import gradio as gr import pandas as pd import plotly.express as px from pycaret.classification import load_model, predict_model import os import ast # --- 1. Definición de Rutas --- DATA_FOLDER = "data" MODELS_FOLDER = "models" PROCESSED_DATA_PATH = os.path.join(DATA_FOLDER, "processed_reddit_data.csv") AUTOML_MODEL_PATH = os.path.join(MODELS_FOLDER, "sentiment_model_v2") # --- 2. Carga de Datos y Modelos --- df_roberta = pd.DataFrame() df_automl_predictions = pd.DataFrame() automl_model_info_str = "El modelo de AutoML no se pudo cargar. Ejecuta python_train.py." try: df_roberta = pd.read_csv(PROCESSED_DATA_PATH) except Exception as e: print(f"Advertencia: No se pudo cargar '{PROCESSED_DATA_PATH}'. {e}") try: automl_model_pipeline = load_model(AUTOML_MODEL_PATH) automl_model_info_str = f"Mejor modelo encontrado por AutoML: 
{str(automl_model_pipeline.steps[-1][1])}
" if not df_roberta.empty: print("Generando predicciones con el modelo AutoML...") data_for_automl = df_roberta[['text', 'brand']] predictions_automl = predict_model(automl_model_pipeline, data=data_for_automl) # Unimos las predicciones de AutoML con las emociones detectadas por RoBERTa para una comparación completa df_automl_predictions = predictions_automl.rename(columns={'prediction_label': 'sentiment'}) df_automl_predictions['emotion'] = df_roberta['emotion'] # Añadir columna de emoción print("Predicciones de AutoML generadas y enriquecidas.") except Exception as e: print(f"Advertencia: No se pudo cargar o usar el modelo AutoML. {e}") # --- 3. Funciones de la App --- def create_sentiment_plot(data, brand_name, model_type): if data.empty or 'sentiment' not in data.columns: return px.pie(title=f"Sin Datos para {brand_name} ({model_type})") sentiment_counts = data['sentiment'].value_counts() fig = px.pie(sentiment_counts, values=sentiment_counts.values, names=sentiment_counts.index, title=f"Sentimiento para {brand_name} ({model_type})", color=sentiment_counts.index, color_discrete_map={'Positive':'#2ca02c', 'Negative':'#d62728', 'Neutral':'#7f7f7f'}) return fig def create_emotion_plot(data, brand_name, title_suffix=""): if data.empty or 'emotion' not in data.columns: return px.bar(title=f"Sin Datos de Emociones para {brand_name}") emotion_counts = data['emotion'].value_counts().nlargest(7) fig = px.bar(emotion_counts, x=emotion_counts.index, y=emotion_counts.values, title=f"Top Emociones Detectadas {title_suffix}", labels={'x':'Emoción', 'y':'Cantidad'}) return fig def update_dashboard(df_source, brand_filter, sentiment_filter): if df_source.empty: return None, None, None, pd.DataFrame() filtered = df_source.copy() if sentiment_filter != "Todos": filtered = filtered[filtered['sentiment'] == sentiment_filter] if brand_filter == "Ambas": intel_df = filtered[filtered['brand'] == 'Intel'] amd_df = filtered[filtered['brand'] == 'AMD'] intel_plot = create_sentiment_plot(intel_df, "Intel", "RoBERTa" if 'emotion' in df_source.columns else "AutoML") amd_plot = create_sentiment_plot(amd_df, "AMD", "RoBERTa" if 'emotion' in df_source.columns else "AutoML") emotion_plot = create_emotion_plot(filtered, "Ambas Marcas", "(RoBERTa)") return intel_plot, amd_plot, emotion_plot, filtered.head(100) else: brand_df = filtered[filtered['brand'] == brand_filter] brand_plot = create_sentiment_plot(brand_df, brand_filter, "RoBERTa" if 'emotion' in df_source.columns else "AutoML") emotion_plot = create_emotion_plot(brand_df, brand_filter, f"({brand_filter} - RoBERTa)") return brand_plot, brand_plot, emotion_plot, brand_df.head(100) # --- 4. Interfaz de Gradio --- custom_theme = gr.themes.Base(font=[gr.themes.GoogleFont("Inter"), "Arial", "sans-serif"]).set( body_background_fill="#f0f2f5", block_background_fill="white", block_border_width="1px", block_shadow="*shadow_drop_lg", button_primary_background_fill="#00a3c0", button_primary_background_fill_hover="#00839c", button_primary_text_color="white", ) custom_css = "h1 { color: #2c3e50; text-align: center; font-weight: 700; } h3 { color: #333d4d; font-weight: 600; } p { color: #5d6776; text-align: center; }" with gr.Blocks(theme=custom_theme, css=custom_css, title="Dashboard Comparativo: Intel vs AMD") as demo: gr.Markdown("

Dashboard Comparativo: Intel vs AMD en Reddit

") with gr.Tabs(): # --- Pestaña 1: Resultados con AutoML (Nuestro Modelo) --- with gr.TabItem("Resultados con AutoML (Nuestro Modelo)"): gr.Markdown("

Análisis utilizando nuestro modelo propio, entrenado con AutoML. Permite comparar su rendimiento con el modelo experto.

") with gr.Row(): brand_filter_automl = gr.Dropdown(choices=["Ambas", "Intel", "AMD"], value="Ambas", label="Filtrar por Marca") sentiment_filter_automl = gr.Dropdown(choices=["Todos"] + (df_automl_predictions['sentiment'].unique().tolist() if not df_automl_predictions.empty else []), value="Todos", label="Filtrar por Sentimiento") with gr.Row(): intel_automl_plot = gr.Plot() amd_automl_plot = gr.Plot() with gr.Row(): emotion_automl_plot = gr.Plot() gr.Markdown("

Vista Previa de Comentarios

") comments_automl = gr.DataFrame(headers=["Marca", "Comentario", "Sentimiento (AutoML)", "Emoción (RoBERTa)"], wrap=True) filters_automl = [brand_filter_automl, sentiment_filter_automl] outputs_automl = [intel_automl_plot, amd_automl_plot, emotion_automl_plot, comments_automl] demo.load(lambda b, s: update_dashboard(df_automl_predictions, b, s), inputs=filters_automl, outputs=outputs_automl) for f in filters_automl: f.change(lambda b, s: update_dashboard(df_automl_predictions, b, s), inputs=filters_automl, outputs=outputs_automl) gr.Markdown("

Detalles del Modelo AutoML

") gr.HTML(value=automl_model_info_str) # --- Pestaña 2: Resultados con Modelo Experto (RoBERTa) --- with gr.TabItem("Resultados con Modelo Experto (RoBERTa)"): gr.Markdown("

Análisis de alta precisión utilizando un modelo Transformer pre-entrenado (State-of-the-Art).

") with gr.Row(): brand_filter_roberta = gr.Dropdown(choices=["Ambas", "Intel", "AMD"], value="Ambas", label="Filtrar por Marca") sentiment_filter_roberta = gr.Dropdown(choices=["Todos"] + (df_roberta['sentiment'].unique().tolist() if not df_roberta.empty else []), value="Todos", label="Filtrar por Sentimiento") with gr.Row(): intel_sentiment_plot = gr.Plot() amd_sentiment_plot = gr.Plot() with gr.Row(): emotion_plot_roberta = gr.Plot() gr.Markdown("

Vista Previa de Comentarios

") comments_roberta = gr.DataFrame(headers=["Marca", "Comentario", "Sentimiento (RoBERTa)", "Emoción"], wrap=True) filters_roberta = [brand_filter_roberta, sentiment_filter_roberta] outputs_roberta = [intel_sentiment_plot, amd_sentiment_plot, emotion_plot_roberta, comments_roberta] demo.load(lambda b, s: update_dashboard(df_roberta, b, s), inputs=filters_roberta, outputs=outputs_roberta) for f in filters_roberta: f.change(lambda b, s: update_dashboard(df_roberta, b, s), inputs=filters_roberta, outputs=outputs_roberta) demo.launch()