app_epico.py

import gradio as gr
import pandas as pd
import numpy as np
import spacy
from textblob import TextBlob
from transformers import pipeline
from langdetect import detect, DetectorFactory
from functools import lru_cache

DetectorFactory.seed = 0  # deterministic langdetect

APP_TITLE = "🚀 Análisis Épico de Sentimientos (Multimodelo + Lingüística)"

# ==============================
# Carga perezosa (lazy) de modelos
# ==============================

@lru_cache(maxsize=1)
def load_spacy():
    try:
        nlp = spacy.load("es_core_news_sm")
        return nlp, "✅ spaCy (es_core_news_sm)"
    except Exception as e:
        return None, f"❌ spaCy no disponible: {e}"

@lru_cache(maxsize=1)
def load_multilingual_sentiment():
    try:
        clf = pipeline("text-classification", model="tabularisai/multilingual-sentiment-analysis")
        return clf, "✅ Multilingual Sentiment cargado"
    except Exception as e:
        return None, f"❌ Multilingual Sentiment no disponible: {e}"

@lru_cache(maxsize=1)
def load_multilingual_bert():
    try:
        clf = pipeline("sentiment-analysis",
                       model="nlptown/bert-base-multilingual-uncased-sentiment",
                       tokenizer="nlptown/bert-base-multilingual-uncased-sentiment")
        return clf, "✅ BERT Multilingual (estrellas) cargado"
    except Exception as e:
        return None, f"❌ BERT Multilingual no disponible: {e}"

# ==============================
# Léxico sencillo ES + resaltado
# ==============================

PAL_POS = {
    'bueno','excelente','fantástico','maravilloso','perfecto','genial',
    'increíble','amo','encanta','feliz','contento','satisfecho','agradable',
    'recomiendo','magnífico','extraordinario','asombroso','estupendo',
    'óptimo','superior','inmejorable','ideal','brutal','espectacular'
}

PAL_NEG = {
    'malo','terrible','horrible','pésimo','odio','decepcionado','fatal',
    'triste','enojado','frustrado','pobre','deficiente','desastroso',
    'insatisfecho','decepcionante','horroroso','malísimo','inútil',
    'defectuoso','deplorable','lamentable','desagradable'
}

def lexical_score(text, nlp):
    text_low = text.lower().strip()
    if not nlp:
        # fallback básico sin lematizar
        tokens = [t for t in ''.join([c if c.isalpha() or c.isspace() else ' ' for c in text_low]).split() if len(t)>2]
        lemmas = tokens
    else:
        doc = nlp(text_low)
        lemmas = [t.lemma_ for t in doc if t.is_alpha and len(t) > 2]

    pos = sum(1 for w in lemmas if w in PAL_POS)
    neg = sum(1 for w in lemmas if w in PAL_NEG)
    total = max(1, len(lemmas))
    raw = (pos - neg) / total
    norm = max(-1.0, min(1.0, raw * 5))
    return {"positivas": pos, "negativas": neg, "total": total, "normalized_score": norm, "lemmas": lemmas}

def highlight_words(text, nlp):
    # Resalta palabras del léxico en el texto original
    if not text:
        return ""
    original = text
    if nlp:
        doc = nlp(original)
        tokens = [t.text for t in doc]
    else:
        tokens = original.split()

    def wrap(tok):
        low = tok.lower()
        if low in PAL_POS:
            return f"<mark style='background:#D1FAE5; padding:2px 4px; border-radius:4px'>+{tok}</mark>"
        if low in PAL_NEG:
            return f"<mark style='background:#FEE2E2; padding:2px 4px; border-radius:4px'>-{tok}</mark>"
        return tok

    return " ".join(wrap(t) for t in tokens)

# ==============================
# Sentimiento por modelos
# ==============================

STAR_MAP = {'1 star': -1.0, '2 stars': -0.5, '3 stars': 0.0, '4 stars': 0.5, '5 stars': 1.0}

def model_scores(text):
    out = {}
    clf1, status1 = load_multilingual_sentiment()
    clf2, status2 = load_multilingual_bert()
    nlp, _ = load_spacy()

    # Multilingual Sentiment
    if clf1:
        try:
            r = clf1(text)[0]
            out['multilingual'] = {
                "label": r['label'], "score": float(r['score']),
                "normalized_score": float(r['score']) if r['label']=='POSITIVE' else -float(r['score'])
            }
        except Exception as e:
            out['multilingual'] = {"error": str(e)}
    else:
        out['multilingual'] = {"error": status1}

    # BERT estrellas
    if clf2:
        try:
            r = clf2(text)[0]
            out['bert'] = {
                "label": r['label'], "score": float(r.get('score', 0.0)),
                "normalized_score": float(STAR_MAP.get(r['label'], 0.0))
            }
        except Exception as e:
            out['bert'] = {"error": str(e)}
    else:
        out['bert'] = {"error": status2}

    # Léxico
    try:
        out['lexico'] = lexical_score(text, nlp)
    except Exception as e:
        out['lexico'] = {"error": str(e)}

    # TextBlob
    try:
        blob = TextBlob(text)
        out['textblob'] = {
            "polarity": float(blob.sentiment.polarity),
            "subjectivity": float(blob.sentiment.subjectivity),
            "normalized_score": float(blob.sentiment.polarity)
        }
    except Exception as e:
        out['textblob'] = {"error": str(e)}

    return out

def fuse_scores(results, w_multi=0.4, w_bert=0.3, w_lex=0.2, w_tb=0.1, thr=0.2):
    scores = []
    if 'normalized_score' in results.get('multilingual', {}):
        scores.append(results['multilingual']['normalized_score'] * w_multi)
    if 'normalized_score' in results.get('bert', {}):
        scores.append(results['bert']['normalized_score'] * w_bert)
    if 'normalized_score' in results.get('lexico', {}):
        scores.append(results['lexico']['normalized_score'] * w_lex)
    if 'normalized_score' in results.get('textblob', {}):
        scores.append(results['textblob']['normalized_score'] * w_tb)

    if not scores:
        return "❓ INDETERMINADO", 0.0, "#FB923C"

    s = float(np.sum(scores))

    if s > thr:
        return "😊 POSITIVO", s, "#10B981"
    elif s < -thr:
        return "😠 NEGATIVO", s, "#EF4444"
    else:
        return "😐 NEUTRO", s, "#6B7280"

def detect_lang(text):
    try:
        return detect(text)
    except Exception:
        return "unknown"

# ==============================
# Análisis de texto (UI)
# ==============================

def analyze_text(text, w_multi, w_bert, w_lex, w_tb, thr):
    text = (text or "").strip()
    if not text:
        return "❌ Ingresa un texto", "", "", ""

    lang = detect_lang(text)
    models = model_scores(text)
    label, final, color = fuse_scores(models, w_multi, w_bert, w_lex, w_tb, thr)

    nlp, _ = load_spacy()

    header = f"""
    <div style='background:{color}22; border-left:6px solid {color}; padding:16px; border-radius:10px'>
      <div style='display:flex; justify-content:space-between; align-items:center'>
        <h2 style='margin:0; color:{color}'>{label}</h2>
        <code style='opacity:0.8'>Idioma detectado: {lang}</code>
      </div>
      <p style='margin:4px 0'><b>Puntuación combinada:</b> {final:.3f}</p>
      <p style='margin:4px 0'><b>Longitud:</b> {len(text)} caracteres</p>
    </div>
    """

    # Detalles por modelo
    def block(name, d):
        if 'error' in d:
            return f"<div><b>{name}</b><br><span style='color:#EF4444'>Error: {d['error']}</span></div>"
        rows = []
        for k,v in d.items():
            if isinstance(v, float):
                rows.append(f"{k}: {v:.3f}")
            else:
                rows.append(f"{k}: {v}")
        return f"<div style='padding:8px; border:1px solid #e5e7eb; border-radius:8px'><b>{name}</b><br>" + "<br>".join(rows) + "</div>"

    details = "<h3>📊 Resultados por método</h3>" +               "<div style='display:grid; gap:10px; grid-template-columns: repeat(auto-fit,minmax(240px,1fr))'>" +               block("Multilingual", models.get('multilingual', {})) +               block("BERT (estrellas)", models.get('bert', {})) +               block("Léxico (ES)", models.get('lexico', {})) +               block("TextBlob", models.get('textblob', {})) +               "</div>"

    # Resaltado léxico
    highlighted = highlight_words(text, nlp)
    highlight_html = f"""
    <h3>🔎 Palabras clave detectadas</h3>
    <div style='padding:12px; border:1px dashed #d1d5db; border-radius:10px'>{highlighted}</div>
    """

    # Lingüística resumida
    if nlp:
        doc = nlp(text)
        ents = "<br>".join([f"• {e.text} ({e.label_})" for e in list(doc.ents)[:8]]) or "—"
        ling = f"""
        <h3>📝 Análisis lingüístico (spaCy)</h3>
        <ul>
          <li>Tokens: {len(doc)}</li>
          <li>Palabras: {len([t for t in doc if t.is_alpha])}</li>
          <li>Oraciones: {len(list(doc.sents))}</li>
          <li>Entidades: {len(doc.ents)}</li>
        </ul>
        <p><b>Entidades detectadas:</b><br>{ents}</p>
        """
    else:
        ling = "<p style='color:#EF4444'>spaCy no disponible (modelo es_core_news_sm no instalado)</p>"

    return header, details, highlight_html, ling

# ==============================
# Excel/CSV
# ==============================

def analyze_file(file, max_rows, text_cols_manual, w_multi, w_bert, w_lex, w_tb, thr):
    if file is None:
        return pd.DataFrame([{"Resultado":"❌ Sube un archivo .xlsx o .csv"}])

    name = getattr(file, "name", "archivo")
    try:
        if name.lower().endswith(".csv"):
            df = pd.read_csv(file)
        else:
            df = pd.read_excel(file)
    except Exception as e:
        return pd.DataFrame([{"Error": f"❌ No pude leer el archivo: {e}"}])

    # Detectar columnas de texto si no se especifican
    if text_cols_manual:
        cols = [c.strip() for c in text_cols_manual.split(",") if c.strip() in df.columns]
    else:
        cols = []
        for c in df.columns:
            if df[c].dtype == "object":
                sample = df[c].dropna().astype(str).head(5).tolist()
                if any(len(s.split()) >= 5 for s in sample):
                    cols.append(c)
        cols = cols[:2]  # máximo 2 columnas por defecto

    if not cols:
        return pd.DataFrame([{"Resultado":"❌ No encontré columnas de texto (o especifica manualmente)"}])

    records = []
    for c in cols:
        for i, text in enumerate(df[c].dropna().astype(str).head(max_rows), start=1):
            models = model_scores(text)
            label, s, _ = fuse_scores(models, w_multi, w_bert, w_lex, w_tb, thr)
            records.append({
                "Columna": c,
                "Fila": i,
                "Texto": (text[:140] + "...") if len(text) > 140 else text,
                "Sentimiento": label.replace("😊 ","").replace("😠 ","").replace("😐 ",""),
                "Score": round(s,3),
                "Len": len(text)
            })

    return pd.DataFrame.from_records(records)

# ==============================
# UI
# ==============================

with gr.Blocks(theme="soft", title=APP_TITLE, css="""
#component-0 .hover\:bg-red-500:hover{ background: none }
.markdown-body h1, .markdown-body h2 { margin-top:0 }
""") as demo:
    gr.Markdown(f"""
    # {APP_TITLE}
    **Combina múltiples modelos, léxico y análisis lingüístico. Ajusta pesos y genera insights épicos.**
    """)

    with gr.Tab("📝 Texto individual"):
        with gr.Row():
            with gr.Column(scale=5):
                text_in = gr.Textbox(label="Texto", lines=6, placeholder="Escribe aquí en ES/EN/FR/PT...")
                with gr.Accordion("⚙️ Pesos y umbral", open=False):
                    w_multi = gr.Slider(0,1,value=0.4,step=0.05,label="Peso Multilingual")
                    w_bert  = gr.Slider(0,1,value=0.3,step=0.05,label="Peso BERT")
                    w_lex   = gr.Slider(0,1,value=0.2,step=0.05,label="Peso Léxico")
                    w_tb    = gr.Slider(0,1,value=0.1,step=0.05,label="Peso TextBlob")
                    thr     = gr.Slider(0,1,value=0.2,step=0.01,label="Umbral de neutro (|score| ≤ umbral)")
                btn = gr.Button("🔍 Analizar", variant="primary")
                gr.Examples(
                    examples=[
                        ["Me encanta este producto, superó mis expectativas y lo recomiendo."],
                        ["Pésimo servicio, llegó tarde y defectuoso. Muy decepcionado."],
                        ["El producto cumple, pero no destaca. Está bien por el precio."],
                        ["I absolutely love it! Great quality and fast delivery."],
                        ["C'est un service horrible, je ne le recommande à personne."],
                        ["O atendimento foi excelente e o produto é ótimo."]
                    ],
                    inputs=[text_in]
                )
            with gr.Column(scale=5):
                head = gr.HTML(label="🎯 Resultado")
                methods = gr.HTML(label="📊 Detalles por modelo")
                highlights = gr.HTML(label="🔎 Palabras clave")
                ling = gr.HTML(label="📝 Lingüística")

        btn.click(analyze_text, [text_in, w_multi, w_bert, w_lex, w_tb, thr], [head, methods, highlights, ling])
        text_in.submit(analyze_text, [text_in, w_multi, w_bert, w_lex, w_tb, thr], [head, methods, highlights, ling])

    with gr.Tab("📈 Lote (Excel/CSV)"):
        with gr.Row():
            with gr.Column(scale=5):
                f = gr.File(label="Sube .xlsx o .csv")
                max_rows = gr.Slider(5, 500, value=100, step=5, label="Filas máximas por columna")
                text_cols_manual = gr.Textbox(label="Columnas de texto (opcional, separadas por coma)")
                with gr.Accordion("⚙️ Pesos y umbral", open=False):
                    w_multi2 = gr.Slider(0,1,value=0.4,step=0.05,label="Peso Multilingual")
                    w_bert2  = gr.Slider(0,1,value=0.3,step=0.05,label="Peso BERT")
                    w_lex2   = gr.Slider(0,1,value=0.2,step=0.05,label="Peso Léxico")
                    w_tb2    = gr.Slider(0,1,value=0.1,step=0.05,label="Peso TextBlob")
                    thr2     = gr.Slider(0,1,value=0.2,step=0.01,label="Umbral de neutro")
                btn2 = gr.Button("🚀 Analizar archivo", variant="primary")
            with gr.Column(scale=5):
                df_out = gr.Dataframe(wrap=True, label="Resultados")
                dl = gr.DownloadButton(label="⬇️ Descargar CSV", value=None)

        def _pipe(file, max_rows, text_cols_manual, w1,w2,w3,w4,thr):
            df = analyze_file(file, int(max_rows), text_cols_manual, w1,w2,w3,w4,thr)
            # generar CSV temporal
            try:
                csv = df.to_csv(index=False).encode("utf-8")
                return df, csv
            except Exception:
                return df, None

        btn2.click(_pipe,
                   [f, max_rows, text_cols_manual, w_multi2, w_bert2, w_lex2, w_tb2, thr2],
                   [df_out, dl])

    with gr.Tab("ℹ️ Sistema & Modelos"):
        spacy_status = load_spacy()[1]
        m1_status = load_multilingual_sentiment()[1]
        m2_status = load_multilingual_bert()[1]
        gr.Markdown(f"""
        ### Estado de modelos
        - {spacy_status}
        - {m1_status}
        - {m2_status}

        ### Cómo mejorar precisión
        - Ajusta pesos según tu dominio (por ejemplo, más peso al léxico para español coloquial).
        - Entrena un diccionario propio con palabras frecuentes de tus clientes.
        - Limpia el texto (remueve spam, URLs, firmas) antes de analizar.
        - Para grandes volúmenes, considera un modelo fine-tuned con tus datos.
        """)

if __name__ == "__main__":
    demo.launch()