app/services/ml/ml_engine.py

"""
Rankora ML Engine
4 ML Features:
1. Fake Review Detector
2. Price Prediction (Linear Regression)
3. Demand Forecasting (Seasonality)
4. Niche Scorer
"""

import math
import random
from typing import Optional
from datetime import datetime, timezone


# ═══════════════════════════════════════════════════════════════
# 1. FAKE REVIEW DETECTOR
# ═══════════════════════════════════════════════════════════════

def detect_fake_reviews(
    rating: float,
    review_count: int,
    rating_distribution: Optional[dict] = None,
    monthly_sales_estimate: int = 100,
    asin: str = ""
) -> dict:
    """
    Detect suspicious review patterns using rule-based ML heuristics.
    Returns a suspicion score 0-100 and detailed signals.
    """
    signals = []
    suspicion_score = 0

    # Signal 1: Perfect or near-perfect rating with many reviews
    if rating >= 4.8 and review_count > 500:
        suspicion_score += 20
        signals.append({
            "signal": "Suspiciously High Rating",
            "detail": f"{rating}★ with {review_count:,} reviews — real products rarely maintain 4.8+ at scale",
            "severity": "high",
            "weight": 20
        })
    elif rating >= 4.9 and review_count > 100:
        suspicion_score += 25
        signals.append({
            "signal": "Near-Perfect Rating",
            "detail": f"{rating}★ is unusually high — may indicate review manipulation",
            "severity": "high",
            "weight": 25
        })

    # Signal 2: Review velocity vs sales ratio
    # If reviews >> expected for sales level, reviews may be incentivized
    expected_reviews = monthly_sales_estimate * 0.02  # ~2% of sales leave reviews
    if review_count > 0 and monthly_sales_estimate > 0:
        review_rate = review_count / max(monthly_sales_estimate * 6, 1)  # 6 months
        if review_rate > 0.15:  # more than 15% of buyers reviewing is suspicious
            suspicion_score += 20
            signals.append({
                "signal": "High Review Velocity",
                "detail": f"Review rate ({review_rate:.1%}) is unusually high vs estimated sales — possible incentivized reviews",
                "severity": "medium",
                "weight": 20
            })

    # Signal 3: Very low review count but high BSR
    # High sales rank but very few reviews = possibly review reset / new ASIN
    if monthly_sales_estimate > 500 and review_count < 20:
        suspicion_score += 15
        signals.append({
            "signal": "Sales/Review Mismatch",
            "detail": f"High estimated sales ({monthly_sales_estimate:,}/mo) but only {review_count} reviews — possible review manipulation reset",
            "severity": "medium",
            "weight": 15
        })

    # Signal 4: Rating distribution analysis (if provided)
    if rating_distribution:
        five_star_pct = rating_distribution.get("5_star", 0)
        one_star_pct  = rating_distribution.get("1_star", 0)
        if five_star_pct > 85:
            suspicion_score += 20
            signals.append({
                "signal": "Extreme 5-Star Concentration",
                "detail": f"{five_star_pct}% five-star reviews — legitimate products rarely exceed 80%",
                "severity": "high",
                "weight": 20
            })
        if one_star_pct < 1 and review_count > 200:
            suspicion_score += 10
            signals.append({
                "signal": "Missing Negative Reviews",
                "detail": f"Only {one_star_pct}% 1-star reviews out of {review_count:,} — statistically unlikely for real products",
                "severity": "medium",
                "weight": 10
            })

    # Signal 5: Round number review counts often indicate manipulation
    if review_count > 100:
        str_count = str(review_count)
        trailing_zeros = len(str_count) - len(str_count.rstrip("0"))
        if trailing_zeros >= 2:
            suspicion_score += 10
            signals.append({
                "signal": "Suspicious Review Count",
                "detail": f"Round number ({review_count:,}) with trailing zeros — may indicate manipulated count",
                "severity": "low",
                "weight": 10
            })

    suspicion_score = min(suspicion_score, 100)

    if suspicion_score >= 60:
        verdict = "HIGH RISK — Likely Fake Reviews"
        verdict_color = "#EF4444"
        recommendation = "Avoid this product — high probability of review manipulation"
    elif suspicion_score >= 35:
        verdict = "MODERATE RISK — Suspicious Patterns"
        verdict_color = "#F59E0B"
        recommendation = "Investigate further before competing in this space"
    elif suspicion_score >= 15:
        verdict = "LOW RISK — Minor Concerns"
        verdict_color = "#FB923C"
        recommendation = "Reviews appear mostly genuine with minor anomalies"
    else:
        verdict = "AUTHENTIC — Reviews Appear Genuine"
        verdict_color = "#10B981"
        recommendation = "No significant red flags detected in review patterns"

    return {
        "suspicion_score": suspicion_score,
        "verdict": verdict,
        "verdict_color": verdict_color,
        "recommendation": recommendation,
        "signals": signals,
        "signals_count": len(signals),
        "is_suspicious": suspicion_score >= 35,
        "analysis_basis": {
            "rating": rating,
            "review_count": review_count,
            "monthly_sales_estimate": monthly_sales_estimate,
        }
    }


# ═══════════════════════════════════════════════════════════════
# 2. PRICE PREDICTION (Linear Regression)
# ═══════════════════════════════════════════════════════════════

def predict_price(
    price_history: list,
    days_ahead: int = 7
) -> dict:
    """
    Predict future price using linear regression on price history.
    price_history: list of {"price": float, "recorded_at": str}
    """
    if not price_history or len(price_history) < 3:
        return {
            "predicted_price": None,
            "confidence": "low",
            "trend": "insufficient_data",
            "message": "Need at least 3 data points for prediction"
        }

    # Extract prices
    prices = [float(p["price"]) for p in price_history if p.get("price")]
    if len(prices) < 3:
        return {"predicted_price": None, "confidence": "low", "trend": "insufficient_data"}

    n = len(prices)
    x = list(range(n))

    # Linear regression: y = mx + b
    sum_x  = sum(x)
    sum_y  = sum(prices)
    sum_xy = sum(x[i] * prices[i] for i in range(n))
    sum_x2 = sum(xi ** 2 for xi in x)

    denom = n * sum_x2 - sum_x ** 2
    if denom == 0:
        slope = 0
    else:
        slope = (n * sum_xy - sum_x * sum_y) / denom
    intercept = (sum_y - slope * sum_x) / n

    # Predict future price
    future_x = n - 1 + days_ahead
    predicted = round(intercept + slope * future_x, 2)
    predicted = max(0.01, predicted)  # price can't be negative

    # Calculate R² for confidence
    y_mean = sum_y / n
    ss_tot = sum((p - y_mean) ** 2 for p in prices)
    ss_res = sum((prices[i] - (intercept + slope * x[i])) ** 2 for i in range(n))
    r2 = 1 - (ss_res / ss_tot) if ss_tot > 0 else 0

    # Determine trend
    current_price = prices[-1]
    price_change_pct = ((predicted - current_price) / current_price) * 100

    if slope > 0.01:
        trend = "rising"
        trend_emoji = "📈"
        trend_color = "#EF4444"
    elif slope < -0.01:
        trend = "falling"
        trend_emoji = "📉"
        trend_color = "#10B981"
    else:
        trend = "stable"
        trend_emoji = "➡️"
        trend_color = "#6B7280"

    confidence = "high" if r2 > 0.7 else "medium" if r2 > 0.4 else "low"

    # Price stats
    min_price = min(prices)
    max_price = max(prices)
    avg_price = sum(prices) / len(prices)

    return {
        "predicted_price": predicted,
        "current_price": current_price,
        "price_change": round(predicted - current_price, 2),
        "price_change_pct": round(price_change_pct, 1),
        "trend": trend,
        "trend_emoji": trend_emoji,
        "trend_color": trend_color,
        "confidence": confidence,
        "r_squared": round(r2, 3),
        "days_ahead": days_ahead,
        "slope_per_day": round(slope, 4),
        "price_stats": {
            "min": round(min_price, 2),
            "max": round(max_price, 2),
            "avg": round(avg_price, 2),
            "volatility": round((max_price - min_price) / avg_price * 100, 1)
        },
        "recommendation": (
            f"Price expected to {'rise' if trend == 'rising' else 'fall' if trend == 'falling' else 'stay stable'} "
            f"by {abs(price_change_pct):.1f}% over next {days_ahead} days"
        )
    }


# ═══════════════════════════════════════════════════════════════
# 3. DEMAND FORECASTING (Seasonality Detection)
# ═══════════════════════════════════════════════════════════════

def forecast_demand(
    bsr_history: list,
    category: str = "general",
    current_month: Optional[int] = None
) -> dict:
    """
    Forecast demand using BSR trends and seasonal patterns.
    Lower BSR = higher demand.
    """
    if current_month is None:
        current_month = datetime.now(timezone.utc).month

    # Category seasonal multipliers (month 1-12)
    seasonal_patterns = {
        "electronics": [0.8, 0.7, 0.8, 0.9, 0.9, 0.8, 0.9, 0.9, 1.0, 1.0, 1.3, 1.4],
        "home": [0.9, 0.8, 1.0, 1.1, 1.2, 1.1, 1.0, 0.9, 0.9, 1.0, 1.1, 1.0],
        "toys": [0.7, 0.6, 0.7, 0.7, 0.8, 0.8, 0.9, 0.9, 1.0, 1.1, 1.3, 1.8],
        "sports": [0.9, 0.9, 1.1, 1.2, 1.3, 1.3, 1.2, 1.1, 1.0, 0.9, 0.8, 0.8],
        "kitchen": [0.9, 0.9, 1.0, 1.0, 1.0, 0.9, 0.9, 0.9, 0.9, 1.0, 1.1, 1.2],
        "general":  [1.0, 0.9, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.1, 1.2],
    }

    # Match category
    cat_key = "general"
    category_lower = category.lower()
    for key in seasonal_patterns:
        if key in category_lower:
            cat_key = key
            break

    pattern = seasonal_patterns[cat_key]
    current_multiplier = pattern[current_month - 1]
    next_month = (current_month % 12)
    next_multiplier = pattern[next_month]

    # BSR trend analysis
    bsr_trend = "stable"
    bsr_change_pct = 0
    demand_trend = "stable"

    if bsr_history and len(bsr_history) >= 3:
        bsr_values = [h.get("bsr") for h in bsr_history if h.get("bsr")]
        if len(bsr_values) >= 3:
            old_bsr = sum(bsr_values[:3]) / 3
            new_bsr = sum(bsr_values[-3:]) / 3
            bsr_change_pct = ((new_bsr - old_bsr) / old_bsr) * 100

            if bsr_change_pct < -10:
                bsr_trend = "improving"  # BSR going down = demand going up
                demand_trend = "increasing"
            elif bsr_change_pct > 10:
                bsr_trend = "declining"
                demand_trend = "decreasing"
            else:
                bsr_trend = "stable"
                demand_trend = "stable"

    # Seasonal score for each month
    months = ["Jan","Feb","Mar","Apr","May","Jun","Jul","Aug","Sep","Oct","Nov","Dec"]
    monthly_forecast = []
    for i, mult in enumerate(pattern):
        monthly_forecast.append({
            "month": months[i],
            "demand_index": round(mult * 100),
            "is_peak": mult >= 1.2,
            "is_low": mult <= 0.8,
        })

    peak_months = [months[i] for i, m in enumerate(pattern) if m >= 1.2]
    low_months  = [months[i] for i, m in enumerate(pattern) if m <= 0.8]

    if current_multiplier >= 1.2:
        season_status = "PEAK SEASON"
        season_color = "#10B981"
        season_advice = "Great time to sell — high demand period"
    elif current_multiplier >= 1.0:
        season_status = "NORMAL SEASON"
        season_color = "#3B82F6"
        season_advice = "Average demand — maintain stock levels"
    elif current_multiplier >= 0.85:
        season_status = "SLOW SEASON"
        season_color = "#F59E0B"
        season_advice = "Reduce inventory — lower demand period"
    else:
        season_status = "OFF SEASON"
        season_color = "#EF4444"
        season_advice = "Avoid heavy stock — very low demand"

    demand_change_next = round((next_multiplier - current_multiplier) / current_multiplier * 100, 1)

    return {
        "current_season": season_status,
        "season_color": season_color,
        "season_advice": season_advice,
        "current_demand_index": round(current_multiplier * 100),
        "next_month_demand_index": round(next_multiplier * 100),
        "demand_change_next_month": demand_change_next,
        "demand_trend": demand_trend,
        "bsr_trend": bsr_trend,
        "bsr_change_pct": round(bsr_change_pct, 1),
        "category": cat_key,
        "peak_months": peak_months,
        "low_months": low_months,
        "monthly_forecast": monthly_forecast,
        "recommendation": (
            f"{season_status}: {season_advice}. "
            f"Next month demand expected to {'increase' if demand_change_next > 5 else 'decrease' if demand_change_next < -5 else 'stay similar'} "
            f"by {abs(demand_change_next):.0f}%."
        )
    }


# ═══════════════════════════════════════════════════════════════
# 4. NICHE SCORER
# ═══════════════════════════════════════════════════════════════

def score_niche(
    keyword: str,
    products: list,  # list of {bsr, price, reviews, rating}
    category: str = "general"
) -> dict:
    """
    Score an entire niche/keyword based on multiple product data points.
    Returns opportunity score 0-100 and detailed breakdown.
    """
    if not products:
        return {"error": "No products provided for niche analysis"}

    prices   = [p.get("price", 0)   for p in products if p.get("price")]
    bsrs     = [p.get("bsr", 0)     for p in products if p.get("bsr")]
    reviews  = [p.get("reviews", 0) for p in products if p.get("reviews") is not None]
    ratings  = [p.get("rating", 0)  for p in products if p.get("rating")]

    def avg(lst): return sum(lst) / len(lst) if lst else 0
    def median(lst):
        s = sorted(lst)
        n = len(s)
        return (s[n//2] + s[n//2-1]) / 2 if n % 2 == 0 else s[n//2]

    avg_price   = avg(prices)
    avg_bsr     = avg(bsrs)
    avg_reviews = avg(reviews)
    avg_rating  = avg(ratings)
    med_reviews = median(reviews) if reviews else 0

    scores = {}

    # 1. Demand Score (based on BSR)
    if avg_bsr < 1000:
        scores["demand"] = 95
    elif avg_bsr < 5000:
        scores["demand"] = 80
    elif avg_bsr < 20000:
        scores["demand"] = 65
    elif avg_bsr < 100000:
        scores["demand"] = 45
    else:
        scores["demand"] = 20

    # 2. Competition Score (lower reviews = easier to compete)
    if avg_reviews < 50:
        scores["competition"] = 90
    elif avg_reviews < 200:
        scores["competition"] = 75
    elif avg_reviews < 500:
        scores["competition"] = 55
    elif avg_reviews < 2000:
        scores["competition"] = 35
    else:
        scores["competition"] = 15

    # 3. Profitability Score (based on price)
    if avg_price >= 25 and avg_price <= 70:
        scores["profitability"] = 85
    elif avg_price >= 15 and avg_price < 25:
        scores["profitability"] = 65
    elif avg_price >= 70 and avg_price <= 150:
        scores["profitability"] = 70
    elif avg_price > 150:
        scores["profitability"] = 50
    else:
        scores["profitability"] = 35

    # 4. Quality Gap Score (lower ratings = easier to beat with better product)
    if avg_rating < 3.8:
        scores["quality_gap"] = 90
    elif avg_rating < 4.2:
        scores["quality_gap"] = 70
    elif avg_rating < 4.5:
        scores["quality_gap"] = 50
    else:
        scores["quality_gap"] = 25

    # 5. Market Size Score
    n = len(products)
    if n >= 10:
        scores["market_size"] = 80
    elif n >= 5:
        scores["market_size"] = 60
    else:
        scores["market_size"] = 40

    # Weighted total
    total = (
        scores["demand"]       * 0.30 +
        scores["competition"]  * 0.30 +
        scores["profitability"]* 0.20 +
        scores["quality_gap"]  * 0.15 +
        scores["market_size"]  * 0.05
    )
    total = round(total)

    if total >= 70:
        verdict = "Excellent Niche"
        verdict_color = "#10B981"
        recommendation = "Strong opportunity — low competition, good demand, profitable price range"
    elif total >= 55:
        verdict = "Good Niche"
        verdict_color = "#3B82F6"
        recommendation = "Decent opportunity — worth pursuing with right differentiation"
    elif total >= 40:
        verdict = "Moderate Niche"
        verdict_color = "#F59E0B"
        recommendation = "Average opportunity — possible but competitive, needs strong USP"
    else:
        verdict = "Tough Niche"
        verdict_color = "#EF4444"
        recommendation = "Difficult market — high competition or low margins"

    # Review gaps — products with < 100 reviews in a market with demand
    low_review_products = [p for p in products if (p.get("reviews") or 0) < 100 and (p.get("bsr") or 999999) < 50000]

    return {
        "keyword": keyword,
        "niche_score": total,
        "verdict": verdict,
        "verdict_color": verdict_color,
        "recommendation": recommendation,
        "scores": {
            "demand":        {"score": scores["demand"],        "label": "Market Demand",     "weight": "30%"},
            "competition":   {"score": scores["competition"],   "label": "Competition Level", "weight": "30%"},
            "profitability": {"score": scores["profitability"], "label": "Profitability",     "weight": "20%"},
            "quality_gap":   {"score": scores["quality_gap"],  "label": "Quality Gap",       "weight": "15%"},
            "market_size":   {"score": scores["market_size"],  "label": "Market Size",       "weight": "5%"},
        },
        "market_stats": {
            "products_analyzed": len(products),
            "avg_price": round(avg_price, 2),
            "avg_bsr": round(avg_bsr),
            "avg_reviews": round(avg_reviews),
            "avg_rating": round(avg_rating, 1),
            "median_reviews": round(med_reviews),
        },
        "opportunities": {
            "low_review_opportunities": len(low_review_products),
            "easy_entry_products": [p.get("asin", "") for p in low_review_products[:3]],
            "price_gap": round(max(prices) - min(prices), 2) if prices else 0,
        }
    }