app.py

import gradio as gr
from transformers import pipeline
from langdetect import detect
from huggingface_hub import InferenceClient
import pandas as pd
import os
import nltk
import asyncio
nltk.download('punkt')  # CORRECT : 'punkt' !

from nltk.tokenize import sent_tokenize

HF_TOKEN = os.getenv("HF_TOKEN")

# Chargement modèles
translator_to_en = pipeline("translation", model="Helsinki-NLP/opus-mt-mul-en")
translator_to_fr = pipeline("translation", model="Helsinki-NLP/opus-mt-en-fr")
classifier = pipeline("sentiment-analysis", model="mrm8488/distilroberta-finetuned-financial-news-sentiment-analysis")

# Fonction traduction segmentée
def safe_translate_to_fr(text, max_length=512):
    sentences = sent_tokenize(text)
    translated_sentences = []
    for sentence in sentences:
        translated = translator_to_fr(sentence, max_length=max_length)[0]['translation_text']
        translated_sentences.append(translated)
    return " ".join(translated_sentences)

# Appel API Zephyr
async def call_zephyr_api(prompt, mode, hf_token=HF_TOKEN):
    client = InferenceClient("HuggingFaceH4/zephyr-7b-beta", token=hf_token)
    try:
        if mode == "Rapide":
            max_new_tokens = 50
            temperature = 0.3
        elif mode == "Équilibré":
            max_new_tokens = 100
            temperature = 0.5
        else:
            max_new_tokens = 150
            temperature = 0.7
        response = await asyncio.to_thread(client.text_generation, prompt, max_new_tokens=max_new_tokens, temperature=temperature)
        return response
    except Exception as e:
        raise gr.Error(f"❌ Erreur API Hugging Face : {str(e)}")

# Suggestion mode
def suggest_model(text):
    word_count = len(text.split())
    if word_count < 50:
        return "Rapide"
    elif word_count <= 200:
        return "Équilibré"
    else:
        return "Précis"

# Jauge sentiment
def create_sentiment_gauge(sentiment, score):
    score_percentage = score * 100
    color = "#A9A9A9"
    if sentiment.lower() == "positive":
        color = "#2E8B57"
    elif sentiment.lower() == "negative":
        color = "#DC143C"

    html = f"""
    <div style='width: 100%; max-width: 300px; margin: 10px 0;'>
        <div style='background-color: #D3D3D3; border-radius: 5px; height: 20px; position: relative;'>
            <div style='background-color: {color}; width: {score_percentage}%; height: 100%; border-radius: 5px;'></div>
            <span style='position: absolute; top: 0; left: 50%; transform: translateX(-50%); font-weight: bold;'>{score_percentage:.1f}%</span>
        </div>
        <div style='text-align: center; margin-top: 5px;'>Sentiment : {sentiment}</div>
    </div>
    """
    return html

# Analyse principale
async def full_analysis(text, mode, detail_mode, count, history):
    if not text:
        yield "Entrez une phrase.", "", "", "", 0, history, "", "Aucune analyse effectuée."
        return

    yield "Analyse en cours... (Étape 1 : Détection de la langue)", "", "", "", count, history, "", "Détection de la langue"

    try:
        lang = detect(text)
    except:
        lang = "unknown"

    if lang != "en":
        text_en = translator_to_en(text, max_length=512)[0]['translation_text']
    else:
        text_en = text

    yield "Analyse en cours... (Étape 2 : Analyse du sentiment)", "", "", "", count, history, "", "Analyse du sentiment"

    result = await asyncio.to_thread(classifier, text_en)
    result = result[0]
    sentiment_output = f"Sentiment prédictif : {result['label']} (Score: {result['score']:.2f})"
    sentiment_gauge = create_sentiment_gauge(result['label'], result['score'])

    yield "Analyse en cours... (Étape 3 : Explication IA)", "", "", "", count, history, "", "Génération de l'explication"

    explanation_prompt = f"""<|system|>
You are a professional financial analyst AI with expertise in economic forecasting.
</s>
<|user|>
Given the following question about a potential economic event: "{text}"

The predicted sentiment for this event is: {result['label'].lower()}.

Assume the event happens. Explain why this event would likely have a {result['label'].lower()} economic impact.
</s>
<|assistant|>"""
    explanation_en = await call_zephyr_api(explanation_prompt, mode)

    yield "Analyse en cours... (Étape 4 : Traduction en français)", "", "", "", count, history, "", "Traduction en français"

    explanation_fr = safe_translate_to_fr(explanation_en)

    count += 1
    history.append({
        "Texte": text,
        "Sentiment": result['label'],
        "Score": f"{result['score']:.2f}",
        "Explication_EN": explanation_en,
        "Explication_FR": explanation_fr
    })

    yield sentiment_output, text, explanation_en, explanation_fr, count, history, sentiment_gauge, "✅ Analyse terminée."

# Historique CSV
def download_history(history):
    if not history:
        return None
    df = pd.DataFrame(history)
    file_path = "/tmp/analysis_history.csv"
    df.to_csv(file_path, index=False)
    return file_path

# Lancement Gradio
def launch_app():
    custom_css = """
    /* Ici tout ton CSS personnalisé pour style de fond, couleurs, boutons */
    body {
        background: linear-gradient(135deg, #0A1D37 0%, #1A3C34 100%);
        font-family: 'Inter', sans-serif;
        color: #E0E0E0;
        padding: 20px;
    }
    .gr-box {
        background: #2A4A43 !important;
        border: 1px solid #FFD700 !important;
        border-radius: 12px !important;
        padding: 20px !important;
        box-shadow: 0px 4px 12px rgba(255, 215, 0, 0.4);
    }
    .gr-button {
        background: linear-gradient(90deg, #FFD700, #D4AF37);
        color: #0A1D37;
        font-weight: bold;
        border: none;
        border-radius: 8px;
        padding: 12px 24px;
        transition: transform 0.2s;
    }
    .gr-button:hover {
        transform: translateY(-2px);
        box-shadow: 0 6px 12px rgba(255, 215, 0, 0.5);
    }
    """

    with gr.Blocks(theme=gr.themes.Base(), css=custom_css) as iface:
        gr.Markdown("# 📈 Analyse Financière Premium avec IA")
        gr.Markdown("**Posez une question économique.** L'IA analyse et explique l'impact.")

        count = gr.State(0)
        history = gr.State([])

        with gr.Row():
            with gr.Column(scale=2):
                input_text = gr.Textbox(lines=4, label="Votre question économique")
            with gr.Column(scale=1):
                mode_selector = gr.Dropdown(choices=["Rapide", "Équilibré", "Précis"], value="Équilibré", label="Mode de réponse")
                detail_mode_selector = gr.Dropdown(choices=["Normal", "Expert"], value="Normal", label="Niveau de détail")

        analyze_btn = gr.Button("Analyser")
        download_btn = gr.Button("Télécharger l'historique")

        with gr.Row():
            sentiment_output = gr.Textbox(label="Sentiment prédictif")
            displayed_prompt = gr.Textbox(label="Texte utilisé (anglais)", interactive=False)
            explanation_output_en = gr.Textbox(label="Explication en anglais")
            explanation_output_fr = gr.Textbox(label="Explication en français")
            sentiment_gauge = gr.HTML()
            progress_message = gr.Textbox(label="Progression", interactive=False)

        download_file = gr.File(label="Fichier CSV")

        input_text.change(lambda t: gr.update(value=suggest_model(t)), inputs=[input_text], outputs=[mode_selector])

        analyze_btn.click(
            full_analysis,
            inputs=[input_text, mode_selector, detail_mode_selector, count, history],
            outputs=[sentiment_output, displayed_prompt, explanation_output_en, explanation_output_fr, count, history, sentiment_gauge, progress_message]
        )

        download_btn.click(
            download_history,
            inputs=[history],
            outputs=[download_file]
        )

    iface.launch(share=True)

if __name__ == "__main__":
    launch_app()