main.py

import re
import string
from fastapi import FastAPI
from pydantic import BaseModel
from transformers import pipeline

# --- 1. Define the Cleaning Function (No Pandas needed) ---
def clean_text(text: str) -> str:
    """
    Comprehensive text cleaning for Bengali text.
    Optimized for API usage (removed pandas dependency).
    """
    if not text or not isinstance(text, str):
        return ""
    
    # Remove URLs
    text = re.sub(r'http\S+|www\S+|https\S+', '', text, flags=re.MULTILINE)
    
    # Remove Emails
    text = re.sub(r'\S+@\S+', '', text)
    
    # Remove Mentions/Hashtags
    text = re.sub(r'@\w+|#\w+', '', text)
    
    # Remove Emojis (Unicode range)
    text = re.sub(r'[\U00010000-\U0010ffff]', '', text)
    
    # Remove Punctuation (English + Bengali Dari/Double Dari)
    # Note: Ensure these characters are correct for your needs
    exclude_chars = string.punctuation + "।॥''"""
    text = text.translate(str.maketrans('', '', exclude_chars))
    
    # Normalize whitespace
    text = ' '.join(text.split())
    
    # Remove very short texts
    if len(text.strip()) < 2:
        return ""
        
    return text.strip()

# --- 2. Initialize App & Model ---
app = FastAPI()

# Load your model here (ensure files are uploaded if local)
# classifier = pipeline("text-classification", model="./my_model_path") 
# Or if testing with a generic one:
classifier = pipeline("sentiment-analysis")

# --- 3. Define Input Structure ---
class TextInput(BaseModel):
    text: str

# --- 4. Define Endpoints ---
@app.get("/")
def home():
    return {"message": "Bengali Model API is running"}

@app.post("/predict")
def predict(input_data: TextInput):
    # STEP 1: Clean the input
    cleaned_text = clean_text(input_data.text)
    
    # STEP 2: Handle empty results (if cleaning removed everything)
    if not cleaned_text:
        return {
            "error": "Input text contained only noise (urls, emojis, etc.)",
            "original_input": input_data.text,
            "cleaned_input": ""
        }

    # STEP 3: Pass CLEANED text to model
    prediction = classifier(cleaned_text)
    
    return {
        "original_input": input_data.text,
        "cleaned_input": cleaned_text,
        "prediction": prediction
    }