   │   ├── momentum.py             # Momentum features
│   │   ├── trend.py                # Trend features
│   │   └── volume.py               # Volume features
│   ├── schemas/
│   │   ├── __init__.py
│   │   ├── prediction.py           # Prediction DTOs
│   │   ├── signal.py               # Signal DTOs
│   │   └── indicator.py            # Indicator DTOs
│   └── tasks/
│       ├── __init__.py
│       ├── market_data_fetcher.py  # Background data fetch
│       ├── training_pipeline.py    # Model training
│       └── signal_publisher.py     # Signal broadcast
├── data/
│   ├── models/                     # Trained model files
│   │   ├── range_predictor/
│   │   ├── tpsl_classifier/
│   │   └── signal_classifier/
│   └── cache/                      # Cached market data
├── tests/
│   ├── __init__.py
│   ├── conftest.py
│   ├── unit/
│   └── integration/
├── scripts/
│   ├── train_models.py
│   └── evaluate_models.py
├── requirements.txt
├── Dockerfile
└── docker-compose.yml

---

Componentes Principales

1. FastAPI Application

# app/main.py
from fastapi import FastAPI
from fastapi.middleware.cors import CORSMiddleware
from contextlib import asynccontextmanager

from app.config.settings import settings
from app.api.routers import predictions, signals, indicators, models, health
from app.tasks.market_data_fetcher import MarketDataFetcher
from app.services.model_manager import ModelManager

@asynccontextmanager
async def lifespan(app: FastAPI):
    # Startup
    model_manager = ModelManager()
    await model_manager.load_all_models()

    market_fetcher = MarketDataFetcher()
    await market_fetcher.start()

    yield

    # Shutdown
    await market_fetcher.stop()

app = FastAPI(
    title="Trading Platform ML Engine",
    version="1.0.0",
    lifespan=lifespan
)

app.add_middleware(
    CORSMiddleware,
    allow_origins=settings.CORS_ORIGINS,
    allow_methods=["*"],
    allow_headers=["*"],
)

app.include_router(health.router, tags=["Health"])
app.include_router(predictions.router, prefix="/predictions", tags=["Predictions"])
app.include_router(signals.router, prefix="/signals", tags=["Signals"])
app.include_router(indicators.router, prefix="/indicators", tags=["Indicators"])
app.include_router(models.router, prefix="/models", tags=["Models"])

2. Configuration

# app/config/settings.py
from pydantic_settings import BaseSettings
from typing import List

class Settings(BaseSettings):
    # API
    API_HOST: str = "0.0.0.0"
    API_PORT: int = 8000
    API_KEY: str

    # Database
    DATABASE_URL: str
    REDIS_URL: str

    # Binance
    BINANCE_API_KEY: str
    BINANCE_API_SECRET: str

    # Models
    MODEL_PATH: str = "./data/models"
    SUPPORTED_SYMBOLS: List[str] = ["BTCUSDT", "ETHUSDT"]

    # Features
    DEFAULT_HORIZONS: List[int] = [6, 18, 36, 72]  # candles

    # CORS
    CORS_ORIGINS: List[str] = ["http://localhost:3000"]

    class Config:
        env_file = ".env"

settings = Settings()

---

Flujos de Datos

1. Flujo de Predicción

┌──────────┐    ┌──────────┐    ┌────────────┐    ┌──────────┐    ┌──────────┐
│  Client  │───▶│  Router  │───▶│  Service   │───▶│ Features │───▶│  Model   │
│          │    │          │    │            │    │ Builder  │    │          │
└──────────┘    └──────────┘    └────────────┘    └──────────┘    └──────────┘
                                      │                                 │
                                      ▼                                 │
                               ┌────────────┐                          │
                               │   Cache    │◀─────────────────────────┘
                               │  (Redis)   │
                               └────────────┘

Secuencia:

1. Cliente solicita predicción vía API
2. Router valida request y extrae parámetros
3. Service verifica cache (Redis)
4. Si no hay cache: Feature Builder genera features
5. Model realiza predicción
6. Resultado se cachea y retorna

2. Flujo de Señales

┌──────────────┐    ┌──────────────┐    ┌──────────────┐    ┌──────────────┐
│ Market Data  │───▶│   Feature    │───▶│    Signal    │───▶│  Publisher   │
│   Fetcher    │    │   Builder    │    │  Generator   │    │   (Redis)    │
└──────────────┘    └──────────────┘    └──────────────┘    └──────────────┘
      │                                        │                    │
      ▼                                        ▼                    ▼
┌──────────────┐                        ┌──────────────┐    ┌──────────────┐
│   Binance    │                        │  PostgreSQL  │    │   Backend    │
│     API      │                        │   (Logs)     │    │   Express    │
└──────────────┘                        └──────────────┘    └──────────────┘

---

Comunicación Entre Servicios

Backend ↔ ML Engine

# Comunicación HTTP
Protocol: REST over HTTPS
Authentication: API Key (X-API-Key header)
Format: JSON
Timeout: 30 seconds

# Endpoints principales
POST /predictions      # Solicitar predicción
POST /signals          # Generar señal
GET  /indicators       # Obtener indicadores
GET  /models/status    # Estado de modelos

ML Engine ↔ Redis

# Cache de predicciones
Key pattern: "prediction:{symbol}:{horizon}:{timestamp}"
TTL: 60 seconds (1 candle de 5min)

# Publicación de señales
Channel: "signals:{symbol}"
Message format: JSON serialized Signal object

ML Engine ↔ PostgreSQL

-- Solo para métricas y logs
CREATE TABLE ml.prediction_logs (
    id UUID PRIMARY KEY,
    symbol VARCHAR(20),
    horizon INTEGER,
    predicted_high DECIMAL(20, 8),
    predicted_low DECIMAL(20, 8),
    actual_high DECIMAL(20, 8),
    actual_low DECIMAL(20, 8),
    mae DECIMAL(10, 6),
    created_at TIMESTAMPTZ DEFAULT NOW()
);

CREATE TABLE ml.signal_logs (
    id UUID PRIMARY KEY,
    symbol VARCHAR(20),
    signal_type VARCHAR(10),
    confidence DECIMAL(5, 4),
    entry_price DECIMAL(20, 8),
    outcome VARCHAR(10),
    pnl_percent DECIMAL(10, 4),
    created_at TIMESTAMPTZ DEFAULT NOW()
);

---

Concurrencia y Escalabilidad

Workers Configuration

# uvicorn config
workers = 4  # CPU cores
worker_class = "uvicorn.workers.UvicornWorker"
timeout = 60
keepalive = 5

# Thread pool for CPU-bound tasks
from concurrent.futures import ThreadPoolExecutor
executor = ThreadPoolExecutor(max_workers=8)

Rate Limiting

from slowapi import Limiter
from slowapi.util import get_remote_address

limiter = Limiter(key_func=get_remote_address)

# Limits
@router.get("/predictions")
@limiter.limit("100/minute")
async def get_prediction():
    ...

Caching Strategy

# Redis caching decorator
from functools import wraps

def cache_prediction(ttl: int = 60):
    def decorator(func):
        @wraps(func)
        async def wrapper(symbol: str, horizon: int, *args, **kwargs):
            cache_key = f"pred:{symbol}:{horizon}:{current_candle_time()}"

            cached = await redis.get(cache_key)
            if cached:
                return json.loads(cached)

            result = await func(symbol, horizon, *args, **kwargs)
            await redis.setex(cache_key, ttl, json.dumps(result))
            return result
        return wrapper
    return decorator

---

Manejo de Errores

Exception Hierarchy

# app/core/exceptions.py
class MLEngineError(Exception):
    """Base exception for ML Engine"""
    pass

class ModelNotFoundError(MLEngineError):
    """Model file not found"""
    pass

class ModelLoadError(MLEngineError):
    """Failed to load model"""
    pass

class PredictionError(MLEngineError):
    """Error during prediction"""
    pass

class MarketDataError(MLEngineError):
    """Error fetching market data"""
    pass

class FeatureError(MLEngineError):
    """Error calculating features"""
    pass

Error Handlers

# app/main.py
from fastapi import Request
from fastapi.responses import JSONResponse

@app.exception_handler(MLEngineError)
async def ml_exception_handler(request: Request, exc: MLEngineError):
    return JSONResponse(
        status_code=500,
        content={
            "error": exc.__class__.__name__,
            "message": str(exc),
            "path": str(request.url)
        }
    )

---

Monitoreo y Observabilidad

Health Checks

# app/api/routers/health.py
from fastapi import APIRouter, Depends
from app.services.model_manager import ModelManager

router = APIRouter()

@router.get("/health")
async def health_check():
    return {"status": "healthy"}

@router.get("/health/detailed")
async def detailed_health(model_manager: ModelManager = Depends()):
    return {
        "status": "healthy",
        "models": model_manager.get_status(),
        "cache": await check_redis(),
        "database": await check_postgres(),
        "binance": await check_binance_connection()
    }

Metrics (Prometheus)

from prometheus_client import Counter, Histogram, Gauge

# Counters
predictions_total = Counter(
    'ml_predictions_total',
    'Total predictions made',
    ['symbol', 'horizon']
)

# Histograms
prediction_latency = Histogram(
    'ml_prediction_latency_seconds',
    'Prediction latency',
    buckets=[.01, .025, .05, .1, .25, .5, 1.0]
)

# Gauges
model_accuracy = Gauge(
    'ml_model_accuracy',
    'Current model accuracy',
    ['model_name']
)

---

Seguridad

API Key Authentication

# app/core/security.py
from fastapi import Security, HTTPException
from fastapi.security import APIKeyHeader

api_key_header = APIKeyHeader(name="X-API-Key")

async def validate_api_key(api_key: str = Security(api_key_header)):
    if api_key != settings.API_KEY:
        raise HTTPException(status_code=401, detail="Invalid API Key")
    return api_key

Input Validation

# app/schemas/prediction.py
from pydantic import BaseModel, validator
from typing import Literal

class PredictionRequest(BaseModel):
    symbol: str
    horizon: int

    @validator('symbol')
    def validate_symbol(cls, v):
        if v not in settings.SUPPORTED_SYMBOLS:
            raise ValueError(f'Symbol {v} not supported')
        return v

    @validator('horizon')
    def validate_horizon(cls, v):
        if v not in settings.DEFAULT_HORIZONS:
            raise ValueError(f'Horizon {v} not supported')
        return v

---

Deployment

Docker Configuration

# Dockerfile
FROM python:3.11-slim

WORKDIR /app

RUN pip install --no-cache-dir poetry
COPY pyproject.toml poetry.lock ./
RUN poetry install --no-dev

COPY app/ ./app/
COPY data/models/ ./data/models/

EXPOSE 8000

CMD ["poetry", "run", "uvicorn", "app.main:app", "--host", "0.0.0.0", "--port", "8000"]

Docker Compose

# docker-compose.yml
version: '3.8'

services:
  ml-engine:
    build: .
    ports:
      - "8000:8000"
    environment:
      - DATABASE_URL=${DATABASE_URL}
      - REDIS_URL=${REDIS_URL}
      - API_KEY=${ML_API_KEY}
      - BINANCE_API_KEY=${BINANCE_API_KEY}
      - BINANCE_API_SECRET=${BINANCE_API_SECRET}
    volumes:
      - ./data/models:/app/data/models
    depends_on:
      - redis
    restart: unless-stopped

  redis:
    image: redis:7-alpine
    ports:
      - "6379:6379"
    volumes:
      - redis_data:/data

volumes:
  redis_data:

---

Referencias

• RF-ML-001: Predicción de Precios
• ET-ML-002: Modelos XGBoost
• ET-ML-003: Feature Engineering
• FastAPI Documentation

---

Autor: Requirements-Analyst Fecha: 2025-12-05