Spaces:
Build error
Build error
File size: 7,286 Bytes
c40c447 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 |
"""
Servicio de dominio para backtesting.
Este servicio encapsula la l贸gica de validaci贸n de modelos,
cumpliendo con SRP y DIP.
"""
import numpy as np
from dataclasses import dataclass
from typing import List
from app.domain.interfaces.forecast_model import IForecastModel
from app.domain.interfaces.data_transformer import IDataTransformer
from app.domain.models.time_series import TimeSeries
from app.domain.models.forecast_config import ForecastConfig
from app.utils.logger import setup_logger
logger = setup_logger(__name__)
@dataclass
class BacktestMetrics:
"""
M茅tricas de evaluaci贸n de un backtest.
Attributes:
mae: Mean Absolute Error
mape: Mean Absolute Percentage Error (%)
rmse: Root Mean Squared Error
wql: Weighted Quantile Loss (para cuantil 0.5)
"""
mae: float
mape: float
rmse: float
wql: float
def to_dict(self) -> dict:
"""Serializa las m茅tricas"""
return {
"mae": self.mae,
"mape": self.mape,
"rmse": self.rmse,
"wql": self.wql
}
@dataclass
class BacktestResult:
"""
Resultado completo de un backtest.
Attributes:
metrics: M茅tricas de evaluaci贸n
forecast: Valores pronosticados
actuals: Valores reales
timestamps: Timestamps del per铆odo de prueba
"""
metrics: BacktestMetrics
forecast: List[float]
actuals: List[float]
timestamps: List[str]
def to_dict(self) -> dict:
"""Serializa el resultado"""
return {
"metrics": self.metrics.to_dict(),
"forecast": self.forecast,
"actuals": self.actuals,
"timestamps": self.timestamps
}
class BacktestService:
"""
Servicio de dominio para backtesting de modelos.
Realiza validaci贸n de modelos separando la serie en train/test
y comparando pron贸sticos con valores reales.
Attributes:
model: Modelo de forecasting
transformer: Transformador de datos
Example:
>>> service = BacktestService(model, transformer)
>>> series = TimeSeries(values=[100, 102, 105, 103, 108, 112, 115])
>>> result = service.simple_backtest(series, test_length=3)
>>> result.metrics.mae < 5 # Buen modelo
True
"""
def __init__(
self,
model: IForecastModel,
transformer: IDataTransformer
):
"""
Inicializa el servicio.
Args:
model: Implementaci贸n de IForecastModel
transformer: Implementaci贸n de IDataTransformer
"""
self.model = model
self.transformer = transformer
logger.info("BacktestService initialized")
def simple_backtest(
self,
series: TimeSeries,
test_length: int,
config: ForecastConfig = None
) -> BacktestResult:
"""
Realiza un backtest simple: train/test split.
Separa la serie en train (hist贸rico) y test (validaci贸n),
genera pron贸stico para el per铆odo test y calcula m茅tricas.
Args:
series: Serie temporal completa
test_length: N煤mero de puntos para test (final de la serie)
config: Configuraci贸n del forecast (opcional)
Returns:
BacktestResult: Resultado con m茅tricas y pron贸sticos
Raises:
ValueError: Si test_length >= longitud de la serie
Example:
>>> series = TimeSeries(values=[100, 102, 105, 103, 108])
>>> result = service.simple_backtest(series, test_length=2)
>>> len(result.forecast)
2
"""
logger.info(
f"Running simple backtest for series '{series.series_id}' "
f"(total_length={series.length}, test_length={test_length})"
)
# Validar
if test_length >= series.length:
raise ValueError(
f"test_length ({test_length}) debe ser menor que "
f"la longitud de la serie ({series.length})"
)
if test_length < 1:
raise ValueError(f"test_length debe ser >= 1, recibido: {test_length}")
# Configuraci贸n por defecto si no se proporciona
if config is None:
config = ForecastConfig(
prediction_length=test_length,
quantile_levels=[0.5], # Solo mediana para backtest
freq=series.freq
)
else:
# Ajustar prediction_length
config.prediction_length = test_length
# Separar train/test
train_length = series.length - test_length
train_series = series.get_subset(0, train_length)
test_values = series.values[train_length:]
logger.debug(f"Train length: {train_length}, Test length: {test_length}")
# Construir DataFrame de train
context_df = self.transformer.build_context_df(
values=train_series.values,
timestamps=train_series.timestamps,
series_id=series.series_id,
freq=config.freq
)
# Predecir
pred_df = self.model.predict(
context_df=context_df,
prediction_length=test_length,
quantile_levels=[0.5]
)
# Parsear resultado
result = self.transformer.parse_prediction_result(
pred_df=pred_df,
quantile_levels=[0.5]
)
forecast = np.array(result["median"], dtype=float)
actuals = np.array(test_values, dtype=float)
# Calcular m茅tricas
metrics = self._calculate_metrics(forecast, actuals)
logger.info(
f"Backtest completed: MAE={metrics.mae:.2f}, "
f"MAPE={metrics.mape:.2f}%, RMSE={metrics.rmse:.2f}"
)
return BacktestResult(
metrics=metrics,
forecast=forecast.tolist(),
actuals=actuals.tolist(),
timestamps=result["timestamps"]
)
def _calculate_metrics(
self,
forecast: np.ndarray,
actuals: np.ndarray
) -> BacktestMetrics:
"""
Calcula m茅tricas de evaluaci贸n.
Args:
forecast: Valores pronosticados
actuals: Valores reales
Returns:
BacktestMetrics: M茅tricas calculadas
"""
# MAE: Mean Absolute Error
mae = float(np.mean(np.abs(actuals - forecast)))
# MAPE: Mean Absolute Percentage Error
eps = 1e-8
mape = float(np.mean(np.abs((actuals - forecast) / (actuals + eps)))) * 100.0
# RMSE: Root Mean Squared Error
rmse = float(np.sqrt(np.mean((actuals - forecast) ** 2)))
# WQL: Weighted Quantile Loss (para cuantil 0.5 = MAE/2)
tau = 0.5
diff = actuals - forecast
wql = float(np.mean(np.maximum(tau * diff, (tau - 1) * diff)))
return BacktestMetrics(
mae=mae,
mape=mape,
rmse=rmse,
wql=wql
)
|