"""
FomoFeed - User Profiler AI
Creates semantic user profiles using Turkish BERT embeddings
"""

from fastapi import FastAPI, HTTPException
from pydantic import BaseModel
from transformers import AutoTokenizer, AutoModel
import torch
import numpy as np
from datetime import datetime
from collections import Counter
import uvicorn

app = FastAPI(title="FomoFeed User Profiler", version="1.0.0")

# Load Turkish BERT model
MODEL_NAME = "dbmdz/bert-base-turkish-cased"
tokenizer = None
model = None

def load_model():
    global tokenizer, model
    try:
        tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
        model = AutoModel.from_pretrained(MODEL_NAME)
        model.eval()
        print(f"✅ Model loaded: {MODEL_NAME}")
    except Exception as e:
        print(f"⚠️ Model load failed: {e}")

# Load model at startup
load_model()

class UserActivity(BaseModel):
    user_id: int
    post_captions: list[str] = []
    moment_captions: list[str] = []
    liked_tags: list[str] = []
    saved_tags: list[str] = []
    commented_tags: list[str] = []
    engagement_hours: list[int] = []
    engagement_types: list[str] = []  # view, like, comment, save

class UserProfileResponse(BaseModel):
    user_id: int
    interests: list[str]
    content_preference: dict
    activity_pattern: dict
    engagement_style: dict
    optimal_hours: list[int]
    confidence: float

def get_bert_embedding(text: str) -> np.ndarray:
    """
    Get BERT embedding for text
    """
    if not text or model is None or tokenizer is None:
        return np.zeros(768)
    
    try:
        inputs = tokenizer(text, return_tensors="pt", max_length=512, truncation=True, padding=True)
        with torch.no_grad():
            outputs = model(**inputs)
        
        # Use CLS token embedding
        embedding = outputs.last_hidden_state[:, 0, :].numpy()[0]
        return embedding
    except Exception as e:
        print(f"Embedding error: {e}")
        return np.zeros(768)

def analyze_interests(activity: UserActivity) -> list[str]:
    """
    Extract user interests from their activity
    """
    all_tags = (
        activity.liked_tags * 2 +  # Weight likes lower
        activity.saved_tags * 5 +  # Weight saves higher
        activity.commented_tags * 3  # Weight comments medium
    )
    
    if not all_tags:
        return []
    
    # Count frequency
    tag_counts = Counter(all_tags)
    
    # Get top 10 tags
    top_tags = [tag for tag, count in tag_counts.most_common(10)]
    
    return top_tags

def analyze_content_preference(activity: UserActivity) -> dict:
    """
    Analyze user's content consumption preferences
    """
    total_engagement = len(activity.engagement_types)
    
    if total_engagement == 0:
        return {
            "passive_consumer": 0.5,
            "active_engager": 0.5,
            "content_creator": 0.0
        }
    
    # Count engagement types
    engagement_counts = Counter(activity.engagement_types)
    
    views = engagement_counts.get("view", 0)
    likes = engagement_counts.get("like", 0)
    comments = engagement_counts.get("comment", 0)
    saves = engagement_counts.get("save", 0)
    
    # Calculate scores
    passive_score = views / total_engagement if views > 0 else 0
    active_score = (likes + saves) / total_engagement if (likes + saves) > 0 else 0
    creator_score = len(activity.post_captions + activity.moment_captions) / 10  # Normalize
    creator_score = min(creator_score, 1.0)
    
    return {
        "passive_consumer": round(passive_score, 2),
        "active_engager": round(active_score, 2),
        "content_creator": round(creator_score, 2),
        "engagement_depth": round((comments + saves) / max(total_engagement, 1), 2)
    }

def analyze_activity_pattern(activity: UserActivity) -> dict:
    """
    Analyze when user is most active
    """
    if not activity.engagement_hours:
        return {
            "peak_hours": [19, 20, 21],
            "activity_distribution": "unknown",
            "timezone_pattern": "evening"
        }
    
    hour_counts = Counter(activity.engagement_hours)
    
    # Get top 5 hours
    peak_hours = [hour for hour, count in hour_counts.most_common(5)]
    
    # Determine pattern
    morning = sum(1 for h in activity.engagement_hours if 6 <= h < 12)
    afternoon = sum(1 for h in activity.engagement_hours if 12 <= h < 18)
    evening = sum(1 for h in activity.engagement_hours if 18 <= h < 24)
    night = sum(1 for h in activity.engagement_hours if 0 <= h < 6)
    
    total = len(activity.engagement_hours)
    
    if total > 0:
        distribution = {
            "morning": round(morning / total, 2),
            "afternoon": round(afternoon / total, 2),
            "evening": round(evening / total, 2),
            "night": round(night / total, 2)
        }
        
        # Primary pattern
        primary = max(distribution.items(), key=lambda x: x[1])[0]
    else:
        distribution = {"morning": 0.25, "afternoon": 0.25, "evening": 0.25, "night": 0.25}
        primary = "evening"
    
    return {
        "peak_hours": peak_hours[:3],
        "activity_distribution": distribution,
        "timezone_pattern": primary
    }

def analyze_engagement_style(activity: UserActivity) -> dict:
    """
    Classify user's engagement style
    """
    engagement_counts = Counter(activity.engagement_types)
    
    total = sum(engagement_counts.values())
    if total == 0:
        return {
            "style": "new_user",
            "interaction_rate": 0.0,
            "content_saver": False,
            "commenter": False
        }
    
    # Calculate metrics
    views = engagement_counts.get("view", 0)
    likes = engagement_counts.get("like", 0)
    comments = engagement_counts.get("comment", 0)
    saves = engagement_counts.get("save", 0)
    
    interaction_rate = (likes + comments + saves) / total
    
    # Classify style
    if interaction_rate < 0.1:
        style = "lurker"
    elif interaction_rate < 0.3:
        style = "casual"
    elif interaction_rate < 0.6:
        style = "active"
    else:
        style = "power_user"
    
    return {
        "style": style,
        "interaction_rate": round(interaction_rate, 2),
        "content_saver": saves > (total * 0.05),
        "commenter": comments > (total * 0.02),
        "engagement_breakdown": {
            "views": views,
            "likes": likes,
            "comments": comments,
            "saves": saves
        }
    }

def calculate_confidence(activity: UserActivity) -> float:
    """
    Calculate confidence score based on data volume
    """
    # Count data points
    total_captions = len(activity.post_captions) + len(activity.moment_captions)
    total_tags = len(activity.liked_tags) + len(activity.saved_tags) + len(activity.commented_tags)
    total_engagements = len(activity.engagement_types)
    
    # Score each dimension
    caption_score = min(total_captions / 20, 1.0) * 0.3
    tag_score = min(total_tags / 50, 1.0) * 0.4
    engagement_score = min(total_engagements / 100, 1.0) * 0.3
    
    confidence = caption_score + tag_score + engagement_score
    
    return round(confidence, 2)

@app.get("/")
def root():
    return {
        "service": "FomoFeed User Profiler",
        "status": "active",
        "model": "turkish-bert" if model else "rule-based",
        "version": "1.0.0"
    }

@app.get("/health")
def health():
    return {
        "status": "healthy",
        "model_loaded": model is not None,
        "timestamp": datetime.now().isoformat()
    }

@app.post("/profile", response_model=UserProfileResponse)
def create_profile(activity: UserActivity):
    """
    Create comprehensive user profile from activity data
    """
    try:
        # Analyze different aspects
        interests = analyze_interests(activity)
        content_pref = analyze_content_preference(activity)
        activity_pattern = analyze_activity_pattern(activity)
        engagement_style = analyze_engagement_style(activity)
        confidence = calculate_confidence(activity)
        
        # Extract optimal hours
        optimal_hours = activity_pattern["peak_hours"]
        
        return UserProfileResponse(
            user_id=activity.user_id,
            interests=interests,
            content_preference=content_pref,
            activity_pattern=activity_pattern,
            engagement_style=engagement_style,
            optimal_hours=optimal_hours,
            confidence=confidence
        )
    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))

@app.post("/embedding")
def get_user_embedding(activity: UserActivity):
    """
    Generate BERT embedding for user based on their content
    """
    try:
        # Combine all text
        all_text = " ".join(
            activity.post_captions +
            activity.moment_captions +
            activity.liked_tags +
            activity.saved_tags
        )
        
        if not all_text.strip():
            return {
                "user_id": activity.user_id,
                "embedding": [0.0] * 768,
                "note": "No text data available"
            }
        
        # Get embedding
        embedding = get_bert_embedding(all_text[:1000])  # Limit to 1000 chars
        
        return {
            "user_id": activity.user_id,
            "embedding": embedding.tolist(),
            "dimension": 768
        }
    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))

@app.post("/batch_profile")
def batch_profile(activities: list[UserActivity]):
    """
    Create profiles for multiple users
    """
    try:
        profiles = []
        for activity in activities:
            interests = analyze_interests(activity)
            content_pref = analyze_content_preference(activity)
            activity_pattern = analyze_activity_pattern(activity)
            engagement_style = analyze_engagement_style(activity)
            confidence = calculate_confidence(activity)
            
            profiles.append({
                "user_id": activity.user_id,
                "interests": interests,
                "content_preference": content_pref,
                "activity_pattern": activity_pattern,
                "engagement_style": engagement_style,
                "optimal_hours": activity_pattern["peak_hours"],
                "confidence": confidence
            })
        
        return {"profiles": profiles}
    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))

if __name__ == "__main__":
    uvicorn.run(app, host="0.0.0.0", port=7860)