from fastapi import FastAPI
from pydantic import BaseModel
from huggingface_hub import hf_hub_download
from sentence_transformers import SentenceTransformer
import xgboost as xgb
import numpy as np

app = FastAPI(
    title="Check-In Classifier API",
    description="Detects whether a message is a CHECKIN or NOT_CHECKIN using XGBoost + MiniLM.",
    version="1.0.0"
)

# -------------------------------------
# Load embedding model (MiniLM)
# -------------------------------------
embedder = SentenceTransformer("all-MiniLM-L6-v2")

# -------------------------------------
# Download XGBoost model from HF Hub
# -------------------------------------
MODEL_REPO = "mjpsm/checkin_or_not_model"
MODEL_FILE = "checkin_or_not_classifierV3.json"

model_path = hf_hub_download(
    repo_id=MODEL_REPO,
    filename=MODEL_FILE
)

booster = xgb.Booster()
booster.load_model(model_path)

# -------------------------------------
# Request/Response Models
# -------------------------------------
class PredictionRequest(BaseModel):
    text: str

class PredictionResponse(BaseModel):
    label: str
    score: float


@app.get("/")
def home():
    return {"status": "online", "message": "Check-In Classification API is running."}


@app.post("/predict", response_model=PredictionResponse)
def predict(req: PredictionRequest):

    # Convert text → embedding
    embedding = embedder.encode([req.text])

    # Prepare for XGBoost
    dmatrix = xgb.DMatrix(embedding)

    # Predict score
    score = float(booster.predict(dmatrix)[0])
    label = "CHECKIN" if score >= 0.5 else "NOT_CHECKIN"

    return PredictionResponse(label=label, score=score)