# -----------------------------
# MCACommentAnalyzerLight.py
# -----------------------------
import pandas as pd
from transformers import pipeline
from wordcloud import WordCloud
import matplotlib.pyplot as plt
from collections import Counter
import nltk
from nltk.corpus import stopwords
import random
from datetime import datetime, timedelta
from langdetect import detect
from deep_translator import GoogleTranslator

# Download stopwords once
nltk.download('stopwords')

class MCACommentAnalyzerLight:
    def __init__(self):
        # Lightweight sentiment model
        self.sentiment_model = pipeline(
            "sentiment-analysis",
            model="cardiffnlp/twitter-roberta-base-sentiment"
        )
        # Lightweight summarizer
        self.summarizer = pipeline(
            "summarization",
            model="t5-small"
        )
        self.stop_words = set(stopwords.words('english'))

    # -----------------------------
    # Translate to English if needed
    # -----------------------------
    def translate_to_english(self, text):
        try:
            lang = detect(text)
            if lang != "en":
                return GoogleTranslator(source='auto', target='en').translate(text)
            return text
        except:
            return text

    # -----------------------------
    # Rule-based sentiment mapping
    # -----------------------------
    def map_sentiment(self, pred, text):
        text_lower = text.lower()
        violation_keywords = ["violation", "violates", "illegal", "non-compliant", "breach", "unlawful", "risk", "penalty"]
        if any(w in text_lower for w in violation_keywords):
            return "Violation"

        suggestion_keywords = ["should", "recommend", "suggest", "advise", "better if", "could", "need to"]
        if any(w in text_lower for w in suggestion_keywords):
            return "Suggestion"

        positive_keywords = ["clear", "helpful", "good", "appreciate", "support"]
        if any(w in text_lower for w in positive_keywords):
            return "Positive"

        negative_keywords = ["confusing", "unclear", "bad", "problem", "needs clarification"]
        if any(w in text_lower for w in negative_keywords):
            return "Negative"

        label = pred['label'].upper()
        if label == "POSITIVE":
            return "Positive"
        elif label == "NEGATIVE":
            return "Negative"
        else:
            return "Neutral"

    # -----------------------------
    # Process single comment
    # -----------------------------
    def process_comment(self, comment):
        translated_comment = self.translate_to_english(comment)
        pred = self.sentiment_model(translated_comment)[0]
        sentiment = self.map_sentiment(pred, translated_comment)

        # Summary: truncate short comments or use summarizer
        if len(translated_comment.split()) < 10:
            summary_text = " ".join(translated_comment.split()[:10])
        else:
            try:
                summary_text = self.summarizer(
                    translated_comment, 
                    max_length=30, 
                    min_length=5, 
                    do_sample=False
                )[0]['summary_text']
            except:
                summary_text = translated_comment

        # Keywords
        words = [w for w in translated_comment.lower().split() if w.isalpha() and w not in self.stop_words]
        keywords = list(Counter(words).keys())
        top_keywords = ", ".join(keywords[:3])

        return sentiment, summary_text, keywords, top_keywords

    # -----------------------------
    # Process multiple comments
    # -----------------------------
    def process_comments(self, comments_list):
        sentiments, summaries, all_keywords, top_keywords_list, timestamps = [], [], [], [], []
        start_date = datetime.now() - timedelta(days=30)

        for comment in comments_list:
            sentiment, summary, keywords, top_kw = self.process_comment(comment)
            sentiments.append(sentiment)
            summaries.append(summary)
            all_keywords.extend(keywords)
            top_keywords_list.append(top_kw)
            timestamps.append(start_date + timedelta(days=random.randint(0, 30)))

        df = pd.DataFrame({
            "Timestamp": timestamps,
            "Comment": comments_list,
            "Summary": summaries,
            "Sentiment": sentiments,
            "Top Keywords": top_keywords_list
        })

        # Sort by Timestamp
        df.sort_values(by='Timestamp', inplace=True, ascending=True)

        # Keyword frequency table
        keyword_freq = pd.DataFrame(
            Counter(all_keywords).items(),
            columns=['Keyword', 'Frequency']
        ).sort_values(by='Frequency', ascending=False)

        return df, keyword_freq

    # -----------------------------
    # Generate WordCloud
    # -----------------------------
    def generate_wordcloud(self, keyword_freq, filename=None):
        wc_dict = dict(zip(keyword_freq['Keyword'], keyword_freq['Frequency']))
        wc = WordCloud(width=800, height=400, background_color="white").generate_from_frequencies(wc_dict)
        plt.figure(figsize=(10,5))
        plt.imshow(wc, interpolation="bilinear")
        plt.axis("off")
        if filename:
            plt.savefig(filename, bbox_inches='tight')
        return plt

# -----------------------------
# Quick Test (Optional)
# -----------------------------
if __name__ == "__main__":
    comments = [
        "The draft is very clear and helpful for companies.",
        "Section 5 is confusing and needs clarification.",
        "It would be better if SMEs get some relief.",
        "I recommend including more examples for clarity.",
        "Section 12 violates the Companies Act rules.",
        "यह टिप्पणी हिंदी में है।",  # Hindi comment example
        "இந்த கருத்து தமிழில் உள்ளது."   # Tamil comment example
    ]

    analyzer = MCACommentAnalyzerLight()
    df, keyword_freq = analyzer.process_comments(comments)
    print(df)
    analyzer.generate_wordcloud(keyword_freq)