Spaces:
Runtime error
Runtime error
First attempt
Browse files- Dockerfile +30 -0
- backend/config.py +37 -0
- backend/main.py +412 -0
- models/label-encoder.joblib +0 -0
- models/model.json +0 -0
- requirements.txt +16 -0
Dockerfile
ADDED
|
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
FROM python:3.13.5-slim
|
| 2 |
+
|
| 3 |
+
RUN useradd -m -u 1000 user
|
| 4 |
+
USER user
|
| 5 |
+
|
| 6 |
+
WORKDIR /app
|
| 7 |
+
|
| 8 |
+
RUN apt-get update && apt-get install -y \
|
| 9 |
+
build-essential \
|
| 10 |
+
curl \
|
| 11 |
+
git \
|
| 12 |
+
&& rm -rf /var/lib/apt/lists/*
|
| 13 |
+
|
| 14 |
+
# Copiar las dependencias primero para aprovechar el cache
|
| 15 |
+
COPY --chown=user ./requirements.txt requirements.txt
|
| 16 |
+
RUN pip install --no-cache-dir --upgrade -r requirements.txt
|
| 17 |
+
|
| 18 |
+
# Copiar el resto del código de la aplicación
|
| 19 |
+
COPY --chown=user . /app
|
| 20 |
+
|
| 21 |
+
# Variables de entorno
|
| 22 |
+
ENV PYTHONUNBUFFERED=1
|
| 23 |
+
ENV MODEL_PATH=/app/models/model.json
|
| 24 |
+
ENV ENCODER_PATH=/app/models/label-encoder.joblib
|
| 25 |
+
|
| 26 |
+
# Exponer el puerto 7860
|
| 27 |
+
EXPOSE 7860
|
| 28 |
+
|
| 29 |
+
# Comando para ejecutar la aplicación
|
| 30 |
+
CMD ["uvicorn", "backend.main:app", "--host", "0.0.0.0", "--port", "7860"]
|
backend/config.py
ADDED
|
@@ -0,0 +1,37 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from pathlib import Path
|
| 2 |
+
import os
|
| 3 |
+
|
| 4 |
+
BASE_DIR = Path(__file__).resolve().parent.parent
|
| 5 |
+
|
| 6 |
+
|
| 7 |
+
def _default_model_path():
|
| 8 |
+
return str(BASE_DIR / "models" / "model.json")
|
| 9 |
+
|
| 10 |
+
|
| 11 |
+
def _default_encoder_path():
|
| 12 |
+
return str(BASE_DIR / "models" / "label-encoder.joblib")
|
| 13 |
+
|
| 14 |
+
|
| 15 |
+
# Prefer pydantic-settings (separate package) for pydantic v2 compatibility.
|
| 16 |
+
try:
|
| 17 |
+
from pydantic_settings import BaseSettings
|
| 18 |
+
|
| 19 |
+
class Settings(BaseSettings):
|
| 20 |
+
MODEL_PATH: str = _default_model_path()
|
| 21 |
+
ENCODER_PATH: str = _default_encoder_path()
|
| 22 |
+
HOST: str = "0.0.0.0"
|
| 23 |
+
PORT: int = 8000
|
| 24 |
+
|
| 25 |
+
class Config:
|
| 26 |
+
env_file = ".env"
|
| 27 |
+
|
| 28 |
+
settings = Settings()
|
| 29 |
+
except Exception:
|
| 30 |
+
# Fallback mínimo sin dependencias: leer variables de entorno o usar defaults.
|
| 31 |
+
class Settings:
|
| 32 |
+
MODEL_PATH: str = os.environ.get("MODEL_PATH", _default_model_path())
|
| 33 |
+
ENCODER_PATH: str = os.environ.get("ENCODER_PATH", _default_encoder_path())
|
| 34 |
+
HOST: str = os.environ.get("HOST", "0.0.0.0")
|
| 35 |
+
PORT: int = int(os.environ.get("PORT", "8000"))
|
| 36 |
+
|
| 37 |
+
settings = Settings()
|
backend/main.py
ADDED
|
@@ -0,0 +1,412 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# backend/main.py
|
| 2 |
+
"""API de predicción para candidatos Kepler (KOI).
|
| 3 |
+
Recibe una fila de características y devuelve una clasificación: `CONFIRMED` o `FALSE POSITIVE`."""
|
| 4 |
+
|
| 5 |
+
from fastapi import FastAPI, HTTPException, UploadFile, File
|
| 6 |
+
from pydantic import BaseModel, Field
|
| 7 |
+
import logging
|
| 8 |
+
from fastapi.middleware.cors import CORSMiddleware
|
| 9 |
+
import joblib
|
| 10 |
+
import pandas as pd
|
| 11 |
+
from typing import List, Dict, Any
|
| 12 |
+
from pathlib import Path
|
| 13 |
+
import xgboost as xgb
|
| 14 |
+
import numpy as np
|
| 15 |
+
import io
|
| 16 |
+
|
| 17 |
+
|
| 18 |
+
from .config import settings
|
| 19 |
+
|
| 20 |
+
|
| 21 |
+
app = FastAPI(
|
| 22 |
+
title="Exoplanet Prediction API",
|
| 23 |
+
description="Servicio de ML para clasificar Candidatos Kepler (KOI) en CONFIRMED o FALSE POSITIVE."
|
| 24 |
+
)
|
| 25 |
+
|
| 26 |
+
# Logging básico
|
| 27 |
+
logging.basicConfig(level=logging.INFO)
|
| 28 |
+
logger = logging.getLogger("exoplanet_api")
|
| 29 |
+
|
| 30 |
+
# Rutas y settings
|
| 31 |
+
MODELO_XGB_PATH = settings.MODEL_PATH
|
| 32 |
+
ENCODER_PATH = settings.ENCODER_PATH
|
| 33 |
+
|
| 34 |
+
MODELO = None
|
| 35 |
+
LE = None
|
| 36 |
+
COLUMNAS_ESPERADAS: List[str] = []
|
| 37 |
+
|
| 38 |
+
selected_features = [
|
| 39 |
+
"koi_prad",
|
| 40 |
+
"koi_dicco_msky",
|
| 41 |
+
"koi_dor",
|
| 42 |
+
"koi_max_mult_ev",
|
| 43 |
+
"koi_model_snr",
|
| 44 |
+
"koi_max_sngle_ev",
|
| 45 |
+
"koi_fwm_stat_sig",
|
| 46 |
+
"koi_num_transits",
|
| 47 |
+
"koi_ror",
|
| 48 |
+
"koi_fwm_srao",
|
| 49 |
+
"koi_srho",
|
| 50 |
+
"koi_insol",
|
| 51 |
+
"koi_duration",
|
| 52 |
+
"koi_teq",
|
| 53 |
+
"koi_period",
|
| 54 |
+
]
|
| 55 |
+
|
| 56 |
+
COLUMNAS_ESPERADAS = selected_features
|
| 57 |
+
|
| 58 |
+
|
| 59 |
+
class PredictPayload(BaseModel):
|
| 60 |
+
koi_prad: float = Field(...)
|
| 61 |
+
koi_dicco_msky: float = Field(...)
|
| 62 |
+
koi_dor: float = Field(...)
|
| 63 |
+
koi_max_mult_ev: float = Field(...)
|
| 64 |
+
koi_model_snr: float = Field(...)
|
| 65 |
+
koi_max_sngle_ev: float = Field(...)
|
| 66 |
+
koi_fwm_stat_sig: float = Field(...)
|
| 67 |
+
koi_num_transits: float = Field(...)
|
| 68 |
+
koi_ror: float = Field(...)
|
| 69 |
+
koi_fwm_srao: float = Field(...)
|
| 70 |
+
koi_srho: float = Field(...)
|
| 71 |
+
koi_insol: float = Field(...)
|
| 72 |
+
koi_duration: float = Field(...)
|
| 73 |
+
koi_teq: float = Field(...)
|
| 74 |
+
koi_period: float = Field(...)
|
| 75 |
+
|
| 76 |
+
class Config:
|
| 77 |
+
example_obj = {
|
| 78 |
+
"koi_prad": 1.0,
|
| 79 |
+
"koi_dicco_msky": 0.5,
|
| 80 |
+
"koi_dor": 0.1,
|
| 81 |
+
"koi_max_mult_ev": 0.2,
|
| 82 |
+
"koi_model_snr": 10.0,
|
| 83 |
+
"koi_max_sngle_ev": 5.0,
|
| 84 |
+
"koi_fwm_stat_sig": 0.01,
|
| 85 |
+
"koi_num_transits": 3,
|
| 86 |
+
"koi_ror": 0.02,
|
| 87 |
+
"koi_fwm_srao": 0.1,
|
| 88 |
+
"koi_srho": 0.05,
|
| 89 |
+
"koi_insol": 1360.0,
|
| 90 |
+
"koi_duration": 2.5,
|
| 91 |
+
"koi_teq": 250.0,
|
| 92 |
+
"koi_period": 365.0
|
| 93 |
+
}
|
| 94 |
+
# Pydantic v1 uses `schema_extra`; v2 uses `json_schema_extra`.
|
| 95 |
+
schema_extra = {"example": example_obj}
|
| 96 |
+
json_schema_extra = {"example": example_obj}
|
| 97 |
+
|
| 98 |
+
|
| 99 |
+
class PredictionResponse(BaseModel):
|
| 100 |
+
status: str
|
| 101 |
+
prediction: str
|
| 102 |
+
probability_confirmed: float
|
| 103 |
+
probability_false_positive: float
|
| 104 |
+
model_used: str
|
| 105 |
+
|
| 106 |
+
class Config:
|
| 107 |
+
example_obj = {
|
| 108 |
+
"status": "success",
|
| 109 |
+
"prediction": "FALSE POSITIVE",
|
| 110 |
+
"probability_confirmed": 0.009942352771759033,
|
| 111 |
+
"probability_false_positive": 0.990057647228241,
|
| 112 |
+
"model_used": "XGBoost (JSON)"
|
| 113 |
+
}
|
| 114 |
+
schema_extra = {"example": example_obj}
|
| 115 |
+
json_schema_extra = {"example": example_obj}
|
| 116 |
+
|
| 117 |
+
@app.on_event("startup")
|
| 118 |
+
def load_model_and_encoder():
|
| 119 |
+
"""Cargar el encoder y el modelo en el evento de arranque.
|
| 120 |
+
|
| 121 |
+
Las rutas vienen de `backend.config.settings` y pueden sobreescribirse por env vars.
|
| 122 |
+
"""
|
| 123 |
+
global MODELO, LE
|
| 124 |
+
try:
|
| 125 |
+
LE = joblib.load(str(ENCODER_PATH))
|
| 126 |
+
|
| 127 |
+
MODELO = xgb.XGBClassifier()
|
| 128 |
+
MODELO.load_model(str(MODELO_XGB_PATH))
|
| 129 |
+
|
| 130 |
+
print(f"Modelo XGBoost (JSON) cargado exitosamente desde: {MODELO_XGB_PATH}")
|
| 131 |
+
print(f"Clases disponibles: {LE.classes_}")
|
| 132 |
+
|
| 133 |
+
except FileNotFoundError:
|
| 134 |
+
print(f"ERROR: No se encontró uno o más archivos. Revise las rutas: {MODELO_XGB_PATH}, {ENCODER_PATH}")
|
| 135 |
+
MODELO = None
|
| 136 |
+
except Exception as e:
|
| 137 |
+
print(f"ERROR al cargar componentes: {e}")
|
| 138 |
+
MODELO = None
|
| 139 |
+
|
| 140 |
+
|
| 141 |
+
origins = ["*"]
|
| 142 |
+
|
| 143 |
+
app.add_middleware(
|
| 144 |
+
CORSMiddleware,
|
| 145 |
+
allow_origins=origins,
|
| 146 |
+
allow_credentials=True,
|
| 147 |
+
allow_methods=["*"],
|
| 148 |
+
allow_headers=["*"],
|
| 149 |
+
)
|
| 150 |
+
# Endpoint de prueba
|
| 151 |
+
|
| 152 |
+
@app.get("/")
|
| 153 |
+
def read_root():
|
| 154 |
+
"""Endpoint de prueba para verificar si la API funciona."""
|
| 155 |
+
ready = MODELO is not None
|
| 156 |
+
return {"message": "Exoplanet ML API está funcionando correctamente.", "model_ready": ready}
|
| 157 |
+
|
| 158 |
+
|
| 159 |
+
@app.get("/health")
|
| 160 |
+
def health_check():
|
| 161 |
+
"""Liveness probe simple."""
|
| 162 |
+
return {"status": "ok"}
|
| 163 |
+
|
| 164 |
+
|
| 165 |
+
@app.get("/ready")
|
| 166 |
+
def readiness_check():
|
| 167 |
+
"""Readiness probe — devuelve 200 solo si el modelo y el encoder están cargados."""
|
| 168 |
+
if MODELO is None or LE is None:
|
| 169 |
+
raise HTTPException(status_code=503, detail="Servicio no listo")
|
| 170 |
+
return {"status": "ready"}
|
| 171 |
+
|
| 172 |
+
|
| 173 |
+
@app.get("/model_metadata")
|
| 174 |
+
def model_metadata():
|
| 175 |
+
"""Devuelve metadata simple del modelo: features esperadas y clases."""
|
| 176 |
+
if LE is None:
|
| 177 |
+
raise HTTPException(status_code=503, detail="El decodificador de etiquetas no cargó.")
|
| 178 |
+
return {"features": COLUMNAS_ESPERADAS, "classes": LE.classes_.tolist()}
|
| 179 |
+
|
| 180 |
+
def imputar_nulos_con_media_condicional(data: pd.DataFrame) -> pd.DataFrame:
|
| 181 |
+
"""Imputa valores nulos usando la media por columna (solo columnas numéricas).
|
| 182 |
+
|
| 183 |
+
Si no hay columnas numéricas, devuelve el DataFrame sin cambios.
|
| 184 |
+
"""
|
| 185 |
+
try:
|
| 186 |
+
return data.fillna(data.mean())
|
| 187 |
+
except Exception:
|
| 188 |
+
return data
|
| 189 |
+
|
| 190 |
+
|
| 191 |
+
@app.post("/predict_csv_row", response_model=PredictionResponse)
|
| 192 |
+
def predict_exoplanet(payload: PredictPayload):
|
| 193 |
+
"""
|
| 194 |
+
Recibe los datos de una fila de exoplanetas (JSON) y devuelve una predicción.
|
| 195 |
+
"""
|
| 196 |
+
if MODELO is None or not COLUMNAS_ESPERADAS:
|
| 197 |
+
raise HTTPException(status_code=503, detail="Servicio de predicción inactivo.")
|
| 198 |
+
|
| 199 |
+
if LE is None:
|
| 200 |
+
raise HTTPException(status_code=503, detail="El decodificador de etiquetas no cargó.")
|
| 201 |
+
|
| 202 |
+
try:
|
| 203 |
+
logger.info("Entrada de predicción recibida")
|
| 204 |
+
data = payload.dict()
|
| 205 |
+
input_data = pd.DataFrame([data])
|
| 206 |
+
|
| 207 |
+
missing_cols = [c for c in COLUMNAS_ESPERADAS if c not in input_data.columns]
|
| 208 |
+
if missing_cols:
|
| 209 |
+
raise ValueError(f"Faltan columnas críticas: {missing_cols}")
|
| 210 |
+
|
| 211 |
+
input_data = input_data[COLUMNAS_ESPERADAS]
|
| 212 |
+
|
| 213 |
+
input_data_processed = imputar_nulos_con_media_condicional(input_data)
|
| 214 |
+
|
| 215 |
+
# Asegurarse de que los datos sean numéricos antes de predecir
|
| 216 |
+
try:
|
| 217 |
+
input_array = input_data_processed.astype(float).values
|
| 218 |
+
except Exception as e:
|
| 219 |
+
raise ValueError(f"Error al convertir columnas a numérico: {e}")
|
| 220 |
+
|
| 221 |
+
# Usar la API sklearn del XGBClassifier (arrays / DataFrame) en lugar de DMatrix
|
| 222 |
+
prediction_num = MODELO.predict(input_array)
|
| 223 |
+
prediction_num_class = np.asarray(prediction_num).astype(int)
|
| 224 |
+
|
| 225 |
+
# Obtener probabilidades de forma segura
|
| 226 |
+
try:
|
| 227 |
+
proba = MODELO.predict_proba(input_array)[0].tolist()
|
| 228 |
+
except Exception:
|
| 229 |
+
# Fallback: construir probabilidades determinísticas según la predicción
|
| 230 |
+
proba = [0.0, 0.0]
|
| 231 |
+
if prediction_num_class[0] == 1:
|
| 232 |
+
proba[1] = 1.0
|
| 233 |
+
else:
|
| 234 |
+
proba[0] = 1.0
|
| 235 |
+
|
| 236 |
+
# Determinar el índice asociado a la clase "CONFIRMED" en el LabelEncoder
|
| 237 |
+
try:
|
| 238 |
+
classes_list = list(LE.classes_)
|
| 239 |
+
idx_confirmed = classes_list.index('CONFIRMED')
|
| 240 |
+
except Exception:
|
| 241 |
+
# Si no existe la etiqueta exacta, asumimos orden [negativo, positivo]
|
| 242 |
+
idx_confirmed = 1 if len(LE.classes_) == 2 else 0
|
| 243 |
+
|
| 244 |
+
prob_confirmed = float(proba[idx_confirmed]) if idx_confirmed < len(proba) else 0.0
|
| 245 |
+
prob_false_positive = float(proba[1 - idx_confirmed]) if (1 - idx_confirmed) < len(proba) else 0.0
|
| 246 |
+
|
| 247 |
+
result = LE.inverse_transform([prediction_num_class[0]])[0]
|
| 248 |
+
|
| 249 |
+
resp = {
|
| 250 |
+
"status": "success",
|
| 251 |
+
"prediction": str(result),
|
| 252 |
+
"probability_confirmed": prob_confirmed,
|
| 253 |
+
"probability_false_positive": prob_false_positive,
|
| 254 |
+
"model_used": "XGBoost (JSON)"
|
| 255 |
+
}
|
| 256 |
+
|
| 257 |
+
logger.info(f"Predicción: {resp}")
|
| 258 |
+
|
| 259 |
+
return resp
|
| 260 |
+
|
| 261 |
+
except ValueError as e:
|
| 262 |
+
raise HTTPException(status_code=400, detail=f"Error en los datos de entrada: {e}")
|
| 263 |
+
|
| 264 |
+
except Exception as e:
|
| 265 |
+
raise HTTPException(status_code=500, detail=f"Error interno del servidor de ML: {str(e)}")
|
| 266 |
+
|
| 267 |
+
|
| 268 |
+
@app.post("/predict_csv", response_model=List[PredictionResponse])
|
| 269 |
+
async def predict_from_csv(file: UploadFile = File(...)):
|
| 270 |
+
"""Recibe un archivo CSV (multipart/form-data) con varias filas y devuelve predicciones para cada fila.
|
| 271 |
+
|
| 272 |
+
Procesa el CSV en chunks para evitar cargar todo en memoria. El CSV debe contener
|
| 273 |
+
las mismas columnas listadas en `COLUMNAS_ESPERADAS`.
|
| 274 |
+
"""
|
| 275 |
+
if MODELO is None or not COLUMNAS_ESPERADAS:
|
| 276 |
+
raise HTTPException(status_code=503, detail="Servicio de predicci\u00f3n inactivo.")
|
| 277 |
+
if LE is None:
|
| 278 |
+
raise HTTPException(status_code=503, detail="El decodificador de etiquetas no carg\u00f3.")
|
| 279 |
+
|
| 280 |
+
try:
|
| 281 |
+
# Leer todo el contenido de UploadFile en memoria como texto y usar
|
| 282 |
+
# pd.read_csv sobre un StringIO con chunksize. Evita wrappers que
|
| 283 |
+
# requieren métodos no disponibles en SpooledTemporaryFile.
|
| 284 |
+
content = await file.read()
|
| 285 |
+
text = content.decode('utf-8')
|
| 286 |
+
text_stream = io.StringIO(text)
|
| 287 |
+
|
| 288 |
+
chunk_size = 1000
|
| 289 |
+
results = []
|
| 290 |
+
|
| 291 |
+
for chunk in pd.read_csv(text_stream, chunksize=chunk_size):
|
| 292 |
+
# Validar columnas en el primer chunk
|
| 293 |
+
missing_cols = [c for c in COLUMNAS_ESPERADAS if c not in chunk.columns]
|
| 294 |
+
if missing_cols:
|
| 295 |
+
raise ValueError(f"Faltan columnas cr\u00edticas en el CSV: {missing_cols}")
|
| 296 |
+
|
| 297 |
+
chunk = chunk[COLUMNAS_ESPERADAS]
|
| 298 |
+
chunk_processed = imputar_nulos_con_media_condicional(chunk)
|
| 299 |
+
|
| 300 |
+
try:
|
| 301 |
+
arr = chunk_processed.astype(float).values
|
| 302 |
+
except Exception as e:
|
| 303 |
+
raise ValueError(f"Error al convertir columnas a num\u00e9rico: {e}")
|
| 304 |
+
|
| 305 |
+
preds = MODELO.predict(arr)
|
| 306 |
+
try:
|
| 307 |
+
probas = MODELO.predict_proba(arr)
|
| 308 |
+
except Exception:
|
| 309 |
+
probas = np.zeros((len(preds), 2))
|
| 310 |
+
for i, p in enumerate(preds):
|
| 311 |
+
if int(p) == 1:
|
| 312 |
+
probas[i, 1] = 1.0
|
| 313 |
+
else:
|
| 314 |
+
probas[i, 0] = 1.0
|
| 315 |
+
|
| 316 |
+
classes_list = list(LE.classes_)
|
| 317 |
+
try:
|
| 318 |
+
idx_confirmed = classes_list.index('CONFIRMED')
|
| 319 |
+
except Exception:
|
| 320 |
+
idx_confirmed = 1 if len(classes_list) == 2 else 0
|
| 321 |
+
|
| 322 |
+
for i in range(len(preds)):
|
| 323 |
+
pred_label = LE.inverse_transform([int(preds[i])])[0]
|
| 324 |
+
prob_confirmed = float(probas[i, idx_confirmed]) if idx_confirmed < probas.shape[1] else 0.0
|
| 325 |
+
prob_false = float(probas[i, 1 - idx_confirmed]) if (1 - idx_confirmed) < probas.shape[1] else 0.0
|
| 326 |
+
|
| 327 |
+
results.append({
|
| 328 |
+
"status": "success",
|
| 329 |
+
"prediction": str(pred_label),
|
| 330 |
+
"probability_confirmed": prob_confirmed,
|
| 331 |
+
"probability_false_positive": prob_false,
|
| 332 |
+
"model_used": "XGBoost (JSON)"
|
| 333 |
+
})
|
| 334 |
+
|
| 335 |
+
return results
|
| 336 |
+
|
| 337 |
+
except ValueError as e:
|
| 338 |
+
raise HTTPException(status_code=400, detail=f"Error en los datos de entrada: {e}")
|
| 339 |
+
except Exception as e:
|
| 340 |
+
raise HTTPException(status_code=500, detail=f"Error interno del servidor de ML: {str(e)}")
|
| 341 |
+
|
| 342 |
+
|
| 343 |
+
@app.post("/predict_csv_raw", response_model=List[PredictionResponse])
|
| 344 |
+
async def predict_from_csv_raw(body: bytes = None):
|
| 345 |
+
"""Aceptar CSV en texto plano (Content-Type: text/csv) en el body de la petición.
|
| 346 |
+
|
| 347 |
+
Esto es útil para integraciones que envían el CSV en el cuerpo (no multipart).
|
| 348 |
+
"""
|
| 349 |
+
if MODELO is None or not COLUMNAS_ESPERADAS:
|
| 350 |
+
raise HTTPException(status_code=503, detail="Servicio de predicci\u00f3n inactivo.")
|
| 351 |
+
if LE is None:
|
| 352 |
+
raise HTTPException(status_code=503, detail="El decodificador de etiquetas no carg\u00f3.")
|
| 353 |
+
|
| 354 |
+
if not body:
|
| 355 |
+
raise HTTPException(status_code=400, detail="Request body vacío. Se espera CSV en el body.")
|
| 356 |
+
|
| 357 |
+
try:
|
| 358 |
+
text = body.decode('utf-8')
|
| 359 |
+
text_stream = io.StringIO(text)
|
| 360 |
+
|
| 361 |
+
chunk_size = 1000
|
| 362 |
+
results = []
|
| 363 |
+
|
| 364 |
+
for chunk in pd.read_csv(text_stream, chunksize=chunk_size):
|
| 365 |
+
missing_cols = [c for c in COLUMNAS_ESPERADAS if c not in chunk.columns]
|
| 366 |
+
if missing_cols:
|
| 367 |
+
raise ValueError(f"Faltan columnas cr\u00edticas en el CSV: {missing_cols}")
|
| 368 |
+
|
| 369 |
+
chunk = chunk[COLUMNAS_ESPERADAS]
|
| 370 |
+
chunk_processed = imputar_nulos_con_media_condicional(chunk)
|
| 371 |
+
|
| 372 |
+
try:
|
| 373 |
+
arr = chunk_processed.astype(float).values
|
| 374 |
+
except Exception as e:
|
| 375 |
+
raise ValueError(f"Error al convertir columnas a num\u00e9rico: {e}")
|
| 376 |
+
|
| 377 |
+
preds = MODELO.predict(arr)
|
| 378 |
+
try:
|
| 379 |
+
probas = MODELO.predict_proba(arr)
|
| 380 |
+
except Exception:
|
| 381 |
+
probas = np.zeros((len(preds), 2))
|
| 382 |
+
for i, p in enumerate(preds):
|
| 383 |
+
if int(p) == 1:
|
| 384 |
+
probas[i, 1] = 1.0
|
| 385 |
+
else:
|
| 386 |
+
probas[i, 0] = 1.0
|
| 387 |
+
|
| 388 |
+
classes_list = list(LE.classes_)
|
| 389 |
+
try:
|
| 390 |
+
idx_confirmed = classes_list.index('CONFIRMED')
|
| 391 |
+
except Exception:
|
| 392 |
+
idx_confirmed = 1 if len(classes_list) == 2 else 0
|
| 393 |
+
|
| 394 |
+
for i in range(len(preds)):
|
| 395 |
+
pred_label = LE.inverse_transform([int(preds[i])])[0]
|
| 396 |
+
prob_confirmed = float(probas[i, idx_confirmed]) if idx_confirmed < probas.shape[1] else 0.0
|
| 397 |
+
prob_false = float(probas[i, 1 - idx_confirmed]) if (1 - idx_confirmed) < probas.shape[1] else 0.0
|
| 398 |
+
|
| 399 |
+
results.append({
|
| 400 |
+
"status": "success",
|
| 401 |
+
"prediction": str(pred_label),
|
| 402 |
+
"probability_confirmed": prob_confirmed,
|
| 403 |
+
"probability_false_positive": prob_false,
|
| 404 |
+
"model_used": "XGBoost (JSON)"
|
| 405 |
+
})
|
| 406 |
+
|
| 407 |
+
return results
|
| 408 |
+
|
| 409 |
+
except ValueError as e:
|
| 410 |
+
raise HTTPException(status_code=400, detail=f"Error en los datos de entrada: {e}")
|
| 411 |
+
except Exception as e:
|
| 412 |
+
raise HTTPException(status_code=500, detail=f"Error interno del servidor de ML: {str(e)}")
|
models/label-encoder.joblib
ADDED
|
Binary file (502 Bytes). View file
|
|
|
models/model.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
requirements.txt
ADDED
|
@@ -0,0 +1,16 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Servidor Web y Utilidades
|
| 2 |
+
fastapi
|
| 3 |
+
uvicorn[standard] # Para correr el servidor de forma asíncrona (ASGI)
|
| 4 |
+
requests # Para scripts cliente y tests
|
| 5 |
+
|
| 6 |
+
# Machine Learning y Serialización
|
| 7 |
+
joblib # Para cargar el modelo serializado (XGBoost + SMOTE)
|
| 8 |
+
pandas # Para manejar y estructurar los datos (DataFrames)
|
| 9 |
+
pydantic # Para la validación de los datos de entrada (JSON) en FastAPI
|
| 10 |
+
numpy # Dependencia base de Pandas y Scikit-learn
|
| 11 |
+
typing # Base de Python, pero bueno listarla si se usa extensivamente
|
| 12 |
+
|
| 13 |
+
# Dependencias del Modelo
|
| 14 |
+
xgboost-cpu
|
| 15 |
+
scikit-learn # Base para el pipeline, el preprocesamiento y LabelEncoder
|
| 16 |
+
python-multipart # Necesario para UploadFile / multipart en FastAPI
|