app/api/routes.py

import json
import unicodedata
import re
import pandas as pd # <--- ADDED THIS IMPORT
from fastapi import APIRouter, HTTPException
from pydantic import BaseModel
from typing import List, Optional
import numpy as np
import faiss
from sentence_transformers import SentenceTransformer

from app.core.ml_manager import ml_manager
from app.services.recommender import RecommenderService
from app.services.explainer import RecommendationExplainer

router = APIRouter()

embedder = None

def get_embedder():
    global embedder
    if embedder is None:
        print("Loading mixedbread-ai embedding model for semantic search...")
        embedder = SentenceTransformer("mixedbread-ai/mxbai-embed-large-v1")
        print("Embedding model ready!")
    return embedder

def safe_str(val) -> str:
    if isinstance(val, (list, np.ndarray)):
        return str(val)
    if pd.isna(val):
        return ""
    return str(val).strip()

def normalize_text(text: str) -> str:
    if not isinstance(text, str): return ""
    text = unicodedata.normalize('NFKD', text).encode('ascii', 'ignore').decode('utf-8')
    text = re.sub(r'[^a-zA-Z0-9\s]', '', text)
    text = re.sub(r'\s+', ' ', text)
    return text.lower().strip()

class Interaction(BaseModel):
    itemIndex: int
    rating: float
    timestamp: int

class RecommendRequest(BaseModel):
    interactions: List[Interaction]
    top_k: int = 15

@router.post("/recommend")
def get_recommendations(req: RecommendRequest):
    interactions_list = [{"itemIndex": i.itemIndex, "rating": i.rating, "timestamp": i.timestamp} for i in req.interactions]
    
    # Extract raw history for the explainer
    history_indices = [i.itemIndex for i in req.interactions]
    history_ratings = [i.rating for i in req.interactions]

    recommendations = RecommenderService.get_recommendations(
        interactions=interactions_list, 
        top_k=req.top_k
    )
    
    results = []
    for rec in recommendations:
        idx = rec["item_index"]
        try:
            meta = ml_manager.movie_meta.loc[idx]
            badges = rec.get("badges", ["Recommended"])
            
            # --- Generate the Explanation on the fly! ---
            reason = RecommendationExplainer.generate_reason(
                target_index=int(idx), 
                history_indices=history_indices, 
                history_ratings=history_ratings, 
                sources=badges
            )
            
            results.append({
                "item_index": int(idx),
                "tmdb_id": str(meta["tmdb_id"]),
                "title": str(meta["title"]),
                "poster_path": safe_str(meta.get("poster_path", "")),
                "sources": badges,
                "reason": reason  # <--- Added to payload
            })
        except KeyError:
            continue
            
    return {"status": "success", "recommendations": results}

@router.get("/trending")
def get_trending():
    # scalar index 7 is popularity (as seen in your cold-start logic)
    top_pop_idx = np.argsort(ml_manager.scalars[:, 7])[-10:][::-1]
    
    results = []
    for idx in top_pop_idx:
        try:
            meta = ml_manager.movie_meta.loc[idx]
            # Ensure we only grab movies that actually have a backdrop image
            if pd.isna(meta.get("backdrop_path")) or not meta.get("backdrop_path"):
                continue
                
            results.append({
                "item_index": int(idx),
                "tmdb_id": str(meta["tmdb_id"]),
                "title": str(meta["title"]),
                "backdrop_path": safe_str(meta.get("backdrop_path", "")),
                "sources": ["Trending This Week"]
            })
        except KeyError:
            continue
            
    return {"status": "success", "trending": results[:8]} # Return top 8

class SemanticSearchRequest(BaseModel):
    query: str
    top_k: int = 15

@router.post("/semantic-search")
def semantic_search(req: SemanticSearchRequest):
    query_norm = normalize_text(req.query)
    
    normalized_titles = ml_manager.movie_meta['title'].apply(normalize_text)
    mask = normalized_titles.str.contains(query_norm, regex=False, na=False)
    
    if not mask.any():
        query_squashed = query_norm.replace(" ", "")
        mask = normalized_titles.str.replace(" ", "").str.contains(query_squashed, regex=False, na=False)

    title_matches_df = ml_manager.movie_meta[mask].copy()
    title_matches_df['title_len'] = title_matches_df['title'].str.len()
    title_matches_df = title_matches_df.sort_values('title_len').head(10)
    
    title_results = []
    for idx, row in title_matches_df.iterrows():
        title_results.append({
            "item_index": int(idx), "tmdb_id": str(row["tmdb_id"]),
            "title": str(row["title"]), "poster_path": safe_str(row.get("poster_path", "")), "sources": ["Exact Match"]
        })

    raw_embedding = get_embedder().encode(req.query, convert_to_numpy=True).astype(np.float32)
    query_vector = raw_embedding.reshape(1, -1)
    faiss.normalize_L2(query_vector)

    distances, indices = ml_manager.faiss_index.search(query_vector, req.top_k)
    
    semantic_results = []
    title_match_indices = set(title_matches_df.index)
    
    for rank, idx in enumerate(indices[0]):
        if idx in title_match_indices: continue
        try:
            meta = ml_manager.movie_meta.loc[idx]
            semantic_results.append({
                "item_index": int(idx), "tmdb_id": str(meta["tmdb_id"]),
                "title": str(meta["title"]), "poster_path": safe_str(meta.get("poster_path", "")), "sources": ["Semantic"]
            })
        except KeyError:
            continue
            
    return {"status": "success", "title_matches": title_results, "semantic_matches": semantic_results}

class MetadataRequest(BaseModel):
    item_indices: List[int]

@router.post("/metadata")
def get_batch_metadata(req: MetadataRequest):
    results = []
    for idx in req.item_indices:
        try:
            meta = ml_manager.movie_meta.loc[idx]
            results.append({
                "item_index": int(idx),
                "tmdb_id": str(meta["tmdb_id"]),
                "title": str(meta["title"]),
                "poster_path": safe_str(meta.get("poster_path", ""))
            })
        except KeyError:
            continue
    return {"metadata": results}

@router.get("/movie/{item_index}")
def get_movie_details(item_index: int):
    try:
        meta = ml_manager.movie_meta.loc[item_index]
    except KeyError:
        raise HTTPException(status_code=404, detail="Movie not found")
        
    tmdb_id = meta["tmdb_id"]
    
    clean_details = {
        "title": meta.get("title", "Unknown Title"),
        "overview": meta.get("overview", "Plot overview not available."),
        "release_date": str(meta.get("release_date", "")), 
        "runtime": int(meta.get("runtime", 0)) if not pd.isna(meta.get("runtime", 0)) else 0,
        "genres": meta.get("genres", []) if isinstance(meta.get("genres", []), list) else [],
        "backdrop_path": safe_str(meta.get("backdrop_path", "")),
        "poster_path": safe_str(meta.get("poster_path", ""))
    }

    ease_row = np.array(ml_manager.ease_matrix[item_index])
    ease_row[item_index] = -np.inf
    
    ease_results = []
    if np.max(ease_row) > 0.015:
        top_similar_idx = np.argsort(ease_row)[-10:][::-1]
        for idx in top_similar_idx:
            try:
                smeta = ml_manager.movie_meta.loc[idx]
                ease_results.append({
                    "item_index": int(idx), "tmdb_id": str(smeta["tmdb_id"]),
                    "title": str(smeta["title"]), "poster_path": safe_str(smeta.get("poster_path", "")), "sources": ["Collaborative"]
                })
            except KeyError: continue

    item_vector = np.zeros((1, ml_manager.d_semantic), dtype=np.float32)
    ml_manager.faiss_index.reconstruct(item_index, item_vector[0])
    distances, semantic_idx = ml_manager.faiss_index.search(item_vector, 11)
    
    semantic_results = []
    for idx in semantic_idx[0]:
        if idx == item_index: continue
        try:
            smeta = ml_manager.movie_meta.loc[idx]
            semantic_results.append({
                "item_index": int(idx), "tmdb_id": str(smeta["tmdb_id"]),
                "title": str(smeta["title"]), "poster_path": safe_str(smeta.get("poster_path", "")), "sources": ["Semantic"]
            })
        except KeyError: continue

    return {
        "status": "success", "details": clean_details,
        "people_also_watch": ease_results, "similar_theme": semantic_results[:10]
    }

# --- BROWSE ENDPOINTS ---

@router.get("/browse/options/{category}")
def get_browse_options(category: str):
    """Returns the list of available keys for a category (e.g., all genres)"""
    if category not in ml_manager.browse_data:
        raise HTTPException(status_code=404, detail="Category not found")
    
    data = ml_manager.browse_data[category]
    if isinstance(data, dict):
        # Return sorted keys (e.g., ["Action", "Comedy", ...])
        return {"options": sorted(list(data.keys()))}
    return {"options": []}

@router.get("/browse/movies")
def get_browse_movies(category: str, key: Optional[str] = None, page: int = 1, limit: int = 30):
    """Returns movies for a specific category and optional key."""
    if category not in ml_manager.browse_data:
        raise HTTPException(status_code=404, detail="Category not found")
        
    data = ml_manager.browse_data[category]
    
    if isinstance(data, dict):
        if not key or key not in data:
            raise HTTPException(status_code=404, detail="Key not found in category")
        item_indices = data[key]
    else:
        # For direct arrays like 'anime', 'family', 'awards'
        item_indices = data

    # Pagination
    start = (page - 1) * limit
    end = start + limit
    paged_indices = item_indices[start:end]

    results = []
    for idx in paged_indices:
        try:
            meta = ml_manager.movie_meta.loc[idx]
            results.append({
                "item_index": int(idx),
                "tmdb_id": str(meta["tmdb_id"]),
                "title": str(meta["title"]),
                "poster_path": safe_str(meta.get("poster_path", "")),
                "sources": [category.capitalize()]
            })
        except KeyError:
            continue
            
    return {
        "status": "success", 
        "total_results": len(item_indices),
        "movies": results
    }