app.py

# --- INSTALACIÓN DE DEPENDENCIAS ADICIONALES ---
import os
import sys
import subprocess

def install_packages():
    packages = ["gradio", "plotly", "seaborn", "pandas", "openpyxl", "scikit-learn",
                "fpdf2", "python-docx", "kaleido"]
    for package in packages:
        try:
            __import__(package)
        except ImportError:
            print(f"Instalando {package}...")
            subprocess.check_call([sys.executable, "-m", "pip", "install", package])

install_packages()

# --- IMPORTACIONES ---
import os
import io
import tempfile
import traceback
import zipfile
from typing import List, Tuple, Dict, Any, Optional
from abc import ABC, abstractmethod
from unittest.mock import MagicMock

from PIL import Image
import gradio as gr
import plotly.graph_objects as go
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
from scipy.integrate import odeint
from scipy.optimize import curve_fit
from sklearn.metrics import mean_squared_error
from docx import Document
from docx.shared import Inches
from fpdf import FPDF
from fpdf.enums import XPos, YPos

# --- CONSTANTES ---
C_TIME = 'tiempo'
C_BIOMASS = 'biomass'
C_SUBSTRATE = 'substrate'
C_PRODUCT = 'product'
COMPONENTS = [C_BIOMASS, C_SUBSTRATE, C_PRODUCT]

# --- BLOQUE 1: ESTRUCTURA DE MODELOS CINÉTICOS ESCALABLE ---

class KineticModel(ABC):
    def __init__(self, name: str, display_name: str, param_names: List[str]):
        self.name, self.display_name, self.param_names = name, display_name, param_names
        self.num_params = len(param_names)

    @abstractmethod
    def model_function(self, t: np.ndarray, *params: float) -> np.ndarray: pass
    def diff_function(self, X: float, t: float, params: List[float]) -> float: return 0.0
    @abstractmethod
    def get_initial_params(self, time: np.ndarray, biomass: np.ndarray) -> List[float]: pass
    @abstractmethod
    def get_param_bounds(self, time: np.ndarray, biomass: np.ndarray) -> Tuple[List[float], List[float]]: pass

class LogisticModel(KineticModel):
    def __init__(self): super().__init__("logistic", "Logístico", ["Xo", "Xm", "um"])
    def model_function(self, t: np.ndarray, *params: float) -> np.ndarray:
        xo, xm, um = params
        if xm <= 0 or xo <= 0 or xm < xo: return np.full_like(t, np.nan)
        exp_arg = np.clip(um * t, -700, 700); term_exp = np.exp(exp_arg)
        denominator = 1 - (xo / xm) * (1 - term_exp); denominator = np.where(denominator == 0, 1e-9, denominator)
        return (xo * term_exp) / denominator
    def diff_function(self, X: float, t: float, params: List[float]) -> float:
        _, xm, um = params; return um * X * (1 - X / xm) if xm > 0 else 0.0
    def get_initial_params(self, time: np.ndarray, biomass: np.ndarray) -> List[float]:
        return [biomass[0] if len(biomass) > 0 and biomass[0] > 1e-6 else 1e-3, max(biomass) if len(biomass) > 0 else 1.0, 0.1]
    def get_param_bounds(self, time: np.ndarray, biomass: np.ndarray) -> Tuple[List[float], List[float]]:
        initial_biomass = biomass[0] if len(biomass) > 0 else 1e-9; max_biomass = max(biomass) if len(biomass) > 0 else 1.0
        return ([1e-9, initial_biomass, 1e-9], [max_biomass * 1.2, max_biomass * 5, np.inf])

class GompertzModel(KineticModel):
    def __init__(self): super().__init__("gompertz", "Gompertz", ["Xm", "um", "lag"])
    def model_function(self, t: np.ndarray, *params: float) -> np.ndarray:
        xm, um, lag = params
        if xm <= 0 or um <= 0: return np.full_like(t, np.nan)
        exp_term = (um * np.e / xm) * (lag - t) + 1; exp_term_clipped = np.clip(exp_term, -700, 700)
        return xm * np.exp(-np.exp(exp_term_clipped))
    def diff_function(self, X: float, t: float, params: List[float]) -> float:
        xm, um, lag = params; k_val = um * np.e / xm
        u_val = k_val * (lag - t) + 1; u_val_clipped = np.clip(u_val, -np.inf, 700)
        return X * k_val * np.exp(u_val_clipped) if xm > 0 and X > 0 else 0.0
    def get_initial_params(self, time: np.ndarray, biomass: np.ndarray) -> List[float]:
        return [max(biomass) if len(biomass) > 0 else 1.0, 0.1, time[np.argmax(np.gradient(biomass))] if len(biomass) > 1 else 0]
    def get_param_bounds(self, time: np.ndarray, biomass: np.ndarray) -> Tuple[List[float], List[float]]:
        initial_biomass = min(biomass) if len(biomass) > 0 else 1e-9; max_biomass = max(biomass) if len(biomass) > 0 else 1.0
        return ([max(1e-9, initial_biomass), 1e-9, 0], [max_biomass * 5, np.inf, max(time) if len(time) > 0 else 1])

class MoserModel(KineticModel):
    def __init__(self): super().__init__("moser", "Moser", ["Xm", "um", "Ks"])
    def model_function(self, t: np.ndarray, *params: float) -> np.ndarray:
        Xm, um, Ks = params; return Xm * (1 - np.exp(-um * (t - Ks))) if Xm > 0 and um > 0 else np.full_like(t, np.nan)
    def diff_function(self, X: float, t: float, params: List[float]) -> float:
        Xm, um, _ = params; return um * (Xm - X) if Xm > 0 else 0.0
    def get_initial_params(self, time: np.ndarray, biomass: np.ndarray) -> List[float]:
        return [max(biomass) if len(biomass) > 0 else 1.0, 0.1, 0]
    def get_param_bounds(self, time: np.ndarray, biomass: np.ndarray) -> Tuple[List[float], List[float]]:
        initial_biomass = min(biomass) if len(biomass) > 0 else 1e-9; max_biomass = max(biomass) if len(biomass) > 0 else 1.0
        return ([max(1e-9, initial_biomass), 1e-9, -np.inf], [max_biomass * 5, np.inf, np.inf])

class BaranyiModel(KineticModel):
    def __init__(self): super().__init__("baranyi", "Baranyi", ["X0", "Xm", "um", "lag"])
    def model_function(self, t: np.ndarray, *params: float) -> np.ndarray:
        X0, Xm, um, lag = params
        if X0 <= 0 or Xm <= X0 or um <= 0 or lag < 0: return np.full_like(t, np.nan)
        A_t = t + (1 / um) * np.log(np.exp(-um * t) + np.exp(-um * lag) - np.exp(-um * (t + lag)))
        exp_um_At = np.exp(np.clip(um * A_t, -700, 700))
        numerator = Xm; denominator = 1 + ((Xm / X0) - 1) * (1 / exp_um_At)
        return numerator / np.where(denominator == 0, 1e-9, denominator)
    def get_initial_params(self, time: np.ndarray, biomass: np.ndarray) -> List[float]:
        return [biomass[0] if len(biomass) > 0 and biomass[0] > 1e-6 else 1e-3, max(biomass) if len(biomass) > 0 else 1.0, 0.1, time[np.argmax(np.gradient(biomass))] if len(biomass) > 1 else 0.0]
    def get_param_bounds(self, time: np.ndarray, biomass: np.ndarray) -> Tuple[List[float], List[float]]:
        initial_biomass = biomass[0] if len(biomass) > 0 else 1e-9; max_biomass = max(biomass) if len(biomass) > 0 else 1.0
        return ([1e-9, max(1e-9, initial_biomass), 1e-9, 0], [max_biomass * 1.2, max_biomass * 10, np.inf, max(time) if len(time) > 0 else 1])

# --- REGISTRO CENTRAL DE MODELOS ---
AVAILABLE_MODELS: Dict[str, KineticModel] = {model.name: model for model in [LogisticModel(), GompertzModel(), MoserModel(), BaranyiModel()]}

# --- CLASE DE AJUSTE DE BIOPROCESOS ---
class BioprocessFitter:
    def __init__(self, kinetic_model: KineticModel, maxfev: int = 50000):
        self.model, self.maxfev = kinetic_model, maxfev
        self.params: Dict[str, Dict[str, float]] = {c: {} for c in COMPONENTS}
        self.r2: Dict[str, float] = {}; self.rmse: Dict[str, float] = {}
        self.data_time: Optional[np.ndarray] = None
        self.data_means: Dict[str, Optional[np.ndarray]] = {c: None for c in COMPONENTS}
        self.data_stds: Dict[str, Optional[np.ndarray]] = {c: None for c in COMPONENTS}

    def _get_biomass_at_t(self, t: np.ndarray, p: List[float]) -> np.ndarray: return self.model.model_function(t, *p)
    def _get_initial_biomass(self, p: List[float]) -> float:
        if not p: return 0.0
        if any(k in self.model.param_names for k in ["Xo", "X0"]):
            try:
                idx = self.model.param_names.index("Xo") if "Xo" in self.model.param_names else self.model.param_names.index("X0")
                return p[idx]
            except (ValueError, IndexError): pass
        return float(self.model.model_function(np.array([0]), *p)[0])
    def _calc_integral(self, t: np.ndarray, p: List[float]) -> np.ndarray:
        X_t = self._get_biomass_at_t(t, p)
        if np.any(np.isnan(X_t)): return np.full_like(t, np.nan)
        integral_X = np.zeros_like(X_t)
        if len(t) > 1:
            dt = np.diff(t, prepend=t[0] - (t[1] - t[0] if len(t) > 1 else 1))
            integral_X = np.cumsum(X_t * dt)
        return integral_X, X_t
    def substrate(self, t: np.ndarray, so: float, p_c: float, q: float, bio_p: List[float]) -> np.ndarray:
        integral, X_t = self._calc_integral(t, bio_p); X0 = self._get_initial_biomass(bio_p)
        return so - p_c * (X_t - X0) - q * integral
    def product(self, t: np.ndarray, po: float, alpha: float, beta: float, bio_p: List[float]) -> np.ndarray:
        integral, X_t = self._calc_integral(t, bio_p); X0 = self._get_initial_biomass(bio_p)
        return po + alpha * (X_t - X0) + beta * integral
    def process_data_from_df(self, df: pd.DataFrame) -> None:
        try:
            time_col = [c for c in df.columns if c[1].strip().lower() == C_TIME][0]
            self.data_time = df[time_col].dropna().to_numpy(); min_len = len(self.data_time)
            def extract(name: str) -> Tuple[np.ndarray, np.ndarray]:
                cols = [c for c in df.columns if c[1].strip().lower() == name.lower()]
                if not cols: return np.array([]), np.array([])
                reps = [df[c].dropna().values[:min_len] for c in cols]; reps = [r for r in reps if len(r) == min_len]
                if not reps: return np.array([]), np.array([])
                arr = np.array(reps); mean = np.mean(arr, axis=0)
                std = np.std(arr, axis=0, ddof=1) if arr.shape[0] > 1 else np.zeros_like(mean)
                return mean, std
            self.data_means[C_BIOMASS], self.data_stds[C_BIOMASS] = extract('Biomasa')
            self.data_means[C_SUBSTRATE], self.data_stds[C_SUBSTRATE] = extract('Sustrato')
            self.data_means[C_PRODUCT], self.data_stds[C_PRODUCT] = extract('Producto')
        except (IndexError, KeyError) as e: raise ValueError(f"Estructura de DataFrame inválida. Error: {e}")
    def _fit_component(self, func, t, data, p0, bounds, sigma=None, *args):
        try:
            if sigma is not None: sigma = np.where(sigma == 0, 1e-9, sigma)
            popt, _ = curve_fit(func, t, data, p0, bounds=bounds, maxfev=self.maxfev, ftol=1e-9, xtol=1e-9, sigma=sigma, absolute_sigma=bool(sigma is not None))
            pred = func(t, *popt, *args)
            if np.any(np.isnan(pred)): return None, np.nan, np.nan
            r2 = 1 - np.sum((data - pred)**2) / np.sum((data - np.mean(data))**2)
            rmse = np.sqrt(mean_squared_error(data, pred))
            return list(popt), r2, rmse
        except (RuntimeError, ValueError): return None, np.nan, np.nan
    def fit_all_models(self) -> None:
        t, bio_m, bio_s = self.data_time, self.data_means[C_BIOMASS], self.data_stds[C_BIOMASS]
        if t is None or bio_m is None or len(bio_m) == 0: return
        popt_bio = self._fit_biomass_model(t, bio_m, bio_s)
        if popt_bio:
            bio_p = list(self.params[C_BIOMASS].values())
            if self.data_means[C_SUBSTRATE] is not None and len(self.data_means[C_SUBSTRATE]) > 0: self._fit_substrate_model(t, self.data_means[C_SUBSTRATE], self.data_stds[C_SUBSTRATE], bio_p)
            if self.data_means[C_PRODUCT] is not None and len(self.data_means[C_PRODUCT]) > 0: self._fit_product_model(t, self.data_means[C_PRODUCT], self.data_stds[C_PRODUCT], bio_p)
    def _fit_biomass_model(self, t, data, std):
        p0, bounds = self.model.get_initial_params(t, data), self.model.get_param_bounds(t, data)
        popt, r2, rmse = self._fit_component(self.model.model_function, t, data, p0, bounds, std)
        if popt: self.params[C_BIOMASS], self.r2[C_BIOMASS], self.rmse[C_BIOMASS] = dict(zip(self.model.param_names, popt)), r2, rmse
        return popt
    def _fit_substrate_model(self, t, data, std, bio_p):
        p0, b = [data[0], 0.1, 0.01], ([0, -np.inf, -np.inf], [np.inf, np.inf, np.inf])
        popt, r2, rmse = self._fit_component(lambda t, so, p, q: self.substrate(t, so, p, q, bio_p), t, data, p0, b, std)
        if popt: self.params[C_SUBSTRATE], self.r2[C_SUBSTRATE], self.rmse[C_SUBSTRATE] = {'So': popt[0], 'p': popt[1], 'q': popt[2]}, r2, rmse
    def _fit_product_model(self, t, data, std, bio_p):
        p0, b = [data[0] if len(data)>0 else 0, 0.1, 0.01], ([0, -np.inf, -np.inf], [np.inf, np.inf, np.inf])
        popt, r2, rmse = self._fit_component(lambda t, po, a, b: self.product(t, po, a, b, bio_p), t, data, p0, b, std)
        if popt: self.params[C_PRODUCT], self.r2[C_PRODUCT], self.rmse[C_PRODUCT] = {'Po': popt[0], 'alpha': popt[1], 'beta': popt[2]}, r2, rmse
    def system_ode(self, y, t, bio_p, sub_p, prod_p):
        X, _, _ = y; dXdt = self.model.diff_function(X, t, bio_p)
        return [dXdt, -sub_p.get('p',0)*dXdt - sub_p.get('q',0)*X, prod_p.get('alpha',0)*dXdt + prod_p.get('beta',0)*X]
    def solve_odes(self, t_fine):
        p = self.params; bio_d, sub_d, prod_d = p[C_BIOMASS], p[C_SUBSTRATE], p[C_PRODUCT]
        if not bio_d: return None, None, None
        try:
            bio_p = list(bio_d.values()); y0 = [self._get_initial_biomass(bio_p), sub_d.get('So',0), prod_d.get('Po',0)]
            sol = odeint(self.system_ode, y0, t_fine, args=(bio_p, sub_d, prod_d))
            return sol[:, 0], sol[:, 1], sol[:, 2]
        except: return None, None, None
    def _generate_fine_time_grid(self, t_exp): return np.linspace(min(t_exp), max(t_exp), 500) if t_exp is not None and len(t_exp) > 1 else np.array([])
    def get_model_curves_for_plot(self, t_fine, use_diff):
        if use_diff and self.model.diff_function(1, 1, [1]*self.model.num_params) != 0: return self.solve_odes(t_fine)
        X, S, P = None, None, None
        if self.params[C_BIOMASS]:
            bio_p = list(self.params[C_BIOMASS].values()); X = self.model.model_function(t_fine, *bio_p)
            if self.params[C_SUBSTRATE]: S = self.substrate(t_fine, *list(self.params[C_SUBSTRATE].values()), bio_p)
            if self.params[C_PRODUCT]: P = self.product(t_fine, *list(self.params[C_PRODUCT].values()), bio_p)
        return X, S, P
    def plot_individual_or_combined(self, cfg, mode):
        t_exp, t_fine = cfg['time_exp'], self._generate_fine_time_grid(cfg['time_exp'])
        X_m, S_m, P_m = self.get_model_curves_for_plot(t_fine, cfg.get('use_differential', False))
        sns.set_style(cfg.get('style', 'whitegrid'))
        if mode == 'average':
            fig, (ax1,ax2,ax3) = plt.subplots(3,1,figsize=(10,15),sharex=True)
            fig.suptitle(f"Análisis: {cfg.get('exp_name','')} ({self.model.display_name})", fontsize=16); axes=[ax1,ax2,ax3]
        else:
            fig, ax1 = plt.subplots(figsize=(12,8)); fig.suptitle(f"Análisis: {cfg.get('exp_name','')} ({self.model.display_name})", fontsize=16)
            ax2,ax3 = ax1.twinx(),ax1.twinx(); ax3.spines["right"].set_position(("axes",1.18)); axes=[ax1,ax2,ax3]
        data_map = {C_BIOMASS:X_m, C_SUBSTRATE:S_m, C_PRODUCT:P_m}
        comb_styles = {C_BIOMASS:{'c':'#0072B2','mc':'#56B4E9','m':'o','ls':'-'}, C_SUBSTRATE:{'c':'#009E73','mc':'#34E499','m':'s','ls':'--'}, C_PRODUCT:{'c':'#D55E00','mc':'#F0E442','m':'^','ls':'-.'}}
        for ax, comp in zip(axes, COMPONENTS):
            ylabel, data, std, model_data = cfg.get('axis_labels',{}).get(f'{comp}_label',comp.capitalize()), cfg.get(f'{comp}_exp'), cfg.get(f'{comp}_std'), data_map.get(comp)
            if mode == 'combined':
                s = comb_styles[comp]; pc, lc, ms, ls = s['c'], s['mc'], s['m'], s['ls']
            else:
                pc,lc,ms,ls = cfg.get(f'{comp}_point_color'), cfg.get(f'{comp}_line_color'), cfg.get(f'{comp}_marker_style'), cfg.get(f'{comp}_line_style')
            ax_c = pc if mode == 'combined' else 'black'; ax.set_ylabel(ylabel,color=ax_c); ax.tick_params(axis='y',labelcolor=ax_c)
            if data is not None and len(data)>0:
                if cfg.get('show_error_bars') and std is not None and np.any(std>0): ax.errorbar(t_exp, data, yerr=std, fmt=ms, color=pc, label=f'{comp.capitalize()} (Datos)', capsize=cfg.get('error_cap_size',3), elinewidth=cfg.get('error_line_width',1))
                else: ax.plot(t_exp, data, ls='', marker=ms, color=pc, label=f'{comp.capitalize()} (Datos)')
            if model_data is not None and len(model_data)>0: ax.plot(t_fine, model_data, ls=ls, color=lc, label=f'{comp.capitalize()} (Modelo)')
            if mode=='average' and cfg.get('show_legend',True): ax.legend(loc=cfg.get('legend_pos','best'))
            if mode=='average' and cfg.get('show_params',True) and self.params[comp]:
                decs = cfg.get('decimal_places',3); p_txt='\n'.join([f"{k}={format_number(v,decs)}" for k,v in self.params[comp].items()])
                full_txt=f"{p_txt}\nR²={format_number(self.r2.get(comp,0),3)}, RMSE={format_number(self.rmse.get(comp,0),3)}"
                pos_x,ha = (0.95,'right') if 'right' in cfg.get('params_pos','upper right') else (0.05,'left')
                ax.text(pos_x,0.95,full_txt,transform=ax.transAxes,va='top',ha=ha,bbox=dict(boxstyle='round,pad=0.4',fc='wheat',alpha=0.7))
        if mode=='combined' and cfg.get('show_legend',True):
            h1,l1=axes[0].get_legend_handles_labels(); h2,l2=axes[1].get_legend_handles_labels(); h3,l3=axes[2].get_legend_handles_labels()
            axes[0].legend(handles=h1+h2+h3, labels=l1+l2+l3, loc=cfg.get('legend_pos','best'))
        axes[-1].set_xlabel(cfg.get('axis_labels',{}).get('x_label','Tiempo')); plt.tight_layout()
        if mode=='combined': fig.subplots_adjust(right=0.8)
        return fig

# --- FUNCIONES AUXILIARES, DE PLOTEO Y REPORTE (COMPLETAS) ---

def format_number(value: Any, decimals: int) -> str:
    """
    Formatea un número para su visualización. Si decimals es 0, usa un formato inteligente.
    """
    if not isinstance(value, (int, float, np.number)) or pd.isna(value):
        return "" if pd.isna(value) else str(value)
    
    decimals = int(decimals)
    
    if decimals == 0:
        if 0 < abs(value) < 1:
            return f"{value:.2e}"
        else:
            return str(int(round(value, 0)))
            
    return str(round(value, decimals))

def plot_model_comparison_matplotlib(plot_config: Dict, models_results: List[Dict]) -> plt.Figure:
    """
    Crea un gráfico de comparación de modelos estático usando Matplotlib/Seaborn.
    """
    time_exp = plot_config['time_exp']
    # Usar un modelo cualquiera solo para generar la rejilla de tiempo
    time_fine = BioprocessFitter(list(AVAILABLE_MODELS.values())[0])._generate_fine_time_grid(time_exp)
    num_models = len(models_results)
    
    palettes = {
        C_BIOMASS: sns.color_palette("Blues", num_models),
        C_SUBSTRATE: sns.color_palette("Greens", num_models),
        C_PRODUCT: sns.color_palette("Reds", num_models)
    }
    line_styles = ['-', '--', '-.', ':']

    sns.set_style(plot_config.get('style', 'whitegrid'))
    fig, ax1 = plt.subplots(figsize=(12, 8))

    # Configuración de los 3 ejes Y
    ax1.set_xlabel(plot_config['axis_labels']['x_label'])
    ax1.set_ylabel(plot_config['axis_labels']['biomass_label'], color="navy", fontsize=12)
    ax1.tick_params(axis='y', labelcolor="navy")
    ax2 = ax1.twinx()
    ax3 = ax1.twinx()
    ax3.spines["right"].set_position(("axes", 1.22))
    ax2.set_ylabel(plot_config['axis_labels']['substrate_label'], color="darkgreen", fontsize=12)
    ax2.tick_params(axis='y', labelcolor="darkgreen")
    ax3.set_ylabel(plot_config['axis_labels']['product_label'], color="darkred", fontsize=12)
    ax3.tick_params(axis='y', labelcolor="darkred")

    # Dibujar datos experimentales
    data_markers = {C_BIOMASS: 'o', C_SUBSTRATE: 's', C_PRODUCT: '^'}
    for ax, key, color, face in [(ax1, C_BIOMASS, 'navy', 'skyblue'), (ax2, C_SUBSTRATE, 'darkgreen', 'lightgreen'), (ax3, C_PRODUCT, 'darkred', 'lightcoral')]:
        data_exp = plot_config.get(f'{key}_exp')
        data_std = plot_config.get(f'{key}_std')
        if data_exp is not None:
            if plot_config.get('show_error_bars') and data_std is not None and np.any(data_std > 0):
                ax.errorbar(time_exp, data_exp, yerr=data_std, fmt=data_markers[key], color=color, label=f'{key.capitalize()} (Datos)', zorder=10, markersize=8, markerfacecolor=face, markeredgecolor=color, capsize=plot_config.get('error_cap_size', 3), elinewidth=plot_config.get('error_line_width', 1))
            else:
                ax.plot(time_exp, data_exp, ls='', marker=data_markers[key], label=f'{key.capitalize()} (Datos)', zorder=10, ms=8, mfc=face, mec=color, mew=1.5)

    # Dibujar curvas de los modelos
    for i, res in enumerate(models_results):
        ls = line_styles[i % len(line_styles)]
        model_info = AVAILABLE_MODELS.get(res["name"], MagicMock(display_name=res["name"]))
        model_display_name = model_info.display_name
        for key_short, ax, name_long in [('X', ax1, C_BIOMASS), ('S', ax2, C_SUBSTRATE), ('P', ax3, C_PRODUCT)]:
            if res.get(key_short) is not None:
                ax.plot(time_fine, res[key_short], color=palettes[name_long][i], ls=ls, label=f'{name_long.capitalize()} ({model_display_name})', alpha=0.9)

    fig.subplots_adjust(left=0.3, right=0.78, top=0.92, bottom=0.35 if plot_config.get('show_params') else 0.1)

    if plot_config.get('show_legend'):
        h1, l1 = ax1.get_legend_handles_labels(); h2, l2 = ax2.get_legend_handles_labels(); h3, l3 = ax3.get_legend_handles_labels()
        fig.legend(h1 + h2 + h3, l1 + l2 + l3, loc='center left', bbox_to_anchor=(0.0, 0.5), fancybox=True, shadow=True, fontsize='small')

    if plot_config.get('show_params'):
        total_width = 0.95; box_width = total_width / num_models; start_pos = (1.0 - total_width) / 2
        for i, res in enumerate(models_results):
            model_info = AVAILABLE_MODELS.get(res["name"], MagicMock(display_name=res["name"]))
            text = f"**{model_info.display_name}**\n" + _generate_model_param_text(res, plot_config.get('decimal_places', 3))
            fig.text(start_pos + i * box_width, 0.01, text, transform=fig.transFigure, fontsize=7.5, va='bottom', ha='left', bbox=dict(boxstyle='round,pad=0.4', fc='ivory', ec='gray', alpha=0.9))

    fig.suptitle(f"Comparación de Modelos: {plot_config.get('exp_name', '')}", fontsize=16)
    return fig

def plot_model_comparison_plotly(plot_config: Dict, models_results: List[Dict]) -> go.Figure:
    """
    Crea un gráfico de comparación de modelos interactivo usando Plotly.
    """
    fig = go.Figure()
    time_exp = plot_config['time_exp']
    time_fine = BioprocessFitter(list(AVAILABLE_MODELS.values())[0])._generate_fine_time_grid(time_exp)
    num_models = len(models_results)
    palettes = {
        C_BIOMASS: sns.color_palette("Blues", n_colors=num_models).as_hex(),
        C_SUBSTRATE: sns.color_palette("Greens", n_colors=num_models).as_hex(),
        C_PRODUCT: sns.color_palette("Reds", n_colors=num_models).as_hex()
    }
    line_styles, data_markers = ['solid', 'dash', 'dot', 'dashdot'], {C_BIOMASS: 'circle-open', C_SUBSTRATE: 'square-open', C_PRODUCT: 'diamond-open'}

    for key, y_axis, color in [(C_BIOMASS, 'y1', 'navy'), (C_SUBSTRATE, 'y2', 'darkgreen'), (C_PRODUCT, 'y3', 'darkred')]:
        data_exp, data_std = plot_config.get(f'{key}_exp'), plot_config.get(f'{key}_std')
        if data_exp is not None:
            error_y_config = dict(type='data', array=data_std, visible=True) if plot_config.get('show_error_bars') and data_std is not None and np.any(data_std > 0) else None
            fig.add_trace(go.Scatter(x=time_exp, y=data_exp, mode='markers', name=f'{key.capitalize()} (Datos)', marker=dict(color=color, size=10, symbol=data_markers[key], line=dict(width=2)), error_y=error_y_config, yaxis=y_axis, legendgroup="data"))

    for i, res in enumerate(models_results):
        ls = line_styles[i % len(line_styles)]
        model_display_name = AVAILABLE_MODELS.get(res["name"], MagicMock(display_name=res["name"])).display_name
        if res.get('X') is not None: fig.add_trace(go.Scatter(x=time_fine, y=res['X'], mode='lines', name=f'Biomasa ({model_display_name})', line=dict(color=palettes[C_BIOMASS][i], dash=ls), legendgroup=res["name"]))
        if res.get('S') is not None: fig.add_trace(go.Scatter(x=time_fine, y=res['S'], mode='lines', name=f'Sustrato ({model_display_name})', line=dict(color=palettes[C_SUBSTRATE][i], dash=ls), yaxis='y2', legendgroup=res["name"]))
        if res.get('P') is not None: fig.add_trace(go.Scatter(x=time_fine, y=res['P'], mode='lines', name=f'Producto ({model_display_name})', line=dict(color=palettes[C_PRODUCT][i], dash=ls), yaxis='y3', legendgroup=res["name"]))

    if plot_config.get('show_params'):
        x_positions = np.linspace(0, 1, num_models * 2 + 1)[1::2]
        for i, res in enumerate(models_results):
            model_display_name = AVAILABLE_MODELS.get(res["name"], MagicMock(display_name=res["name"])).display_name
            text = f"<b>{model_display_name}</b><br>" + _generate_model_param_text(res, plot_config.get('decimal_places', 3)).replace('\n', '<br>')
            fig.add_annotation(text=text, align='left', showarrow=False, xref='paper', yref='paper', x=x_positions[i], y=-0.35, bordercolor='gray', borderwidth=1, bgcolor='ivory', opacity=0.9)

    fig.update_layout(
        title=f"Comparación de Modelos (Interactivo): {plot_config.get('exp_name', '')}",
        xaxis=dict(domain=[0.18, 0.82]),
        yaxis=dict(title=plot_config['axis_labels']['biomass_label'], titlefont=dict(color='navy'), tickfont=dict(color='navy')),
        yaxis2=dict(title=plot_config['axis_labels']['substrate_label'], titlefont=dict(color='darkgreen'), tickfont=dict(color='darkgreen'), overlaying='y', side='right'),
        yaxis3=dict(title=plot_config['axis_labels']['product_label'], titlefont=dict(color='darkred'), tickfont=dict(color='darkred'), overlaying='y', side='right', position=0.85),
        legend=dict(traceorder="grouped", yanchor="middle", y=0.5, xanchor="right", x=-0.15),
        margin=dict(l=200, r=150, b=250 if plot_config.get('show_params') else 80, t=80),
        template="seaborn",
        showlegend=plot_config.get('show_legend', True)
    )
    return fig

def _generate_model_param_text(result: Dict, decimals: int) -> str:
    """Genera el texto formateado de los parámetros para las cajas de anotación."""
    text = ""
    for comp in COMPONENTS:
        if params := result.get('params', {}).get(comp):
            p_str = ', '.join([f"{k}={format_number(v, decimals)}" for k, v in params.items()])
            r2 = result.get('r2', {}).get(comp, 0)
            rmse = result.get('rmse', {}).get(comp, 0)
            text += f"<i>{comp[:4].capitalize()}:</i> {p_str}\n(R²={format_number(r2, 3)}, RMSE={format_number(rmse, 3)})\n"
    return text.strip()

def create_zip_file(image_list: List[Any]) -> Optional[str]:
    if not image_list:
        gr.Warning("No hay gráficos para descargar.")
        return None
    try:
        zip_buffer = io.BytesIO()
        with zipfile.ZipFile(zip_buffer, "w", zipfile.ZIP_DEFLATED) as zf:
            for i, fig in enumerate(image_list):
                buf = io.BytesIO()
                if isinstance(fig, go.Figure): buf.write(fig.to_image(format="png", scale=2, engine="kaleido"))
                elif isinstance(fig, plt.Figure): fig.savefig(buf, format='png', dpi=200, bbox_inches='tight'); plt.close(fig)
                elif isinstance(fig, Image.Image): fig.save(buf, 'PNG')
                else: continue
                buf.seek(0)
                zf.writestr(f"grafico_{i+1}.png", buf.read())
        with tempfile.NamedTemporaryFile(delete=False, suffix=".zip") as tmp:
            tmp.write(zip_buffer.getvalue())
            return tmp.name
    except Exception as e:
        traceback.print_exc()
        gr.Error(f"Error al crear el archivo ZIP: {e}")
        return None

def create_word_report(image_list: List[Any], table_df: pd.DataFrame, decimals: int) -> Optional[str]:
    if not image_list and (table_df is None or table_df.empty):
        gr.Warning("No hay datos ni gráficos para crear el reporte.")
        return None
    try:
        doc = Document()
        doc.add_heading('Reporte de Análisis de Cinéticas', 0)
        if table_df is not None and not table_df.empty:
            doc.add_heading('Tabla de Resultados', level=1)
            table = doc.add_table(rows=1, cols=len(table_df.columns), style='Table Grid')
            for i, col in enumerate(table_df.columns): table.cell(0, i).text = str(col)
            for _, row in table_df.iterrows():
                cells = table.add_row().cells
                for i, val in enumerate(row): cells[i].text = str(format_number(val, decimals))
        if image_list:
            doc.add_page_break()
            doc.add_heading('Gráficos Generados', level=1)
            for i, fig in enumerate(image_list):
                buf = io.BytesIO()
                if isinstance(fig, go.Figure): buf.write(fig.to_image(format="png", scale=2, engine="kaleido"))
                elif isinstance(fig, plt.Figure): fig.savefig(buf, format='png', dpi=200, bbox_inches='tight'); plt.close(fig)
                elif isinstance(fig, Image.Image): fig.save(buf, 'PNG')
                else: continue
                buf.seek(0)
                doc.add_paragraph(f'Gráfico {i+1}', style='Heading 3')
                doc.add_picture(buf, width=Inches(6.0))
        with tempfile.NamedTemporaryFile(delete=False, suffix=".docx") as tmp:
            doc.save(tmp.name)
            return tmp.name
    except Exception as e:
        traceback.print_exc()
        gr.Error(f"Error al crear el reporte de Word: {e}")
        return None

def create_pdf_report(image_list: List[Any], table_df: pd.DataFrame, decimals: int) -> Optional[str]:
    if not image_list and (table_df is None or table_df.empty):
        gr.Warning("No hay datos ni gráficos para crear el reporte.")
        return None
    try:
        pdf = FPDF()
        pdf.set_auto_page_break(auto=True, margin=15)
        pdf.add_page()
        pdf.set_font("Helvetica", 'B', 16)
        pdf.cell(0, 10, 'Reporte de Análisis de Cinéticas', new_x=XPos.LMARGIN, new_y=YPos.NEXT, align='C')
        if table_df is not None and not table_df.empty:
            pdf.ln(10)
            pdf.set_font("Helvetica", 'B', 12)
            pdf.cell(0, 10, 'Tabla de Resultados', new_x=XPos.LMARGIN, new_y=YPos.NEXT, align='L')
            pdf.set_font("Helvetica", 'B', 8)
            effective_page_width = pdf.w - 2 * pdf.l_margin
            num_cols = len(table_df.columns)
            col_width = effective_page_width / num_cols if num_cols > 0 else 0
            if num_cols > 15: pdf.set_font_size(6)
            elif num_cols > 10: pdf.set_font_size(7)
            for col in table_df.columns: pdf.cell(col_width, 10, str(col), border=1, align='C')
            pdf.ln()
            pdf.set_font("Helvetica", '', 7)
            if num_cols > 15: pdf.set_font_size(5)
            elif num_cols > 10: pdf.set_font_size(6)
            for _, row in table_df.iterrows():
                for val in row: pdf.cell(col_width, 10, str(format_number(val, decimals)), border=1, align='R')
                pdf.ln()
        if image_list:
            for i, fig in enumerate(image_list):
                pdf.add_page()
                pdf.set_font("Helvetica", 'B', 12)
                pdf.cell(0, 10, f'Gráfico {i+1}', new_x=XPos.LMARGIN, new_y=YPos.NEXT, align='L')
                pdf.ln(5)
                buf = io.BytesIO()
                if isinstance(fig, go.Figure): buf.write(fig.to_image(format="png", scale=2, engine="kaleido"))
                elif isinstance(fig, plt.Figure): fig.savefig(buf, format='png', dpi=200, bbox_inches='tight'); plt.close(fig)
                elif isinstance(fig, Image.Image): fig.save(buf, 'PNG')
                else: continue
                buf.seek(0)
                with tempfile.NamedTemporaryFile(delete=False, suffix=".png") as tmp_img:
                    tmp_img.write(buf.read())
                    pdf.image(tmp_img.name, x=None, y=None, w=pdf.w - 20)
                os.remove(tmp_img.name)
        pdf_bytes = pdf.output()
        with tempfile.NamedTemporaryFile(delete=False, suffix=".pdf") as tmp:
            tmp.write(pdf_bytes)
            return tmp.name
    except Exception as e:
        traceback.print_exc()
        gr.Error(f"Error al crear el reporte PDF: {e}")
        return None

# --- FUNCIÓN PRINCIPAL DE ANÁLISIS ---
def run_analysis(file, model_names, mode, engine, exp_names, settings):
    if not file: return [], pd.DataFrame(), "Error: Sube un archivo Excel.", pd.DataFrame()
    if not model_names: return [], pd.DataFrame(), "Error: Selecciona un modelo.", pd.DataFrame()
    try: xls = pd.ExcelFile(file.name)
    except Exception as e: return [], pd.DataFrame(), f"Error al leer archivo: {e}", pd.DataFrame()
    figs, results_data, msgs = [], [], []
    exp_list = [n.strip() for n in exp_names.split('\n') if n.strip()]
    for i, sheet in enumerate(xls.sheet_names):
        exp_name = exp_list[i] if i < len(exp_list) else f"Hoja '{sheet}'"
        try:
            df = pd.read_excel(xls, sheet_name=sheet, header=[0,1])
            reader = BioprocessFitter(list(AVAILABLE_MODELS.values())[0])
            reader.process_data_from_df(df)
            if reader.data_time is None: msgs.append(f"WARN: Sin datos de tiempo en '{sheet}'."); continue
            cfg = settings.copy(); cfg.update({'exp_name':exp_name, 'time_exp':reader.data_time})
            for c in COMPONENTS: cfg[f'{c}_exp'], cfg[f'{c}_std'] = reader.data_means[c], reader.data_stds[c]
            t_fine, plot_results = reader._generate_fine_time_grid(reader.data_time), []
            for m_name in model_names:
                if m_name not in AVAILABLE_MODELS: msgs.append(f"WARN: Modelo '{m_name}' no disponible."); continue
                fitter = BioprocessFitter(AVAILABLE_MODELS[m_name], maxfev=int(settings.get('maxfev',50000)))
                fitter.data_time, fitter.data_means, fitter.data_stds = reader.data_time, reader.data_means, reader.data_stds
                fitter.fit_all_models()
                row = {'Experimento':exp_name, 'Modelo':fitter.model.display_name}
                for c in COMPONENTS:
                    if fitter.params[c]: row.update({f'{c.capitalize()}_{k}':v for k,v in fitter.params[c].items()})
                    row[f'R2_{c.capitalize()}'], row[f'RMSE_{c.capitalize()}'] = fitter.r2.get(c), fitter.rmse.get(c)
                results_data.append(row)
                if mode in ["average","combined"]:
                    if hasattr(fitter,'plot_individual_or_combined'): figs.append(fitter.plot_individual_or_combined(cfg,mode))
                else:
                    X,S,P = fitter.get_model_curves_for_plot(t_fine, settings.get('use_differential',False))
                    plot_results.append({'name':m_name, 'X':X, 'S':S, 'P':P, 'params':fitter.params, 'r2':fitter.r2, 'rmse':fitter.rmse})
            if mode=="model_comparison" and plot_results:
                plot_func = plot_model_comparison_plotly if engine=='Plotly (Interactivo)' else plot_model_comparison_matplotlib
                if 'plot_model_comparison_plotly' in globals(): figs.append(plot_func(cfg, plot_results))
        except Exception as e: msgs.append(f"ERROR en '{sheet}': {e}"); traceback.print_exc()
    msg = "Análisis completado."+("\n"+"\n".join(msgs) if msgs else "")
    df_res = pd.DataFrame(results_data).dropna(axis=1,how='all')
    if not df_res.empty:
        id_c, p_c, m_c = ['Experimento','Modelo'], sorted([c for c in df_res.columns if '_' in c and 'R2' not in c and 'RMSE' not in c]), sorted([c for c in df_res.columns if 'R2' in c or 'RMSE' in c])
        df_res = df_res[[c for c in id_c+p_c+m_c if c in df_res.columns]]
        df_ui = df_res.copy()
        for c in df_ui.select_dtypes(include=np.number).columns: df_ui[c] = df_ui[c].apply(lambda x:format_number(x,settings.get('decimal_places',3)) if pd.notna(x) else '')
    else: df_ui = pd.DataFrame()
    return figs, df_ui, msg, df_res

# --- INTERFAZ DE USUARIO DE GRADIO (COMPLETA) ---

def create_gradio_interface() -> gr.Blocks:
    """
    Crea y configura la interfaz de usuario completa con Gradio.
    """
    # Obtener las opciones de modelo dinámicamente del registro
    MODEL_CHOICES = [(model.display_name, model.name) for model in AVAILABLE_MODELS.values()]
    # Seleccionar por defecto los primeros 3 modelos o todos si hay menos de 3
    DEFAULT_MODELS = [m.name for m in list(AVAILABLE_MODELS.values())[:3]]

    with gr.Blocks(theme=gr.themes.Soft(primary_hue="blue", secondary_hue="sky")) as demo:
        gr.Markdown("# 🔬 Analizador de Cinéticas de Bioprocesos")
        gr.Markdown("Sube tus datos, selecciona modelos, personaliza los gráficos y exporta los resultados.")
        
        with gr.Tabs():
            # --- PESTAÑA 1: GUÍA Y FORMATO ---
            with gr.TabItem("1. Guía y Formato de Datos"):
                with gr.Row():
                    with gr.Column(scale=2):
                        gr.Markdown(
                            """
                            ### Bienvenido al Analizador de Cinéticas
                            Esta herramienta te permite ajustar modelos matemáticos a tus datos de crecimiento microbiano.
                            
                            **Pasos a seguir:**
                            1.  Prepara tu archivo Excel según el formato especificado a la derecha.
                            2.  Ve a la pestaña **"2. Configuración y Ejecución"**.
                            3.  Sube tu archivo y selecciona los modelos cinéticos que deseas probar.
                            4.  Ajusta las opciones de visualización y análisis según tus preferencias.
                            5.  Haz clic en **"Analizar y Graficar"**.
                            6.  Explora los resultados en la pestaña **"3. Resultados"**.

                            ### Fórmulas de los Modelos
                            - **Logístico:** $ X(t) = \\frac{X_0 X_m e^{\\mu_m t}}{X_m - X_0 + X_0 e^{\\mu_m t}} $
                            - **Gompertz:** $ X(t) = X_m \\exp\\left(-\\exp\\left(\\frac{\\mu_m e}{X_m}(\\lambda-t)+1\\right)\\right) $
                            - **Moser:** $X(t) = X_m (1 - e^{-\\mu_m (t - K_s)})$
                            """
                        )
                    with gr.Column(scale=3):
                        gr.Markdown("### Formato del Archivo Excel")
                        gr.Markdown("Usa una **cabecera de dos niveles** para tus datos. La primera fila es el nombre de la réplica (ej. 'Rep1', 'Rep2') y la segunda el tipo de dato ('Tiempo', 'Biomasa', 'Sustrato', 'Producto').")
                        df_ejemplo = pd.DataFrame({
                            ('Rep1', 'Tiempo'): [0, 2, 4, 6], ('Rep1', 'Biomasa'): [0.1, 0.5, 2.5, 5.0], ('Rep1', 'Sustrato'): [10.0, 9.5, 7.0, 2.0],
                            ('Rep2', 'Tiempo'): [0, 2, 4, 6], ('Rep2', 'Biomasa'): [0.12, 0.48, 2.6, 5.2], ('Rep2', 'Sustrato'): [10.2, 9.6, 7.1, 2.1],
                        })
                        gr.DataFrame(df_ejemplo, interactive=False, label="Ejemplo de Formato")

            # --- PESTAÑA 2: CONFIGURACIÓN Y EJECUCIÓN ---
            with gr.TabItem("2. Configuración y Ejecución"):
                with gr.Row():
                    with gr.Column(scale=1):
                        file_input = gr.File(label="Sube tu archivo Excel (.xlsx)", file_types=['.xlsx'])
                        exp_names_input = gr.Textbox(label="Nombres de Experimentos (opcional)", placeholder="Nombre Hoja 1\nNombre Hoja 2\n...", lines=3, info="Un nombre por línea, en el mismo orden que las hojas del Excel.")
                        model_selection_input = gr.CheckboxGroup(choices=MODEL_CHOICES, label="Modelos a Probar", value=DEFAULT_MODELS)
                        analysis_mode_input = gr.Radio(["average", "combined", "model_comparison"], label="Modo de Análisis", value="average", info="Average: Gráficos separados.\nCombined: Un gráfico con 3 ejes.\nComparación: Gráfico global comparativo.")
                        plotting_engine_input = gr.Radio(["Seaborn (Estático)", "Plotly (Interactivo)"], label="Motor Gráfico (en modo Comparación)", value="Plotly (Interactivo)")
                    
                    with gr.Column(scale=2):
                        with gr.Accordion("Opciones Generales de Análisis", open=True):
                            decimal_places_input = gr.Slider(0, 10, value=3, step=1, label="Precisión Decimal de Parámetros", info="0 para notación científica automática.")
                            show_params_input = gr.Checkbox(label="Mostrar Parámetros en Gráfico", value=True)
                            show_legend_input = gr.Checkbox(label="Mostrar Leyenda en Gráfico", value=True)
                            use_differential_input = gr.Checkbox(label="Usar EDO para graficar", value=False, info="Simula con ecuaciones diferenciales en lugar de la fórmula integral.")
                            maxfev_input = gr.Number(label="Iteraciones Máximas de Ajuste (maxfev)", value=50000)

                        with gr.Accordion("Etiquetas de los Ejes", open=True):
                            with gr.Row(): xlabel_input = gr.Textbox(label="Etiqueta Eje X", value="Tiempo (h)", interactive=True)
                            with gr.Row():
                                ylabel_biomass_input = gr.Textbox(label="Etiqueta Biomasa", value="Biomasa (g/L)", interactive=True)
                                ylabel_substrate_input = gr.Textbox(label="Etiqueta Sustrato", value="Sustrato (g/L)", interactive=True)
                                ylabel_product_input = gr.Textbox(label="Etiqueta Producto", value="Producto (g/L)", interactive=True)
                        
                        with gr.Accordion("Opciones de Estilo (Modo 'Average' y 'Combined')", open=False):
                            style_input = gr.Dropdown(['whitegrid', 'darkgrid', 'white', 'dark', 'ticks'], label="Estilo General (Matplotlib)", value='whitegrid')
                            with gr.Row():
                                with gr.Column():
                                    gr.Markdown("**Biomasa**"); biomass_point_color_input = gr.ColorPicker(label="Color Puntos", value='#0072B2'); biomass_line_color_input = gr.ColorPicker(label="Color Línea", value='#56B4E9'); biomass_marker_style_input = gr.Dropdown(['o','s','^','D','p','*','X'], label="Marcador", value='o'); biomass_line_style_input = gr.Dropdown(['-','--','-.',':'], label="Estilo Línea", value='-')
                                with gr.Column():
                                    gr.Markdown("**Sustrato**"); substrate_point_color_input = gr.ColorPicker(label="Color Puntos", value='#009E73'); substrate_line_color_input = gr.ColorPicker(label="Color Línea", value='#34E499'); substrate_marker_style_input = gr.Dropdown(['o','s','^','D','p','*','X'], label="Marcador", value='s'); substrate_line_style_input = gr.Dropdown(['-','--','-.',':'], label="Estilo Línea", value='--')
                                with gr.Column():
                                    gr.Markdown("**Producto**"); product_point_color_input = gr.ColorPicker(label="Color Puntos", value='#D55E00'); product_line_color_input = gr.ColorPicker(label="Color Línea", value='#F0E442'); product_marker_style_input = gr.Dropdown(['o','s','^','D','p','*','X'], label="Marcador", value='^'); product_line_style_input = gr.Dropdown(['-','--','-.',':'], label="Estilo Línea", value='-.')
                            with gr.Row():
                                legend_pos_input = gr.Radio(["best","upper right","upper left","lower left","lower right","center"], label="Posición Leyenda", value="best")
                                params_pos_input = gr.Radio(["upper right","upper left","lower right","lower left"], label="Posición Parámetros", value="upper right")
                        
                        with gr.Accordion("Opciones de Barra de Error", open=False):
                            show_error_bars_input = gr.Checkbox(label="Mostrar barras de error (si hay réplicas)", value=True)
                            error_cap_size_input = gr.Slider(1, 10, 3, step=1, label="Tamaño Tapa Error")
                            error_line_width_input = gr.Slider(0.5, 5, 1.0, step=0.5, label="Grosor Línea Error")

                simulate_btn = gr.Button("Analizar y Graficar", variant="primary")

            # --- PESTAÑA 3: RESULTADOS ---
            with gr.TabItem("3. Resultados"):
                status_output = gr.Textbox(label="Estado del Análisis", interactive=False, lines=2)
                gallery_output = gr.Gallery(label="Gráficos Generados", columns=1, height=600, object_fit="contain", preview=True)
                with gr.Accordion("Descargar Reportes y Gráficos", open=True):
                    with gr.Row():
                        zip_btn = gr.Button("Descargar Gráficos (.zip)"); word_btn = gr.Button("Descargar Reporte (.docx)"); pdf_btn = gr.Button("Descargar Reporte (.pdf)")
                    download_output = gr.File(label="Archivo de Descarga", interactive=False)
                gr.Markdown("### Tabla de Resultados Numéricos"); table_output = gr.DataFrame(wrap=True)
                with gr.Row():
                    excel_btn = gr.Button("Descargar Tabla (.xlsx)"); csv_btn = gr.Button("Descargar Tabla (.csv)")
                download_table_output = gr.File(label="Descargar Tabla", interactive=False)
                df_for_export = gr.State(pd.DataFrame()); figures_for_export = gr.State([])
        
        # --- LÓGICA DE CONEXIÓN (WRAPPER Y EVENTOS .CLICK()) ---
        demo.queue()

        def simulation_wrapper(file, models, mode, engine, names, use_diff, s_par, s_leg, maxfev, decimals, x_label, bio_label, sub_label, prod_label, style, s_err, cap, lw, l_pos, p_pos, bio_pc, bio_lc, bio_ms, bio_ls, sub_pc, sub_lc, sub_ms, sub_ls, prod_pc, prod_lc, prod_ms, prod_ls):
            try:
                def rgba_to_hex(rgba_string: str) -> str:
                    if not isinstance(rgba_string, str) or rgba_string.startswith('#'): return rgba_string
                    try:
                        parts = rgba_string.lower().replace('rgba', '').replace('rgb', '').replace('(', '').replace(')', '')
                        r, g, b, *_ = map(float, parts.split(',')); return f'#{int(r):02x}{int(g):02x}{int(b):02x}'
                    except (ValueError, TypeError): return "#000000"

                plot_settings = {
                    'decimal_places': int(decimals), 'use_differential': use_diff, 'style': style, 'show_legend': s_leg, 'show_params': s_par, 'maxfev': int(maxfev),
                    'axis_labels': {'x_label': x_label, 'biomass_label': bio_label, 'substrate_label': sub_label, 'product_label': prod_label},
                    'legend_pos': l_pos, 'params_pos': p_pos, 'show_error_bars': s_err, 'error_cap_size': cap, 'error_line_width': lw,
                    f'{C_BIOMASS}_point_color': rgba_to_hex(bio_pc), f'{C_BIOMASS}_line_color': rgba_to_hex(bio_lc), f'{C_BIOMASS}_marker_style': bio_ms, f'{C_BIOMASS}_line_style': bio_ls,
                    f'{C_SUBSTRATE}_point_color': rgba_to_hex(sub_pc), f'{C_SUBSTRATE}_line_color': rgba_to_hex(sub_lc), f'{C_SUBSTRATE}_marker_style': sub_ms, f'{C_SUBSTRATE}_line_style': sub_ls,
                    f'{C_PRODUCT}_point_color': rgba_to_hex(prod_pc), f'{C_PRODUCT}_line_color': rgba_to_hex(prod_lc), f'{C_PRODUCT}_marker_style': prod_ms, f'{C_PRODUCT}_line_style': prod_ls,
                }
                figures, df_ui, msg, df_export = run_analysis(file, models, mode, engine, names, plot_settings)
                image_list = []
                for fig in figures:
                    buf = io.BytesIO()
                    if isinstance(fig, go.Figure): buf.write(fig.to_image(format="png", scale=2, engine="kaleido"))
                    elif isinstance(fig, plt.Figure): fig.savefig(buf, format='png', bbox_inches='tight', dpi=150); plt.close(fig)
                    buf.seek(0); image_list.append(Image.open(buf).convert("RGB"))
                return image_list, df_ui, msg, df_export, figures
            except Exception as e:
                print(f"--- ERROR CAPTURADO EN WRAPPER ---\n{traceback.format_exc()}"); return [], pd.DataFrame(), f"Error Crítico: {e}", pd.DataFrame(), []

        all_inputs = [
            file_input, model_selection_input, analysis_mode_input, plotting_engine_input, exp_names_input, 
            use_differential_input, show_params_input, show_legend_input, maxfev_input, decimal_places_input,
            xlabel_input, ylabel_biomass_input, ylabel_substrate_input, ylabel_product_input,
            style_input, show_error_bars_input, error_cap_size_input, error_line_width_input, legend_pos_input, params_pos_input,
            biomass_point_color_input, biomass_line_color_input, biomass_marker_style_input, biomass_line_style_input,
            substrate_point_color_input, substrate_line_color_input, substrate_marker_style_input, substrate_line_style_input,
            product_point_color_input, product_line_color_input, product_marker_style_input, product_line_style_input
        ]
        all_outputs = [gallery_output, table_output, status_output, df_for_export, figures_for_export]
        
        simulate_btn.click(fn=simulation_wrapper, inputs=all_inputs, outputs=all_outputs)
        zip_btn.click(fn=create_zip_file, inputs=[figures_for_export], outputs=[download_output])
        word_btn.click(fn=create_word_report, inputs=[figures_for_export, df_for_export, decimal_places_input], outputs=[download_output])
        pdf_btn.click(fn=create_pdf_report, inputs=[figures_for_export, df_for_export, decimal_places_input], outputs=[download_output])
        
        def export_table_to_file(df: pd.DataFrame, file_format: str) -> Optional[str]:
            if df is None or df.empty: gr.Warning("No hay datos para exportar."); return None
            suffix = ".xlsx" if file_format == "excel" else ".csv"
            with tempfile.NamedTemporaryFile(delete=False, suffix=suffix) as tmp:
                if file_format == "excel": df.to_excel(tmp.name, index=False)
                else: df.to_csv(tmp.name, index=False, encoding='utf-8-sig')
                return tmp.name
        
        excel_btn.click(fn=lambda df: export_table_to_file(df, "excel"), inputs=[df_for_export], outputs=[download_table_output])
        csv_btn.click(fn=lambda df: export_table_to_file(df, "csv"), inputs=[df_for_export], outputs=[download_table_output])
    
    return demo

# --- PUNTO DE ENTRADA PRINCIPAL ---

if __name__ == '__main__':
    """
    Este bloque se ejecuta solo cuando el script es llamado directamente.
    Crea la interfaz de Gradio y la lanza, haciendo que la aplicación
    esté disponible en una URL local (y opcionalmente pública si share=True).
    """
    
    # Todas las funciones necesarias (create_gradio_interface, run_analysis,
    # funciones de ploteo y reporte) ya están definidas en el alcance global,
    # por lo que no es necesario rellenar nada aquí.
    
    # Crear la aplicación Gradio llamando a la función que la construye.
    gradio_app = create_gradio_interface()
    
    # Lanzar la aplicación.
    # share=True: Crea un túnel público temporal a tu aplicación (útil para compartir).
    # debug=True: Muestra más información de depuración en la consola si ocurren errores.
    gradio_app.launch(share=True, debug=True)