# app.py — HuggingFace Space (FastAPI)
# FREE CPU tier hosting for CreatorPulse sentiment model
# Space URL: https://ningaraddi-creatorpulse-api.hf.space

from fastapi import FastAPI
from fastapi.middleware.cors import CORSMiddleware
from pydantic import BaseModel
from typing import List
from transformers import pipeline
import os

app = FastAPI(title="CreatorPulse Sentiment API")

# ── CORS — allow calls from any origin (our React app) ─────────────────────────
app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_methods=["POST", "GET"],
    allow_headers=["*"],
)

# ── Load model once on startup ─────────────────────────────────────────────────
MODEL_REPO = "ningaraddi/creatorpulse-sentiment"

print(f"Loading model: {MODEL_REPO}")

classifier = pipeline(
    "text-classification",
    model=MODEL_REPO,
    tokenizer=MODEL_REPO,
    truncation=True,
    max_length=128,
    device=-1,        # CPU — free tier
    top_k=None,       # Return all labels with scores
)

print("Model loaded!")

# ── Schemas ────────────────────────────────────────────────────────────────────
class ClassifyRequest(BaseModel):
    inputs: List[str]

class Prediction(BaseModel):
    label:      str
    confidence: float
    text:       str

class ClassifyResponse(BaseModel):
    predictions: List[Prediction]

# ── Health check ───────────────────────────────────────────────────────────────
@app.get("/")
def health():
    return {"status": "ok", "model": MODEL_REPO}

# ── Classify endpoint ──────────────────────────────────────────────────────────
@app.post("/classify", response_model=ClassifyResponse)
def classify(request: ClassifyRequest):
    texts = request.inputs[:100]   # Max 100 per call

    results = classifier(texts, batch_size=8)

    predictions = []
    for text, label_list in zip(texts, results):
        top = max(label_list, key=lambda x: x["score"])

        # Normalize label names
        label_map = {"LABEL_0": "NEGATIVE", "LABEL_1": "NEUTRAL", "LABEL_2": "POSITIVE"}
        label = label_map.get(top["label"], top["label"])

        predictions.append(Prediction(
            text=text,
            label=label,
            confidence=round(top["score"], 4),
        ))

    return ClassifyResponse(predictions=predictions)