import logging
import os
import tempfile
from datetime import datetime

import gradio as gr
import pandas as pd
import plotly.express as px
import torch
import yfinance as yf
from GoogleNews import GoogleNews
from transformers import pipeline

logging.basicConfig(
    level=logging.INFO,
    format="%(asctime)s - %(levelname)s - %(message)s"
)

SENTIMENT_ANALYSIS_MODEL = (
    "mrm8488/distilroberta-finetuned-financial-news-sentiment-analysis"
)

DEVICE = 0 if torch.cuda.is_available() else -1
logging.info(f"Using device: {'cuda' if DEVICE == 0 else 'cpu'}")

sentiment_analyzer = pipeline(
    "sentiment-analysis",
    model=SENTIMENT_ANALYSIS_MODEL,
    device=DEVICE
)


def fetch_articles(query, max_articles=20):
    try:
        googlenews = GoogleNews(lang="en")
        googlenews.search(query)
        articles = googlenews.result()

        if not articles:
            return []

        df = pd.DataFrame(articles)

        required_cols = ["title", "desc", "link", "date", "media"]
        for col in required_cols:
            if col not in df.columns:
                df[col] = ""

        df = df.drop_duplicates(subset=["title", "link"])
        df = df.head(max_articles)

        return df.to_dict("records")

    except Exception as e:
        logging.error(f"News fetch failed for {query}: {e}")
        raise gr.Error("Unable to fetch news. Try again later.")


def analyze_sentiments_batch(articles):
    if not articles:
        return []

    texts = [
        f"{article.get('title', '')}. {article.get('desc', '')}"
        for article in articles
    ]

    sentiments = sentiment_analyzer(
        texts,
        batch_size=8,
        truncation=True,
        max_length=512
    )

    for article, sentiment in zip(articles, sentiments):
        article["sentiment_label"] = sentiment["label"].lower()
        article["sentiment_score"] = round(float(sentiment["score"]), 4)

    return articles


def get_price_context(asset):
    try:
        ticker = yf.Ticker(asset)
        hist = ticker.history(period="1mo")

        if hist.empty:
            return "No price data found. Try using ticker symbols like AAPL, TSLA, BTC-USD, ETH-USD."

        latest_price = hist["Close"].iloc[-1]
        prev_price = hist["Close"].iloc[-2] if len(hist) > 1 else latest_price
        first_price = hist["Close"].iloc[0]

        one_day_return = ((latest_price - prev_price) / prev_price) * 100
        one_month_return = ((latest_price - first_price) / first_price) * 100
        volatility = hist["Close"].pct_change().std() * 100

        return (
            f"### Price Context\n"
            f"- Latest close: **{latest_price:.2f}**\n"
            f"- 1D return: **{one_day_return:.2f}%**\n"
            f"- 1M return: **{one_month_return:.2f}%**\n"
            f"- 1M daily volatility: **{volatility:.2f}%**"
        )

    except Exception as e:
        logging.error(f"Price fetch failed for {asset}: {e}")
        return "Price data unavailable for this asset."


def sentiment_badge(label):
    colors = {
        "negative": "#dc2626",
        "neutral": "#6b7280",
        "positive": "#16a34a",
    }

    color = colors.get(label.lower(), "#6b7280")

    return (
        f'<span style="background-color:{color}; color:white; '
        f'padding:4px 8px; border-radius:6px; font-weight:600;">'
        f'{label.upper()}</span>'
    )


def build_article_table(articles):
    if not articles:
        return pd.DataFrame()

    df = pd.DataFrame(articles)

    df["Sentiment"] = df["sentiment_label"].apply(sentiment_badge)
    df["Confidence"] = df["sentiment_score"]
    df["Title"] = df.apply(
        lambda row: f'<a href="{row["link"]}" target="_blank">{row["title"]}</a>',
        axis=1
    )
    df["Description"] = df["desc"]
    df["Source"] = df["media"]
    df["Date"] = df["date"]

    return df[
        [
            "Sentiment",
            "Confidence",
            "Title",
            "Description",
            "Source",
            "Date",
        ]
    ]


def create_sentiment_summary(articles):
    if not articles:
        return "No articles found."

    df = pd.DataFrame(articles)

    counts = df["sentiment_label"].value_counts(normalize=True) * 100
    avg_confidence = df["sentiment_score"].mean()

    positive = counts.get("positive", 0)
    negative = counts.get("negative", 0)
    neutral = counts.get("neutral", 0)

    bullish_score = positive - negative

    if bullish_score > 20:
        verdict = "Bullish"
    elif bullish_score < -20:
        verdict = "Bearish"
    else:
        verdict = "Neutral / Mixed"

    return (
        f"### Sentiment Summary\n"
        f"- Overall view: **{verdict}**\n"
        f"- Positive: **{positive:.1f}%**\n"
        f"- Neutral: **{neutral:.1f}%**\n"
        f"- Negative: **{negative:.1f}%**\n"
        f"- Average model confidence: **{avg_confidence:.2f}**\n"
        f"- Bullish score: **{bullish_score:.1f}**"
    )


def generate_llm_style_summary(asset, articles):
    if not articles:
        return "No summary available."

    df = pd.DataFrame(articles)

    top_sentiment = df["sentiment_label"].value_counts().idxmax()
    top_articles = df.head(5)

    headlines = "\n".join(
        [f"- {row['title']}" for _, row in top_articles.iterrows()]
    )

    return (
        f"### AI-Style Market Brief\n"
        f"The current news sentiment for **{asset}** appears mostly "
        f"**{top_sentiment.upper()}** based on recent headlines.\n\n"
        f"Key headlines influencing this reading:\n"
        f"{headlines}\n\n"
        f"This should not be treated as a trading signal by itself. "
        f"Use it alongside price action, volume, volatility, macro events, "
        f"and risk management."
    )


def create_sentiment_chart(articles):
    if not articles:
        return None

    df = pd.DataFrame(articles)
    sentiment_counts = df["sentiment_label"].value_counts().reset_index()
    sentiment_counts.columns = ["Sentiment", "Count"]

    fig = px.bar(
        sentiment_counts,
        x="Sentiment",
        y="Count",
        title="Sentiment Distribution"
    )

    return fig


def create_source_chart(articles):
    if not articles:
        return None

    df = pd.DataFrame(articles)

    if "media" not in df.columns:
        return None

    source_counts = df["media"].replace("", "Unknown").value_counts().head(10)
    source_df = source_counts.reset_index()
    source_df.columns = ["Source", "Count"]

    fig = px.bar(
        source_df,
        x="Source",
        y="Count",
        title="Top News Sources"
    )

    return fig


def export_csv(articles):
    if not articles:
        return None

    df = pd.DataFrame(articles)

    filename = f"sentiment_results_{datetime.now().strftime('%Y%m%d_%H%M%S')}.csv"
    filepath = os.path.join(tempfile.gettempdir(), filename)

    df.to_csv(filepath, index=False)
    return filepath


def analyze_single_asset(asset_name, max_articles):
    if not asset_name.strip():
        raise gr.Error("Please enter an asset name or ticker.")

    articles = fetch_articles(asset_name, max_articles)
    analyzed_articles = analyze_sentiments_batch(articles)

    article_table = build_article_table(analyzed_articles)
    sentiment_summary = create_sentiment_summary(analyzed_articles)
    price_context = get_price_context(asset_name)
    ai_summary = generate_llm_style_summary(asset_name, analyzed_articles)

    sentiment_chart = create_sentiment_chart(analyzed_articles)
    source_chart = create_source_chart(analyzed_articles)

    csv_file = export_csv(analyzed_articles)

    return (
        sentiment_summary,
        price_context,
        ai_summary,
        article_table,
        sentiment_chart,
        source_chart,
        csv_file,
    )


def analyze_multiple_assets(asset_list, max_articles):
    assets = [
        asset.strip()
        for asset in asset_list.split(",")
        if asset.strip()
    ]

    if not assets:
        raise gr.Error("Please enter at least one asset.")

    rows = []

    for asset in assets:
        articles = fetch_articles(asset, max_articles)
        analyzed_articles = analyze_sentiments_batch(articles)

        if not analyzed_articles:
            rows.append(
                {
                    "Asset": asset,
                    "Overall Sentiment": "No data",
                    "Positive %": 0,
                    "Neutral %": 0,
                    "Negative %": 0,
                    "Bullish Score": 0,
                    "Articles": 0,
                }
            )
            continue

        df = pd.DataFrame(analyzed_articles)
        counts = df["sentiment_label"].value_counts(normalize=True) * 100

        positive = counts.get("positive", 0)
        neutral = counts.get("neutral", 0)
        negative = counts.get("negative", 0)
        bullish_score = positive - negative

        if bullish_score > 20:
            overall = "Bullish"
        elif bullish_score < -20:
            overall = "Bearish"
        else:
            overall = "Neutral / Mixed"

        rows.append(
            {
                "Asset": asset,
                "Overall Sentiment": overall,
                "Positive %": round(positive, 2),
                "Neutral %": round(neutral, 2),
                "Negative %": round(negative, 2),
                "Bullish Score": round(bullish_score, 2),
                "Articles": len(analyzed_articles),
            }
        )

    result_df = pd.DataFrame(rows)

    fig = px.bar(
        result_df,
        x="Asset",
        y="Bullish Score",
        title="Bullish Score Comparison"
    )

    return result_df, fig


with gr.Blocks(title="FinSentinel") as app:
    gr.Markdown("# FinSentinel")
    gr.Markdown(
        "Financial news sentiment dashboard for stocks, crypto, ETFs, commodities, and market themes."
    )

    with gr.Tabs():
        with gr.Tab("Single Asset Analysis"):
            with gr.Row():
                asset_input = gr.Textbox(
                    label="Asset Name or Ticker",
                    placeholder="Example: AAPL, TSLA, Bitcoin, BTC-USD, Nvidia",
                )

                max_articles = gr.Slider(
                    minimum=5,
                    maximum=50,
                    value=20,
                    step=5,
                    label="Number of Articles",
                )

            analyze_button = gr.Button("Analyze Asset", variant="primary")

            gr.Examples(
                examples=[
                    "AAPL",
                    "TSLA",
                    "NVDA",
                    "BTC-USD",
                    "Bitcoin",
                    "Gold",
                ],
                inputs=asset_input,
            )

            with gr.Row():
                sentiment_summary_output = gr.Markdown()
                price_context_output = gr.Markdown()

            ai_summary_output = gr.Markdown()

            with gr.Row():
                sentiment_chart_output = gr.Plot()
                source_chart_output = gr.Plot()

            articles_output = gr.Dataframe(
                label="Articles and Sentiment Analysis",
                headers=[
                    "Sentiment",
                    "Confidence",
                    "Title",
                    "Description",
                    "Source",
                    "Date",
                ],
                datatype=[
                    "html",
                    "number",
                    "html",
                    "str",
                    "str",
                    "str",
                ],
                wrap=True,
                interactive=False,
            )

            csv_output = gr.File(label="Download CSV")

            analyze_button.click(
                analyze_single_asset,
                inputs=[asset_input, max_articles],
                outputs=[
                    sentiment_summary_output,
                    price_context_output,
                    ai_summary_output,
                    articles_output,
                    sentiment_chart_output,
                    source_chart_output,
                    csv_output,
                ],
            )

        with gr.Tab("Portfolio Comparison"):
            portfolio_input = gr.Textbox(
                label="Assets",
                placeholder="Example: AAPL, TSLA, NVDA, BTC-USD, ETH-USD",
                lines=2,
            )

            portfolio_articles = gr.Slider(
                minimum=5,
                maximum=30,
                value=10,
                step=5,
                label="Articles Per Asset",
            )

            portfolio_button = gr.Button("Compare Assets", variant="primary")

            portfolio_table = gr.Dataframe(
                label="Portfolio Sentiment Comparison",
                interactive=False,
            )

            portfolio_chart = gr.Plot()

            portfolio_button.click(
                analyze_multiple_assets,
                inputs=[portfolio_input, portfolio_articles],
                outputs=[portfolio_table, portfolio_chart],
            )

if __name__ == "__main__":
    app.queue().launch(server_name="0.0.0.0", server_port=7860)