Spaces:
Running
Running
# modules/studentact/current_situation_interface.py | |
import streamlit as st | |
import logging | |
from ..utils.widget_utils import generate_unique_key | |
import matplotlib.pyplot as plt | |
import numpy as np | |
from ..database.current_situation_mongo_db import store_current_situation_result | |
from ..database.writing_progress_mongo_db import ( | |
store_writing_baseline, | |
store_writing_progress, | |
get_writing_baseline, | |
get_writing_progress, | |
get_latest_writing_metrics | |
) | |
from .current_situation_analysis import ( | |
analyze_text_dimensions, | |
analyze_clarity, | |
analyze_vocabulary_diversity, | |
analyze_cohesion, | |
analyze_structure, | |
get_dependency_depths, | |
normalize_score, | |
generate_sentence_graphs, | |
generate_word_connections, | |
generate_connection_paths, | |
create_vocabulary_network, | |
create_syntax_complexity_graph, | |
create_cohesion_heatmap | |
) | |
# Configuración del estilo de matplotlib para el gráfico de radar | |
plt.rcParams['font.family'] = 'sans-serif' | |
plt.rcParams['axes.grid'] = True | |
plt.rcParams['axes.spines.top'] = False | |
plt.rcParams['axes.spines.right'] = False | |
logger = logging.getLogger(__name__) | |
#################################### | |
TEXT_TYPES = { | |
'academic_article': { | |
'name': 'Artículo Académico', | |
'thresholds': { | |
'vocabulary': {'min': 0.70, 'target': 0.85}, | |
'structure': {'min': 0.75, 'target': 0.90}, | |
'cohesion': {'min': 0.65, 'target': 0.80}, | |
'clarity': {'min': 0.70, 'target': 0.85} | |
} | |
}, | |
'student_essay': { | |
'name': 'Trabajo Universitario', | |
'thresholds': { | |
'vocabulary': {'min': 0.60, 'target': 0.75}, | |
'structure': {'min': 0.65, 'target': 0.80}, | |
'cohesion': {'min': 0.55, 'target': 0.70}, | |
'clarity': {'min': 0.60, 'target': 0.75} | |
} | |
}, | |
'general_communication': { | |
'name': 'Comunicación General', | |
'thresholds': { | |
'vocabulary': {'min': 0.50, 'target': 0.65}, | |
'structure': {'min': 0.55, 'target': 0.70}, | |
'cohesion': {'min': 0.45, 'target': 0.60}, | |
'clarity': {'min': 0.50, 'target': 0.65} | |
} | |
} | |
} | |
#################################### | |
ANALYSIS_DIMENSION_MAPPING = { | |
'morphosyntactic': { | |
'primary': ['vocabulary', 'clarity'], | |
'secondary': ['structure'], | |
'tools': ['arc_diagrams', 'word_repetition'] | |
}, | |
'semantic': { | |
'primary': ['cohesion', 'structure'], | |
'secondary': ['vocabulary'], | |
'tools': ['concept_graphs', 'semantic_networks'] | |
}, | |
'discourse': { | |
'primary': ['cohesion', 'structure'], | |
'secondary': ['clarity'], | |
'tools': ['comparative_analysis'] | |
} | |
} | |
############################################################################## | |
# FUNCIÓN PRINCIPAL | |
############################################################################## | |
def display_current_situation_interface(lang_code, nlp_models, t): | |
""" | |
TAB: | |
- Expander con radio para tipo de texto | |
Contenedor-1 (2 "filas" con borde): | |
- Fila 1: Métricas base | |
- Fila 2: Métricas iteración | |
Contenedor-2 (2 columnas): | |
- Col1: Texto base | |
- Col2: Texto iteración | |
Al final, Recomendaciones en una sola línea. | |
""" | |
# --- Inicializar session_state --- | |
if 'base_text' not in st.session_state: | |
st.session_state.base_text = "" | |
if 'iter_text' not in st.session_state: | |
st.session_state.iter_text = "" | |
if 'base_metrics' not in st.session_state: | |
st.session_state.base_metrics = {} | |
if 'iter_metrics' not in st.session_state: | |
st.session_state.iter_metrics = {} | |
if 'show_base' not in st.session_state: | |
st.session_state.show_base = False | |
if 'show_iter' not in st.session_state: | |
st.session_state.show_iter = False | |
# Creamos un TAB | |
tabs = st.tabs(["Análisis de Texto"]) | |
with tabs[0]: | |
# 1) Expander con radio | |
with st.expander("Selecciona el tipo de texto", expanded=True): | |
text_type = st.radio( | |
"¿Qué tipo de texto quieres analizar?", | |
options=list(TEXT_TYPES.keys()), | |
format_func=lambda x: TEXT_TYPES[x]['name'], | |
index=0 | |
) | |
st.session_state.current_text_type = text_type | |
st.markdown("---") | |
# --------------------------------------------------------------------- | |
# CONTENEDOR-1: Métricas (2 filas con borde) | |
# --------------------------------------------------------------------- | |
with st.container(): | |
# Fila 1: Métricas base | |
if st.session_state.show_base: | |
# Usamos HTML + CSS para enmarcar las métricas en un div con borde | |
st.markdown( | |
""" | |
<div style="border:1px solid black; padding:10px; margin-bottom:10px;"> | |
<p style="font-weight:bold;">Métricas de la línea base</p> | |
""", | |
unsafe_allow_html=True | |
) | |
# ---- Aquí llamamos a la función que muestra las 4 métricas ---- | |
display_metrics_in_one_row(st.session_state.base_metrics, text_type) | |
# Cerramos el div | |
st.markdown("</div>", unsafe_allow_html=True) | |
# Fila 2: Métricas de iteración | |
if st.session_state.show_iter: | |
st.markdown( | |
""" | |
<div style="border:1px solid black; padding:10px; margin-bottom:10px;"> | |
<p style="font-weight:bold;">Métricas de la iteración</p> | |
""", | |
unsafe_allow_html=True | |
) | |
display_metrics_in_one_row(st.session_state.iter_metrics, text_type) | |
st.markdown("</div>", unsafe_allow_html=True) | |
# --------------------------------------------------------------------- | |
# CONTENEDOR-2: 2 columnas (texto base | texto iteración) | |
# --------------------------------------------------------------------- | |
with st.container(): | |
col_left, col_right = st.columns(2) | |
with col_left: | |
st.markdown("**Texto base**") | |
text_base = st.text_area( | |
label="", | |
value=st.session_state.base_text, | |
key="text_base_area", | |
placeholder="Pega aquí tu texto base" | |
) | |
if st.button("Analizar Base"): | |
with st.spinner("Analizando texto base..."): | |
doc = nlp_models[lang_code](text_base) | |
metrics = analyze_text_dimensions(doc) | |
st.session_state.base_text = text_base | |
st.session_state.base_metrics = metrics | |
st.session_state.show_base = True | |
# Al analizar base, reiniciamos la iteración | |
st.session_state.show_iter = False | |
with col_right: | |
st.markdown("**Texto de iteración**") | |
text_iter = st.text_area( | |
label="", | |
value=st.session_state.iter_text, | |
key="text_iter_area", | |
placeholder="Edita y mejora tu texto...", | |
disabled=not st.session_state.show_base | |
) | |
if st.button("Analizar Iteración", disabled=not st.session_state.show_base): | |
with st.spinner("Analizando iteración..."): | |
doc = nlp_models[lang_code](text_iter) | |
metrics = analyze_text_dimensions(doc) | |
st.session_state.iter_text = text_iter | |
st.session_state.iter_metrics = metrics | |
st.session_state.show_iter = True | |
# --------------------------------------------------------------------- | |
# Recomendaciones al final (una sola línea, por ejemplo) | |
# --------------------------------------------------------------------- | |
# Solo mostramos si tenemos iteración | |
if st.session_state.show_iter: | |
# Podríamos mostrar todo en una misma línea usando st.columns(1) | |
# o simplemente un st.container(). Lo muestro en un solo contenedor: | |
with st.container(): | |
st.markdown("**Recomendaciones**") | |
# Si las quieres en una sola línea, podrías hacer algo como: | |
# rec_col = st.columns(1)[0] # Solo 1 columna | |
# con rec_col: ... | |
# Pero en general, un for con st.write estará apilado verticalmente. | |
# Ejemplo de iterar dimensiones | |
any_reco = False | |
for dimension, values in st.session_state.iter_metrics.items(): | |
score = values['normalized_score'] | |
target = TEXT_TYPES[text_type]['thresholds'][dimension]['target'] | |
if score < target: | |
suggest_improvement_tools(dimension) | |
any_reco = True | |
if not any_reco: | |
st.info("¡No hay recomendaciones! Todas las métricas superan la meta.") | |
#Funciones de visualización ################################## | |
############################################################################## | |
# Función para mostrar métricas en una sola fila | |
############################################################################## | |
def display_metrics_in_one_line(metrics, text_type): | |
""" | |
Muestra las cuatro dimensiones (Vocabulario, Estructura, Cohesión, Claridad) | |
en una sola línea, usando 4 columnas. | |
""" | |
thresholds = TEXT_TYPES[text_type]['thresholds'] | |
dimensions = ["vocabulary", "structure", "cohesion", "clarity"] | |
col1, col2, col3, col4 = st.columns(4) | |
cols = [col1, col2, col3, col4] | |
for dim, col in zip(dimensions, cols): | |
score = metrics[dim]['normalized_score'] | |
target = thresholds[dim]['target'] | |
min_val = thresholds[dim]['min'] | |
# Determinar estado y color | |
if score < min_val: | |
status = "⚠️ Por mejorar" | |
color = "inverse" | |
elif score < target: | |
status = "📈 Aceptable" | |
color = "off" | |
else: | |
status = "✅ Óptimo" | |
color = "normal" | |
with col: | |
col.metric( | |
label=dim.capitalize(), | |
value=f"{score:.2f}", | |
delta=f"{status} (Meta: {target:.2f})", | |
delta_color=color | |
border=True | |
) | |
#################################################################### | |
def display_metrics_analysis(metrics, text_type=None): | |
""" | |
Muestra los resultados del análisis: métricas verticalmente y gráfico radar. | |
""" | |
try: | |
# Usar valor por defecto si no se especifica tipo | |
text_type = text_type or 'student_essay' | |
# Obtener umbrales según el tipo de texto | |
thresholds = TEXT_TYPES[text_type]['thresholds'] | |
# Crear dos columnas para las métricas y el gráfico | |
metrics_col, graph_col = st.columns([1, 1.5]) | |
# Columna de métricas | |
with metrics_col: | |
metrics_config = [ | |
{ | |
'label': "Vocabulario", | |
'key': 'vocabulary', | |
'value': metrics['vocabulary']['normalized_score'], | |
'help': "Riqueza y variedad del vocabulario", | |
'thresholds': thresholds['vocabulary'] | |
}, | |
{ | |
'label': "Estructura", | |
'key': 'structure', | |
'value': metrics['structure']['normalized_score'], | |
'help': "Organización y complejidad de oraciones", | |
'thresholds': thresholds['structure'] | |
}, | |
{ | |
'label': "Cohesión", | |
'key': 'cohesion', | |
'value': metrics['cohesion']['normalized_score'], | |
'help': "Conexión y fluidez entre ideas", | |
'thresholds': thresholds['cohesion'] | |
}, | |
{ | |
'label': "Claridad", | |
'key': 'clarity', | |
'value': metrics['clarity']['normalized_score'], | |
'help': "Facilidad de comprensión del texto", | |
'thresholds': thresholds['clarity'] | |
} | |
] | |
# Mostrar métricas | |
for metric in metrics_config: | |
value = metric['value'] | |
if value < metric['thresholds']['min']: | |
status = "⚠️ Por mejorar" | |
color = "inverse" | |
elif value < metric['thresholds']['target']: | |
status = "📈 Aceptable" | |
color = "off" | |
else: | |
status = "✅ Óptimo" | |
color = "normal" | |
st.metric( | |
metric['label'], | |
f"{value:.2f}", | |
f"{status} (Meta: {metric['thresholds']['target']:.2f})", | |
delta_color=color, | |
help=metric['help'] | |
) | |
st.markdown("<div style='margin-bottom: 0.5rem;'></div>", unsafe_allow_html=True) | |
except Exception as e: | |
logger.error(f"Error mostrando resultados: {str(e)}") | |
st.error("Error al mostrar los resultados") | |
def display_comparison_results(baseline_metrics, current_metrics): | |
"""Muestra comparación entre línea base y métricas actuales""" | |
# Crear columnas para métricas y gráfico | |
metrics_col, graph_col = st.columns([1, 1.5]) | |
with metrics_col: | |
for dimension in ['vocabulary', 'structure', 'cohesion', 'clarity']: | |
baseline = baseline_metrics[dimension]['normalized_score'] | |
current = current_metrics[dimension]['normalized_score'] | |
delta = current - baseline | |
st.metric( | |
dimension.title(), | |
f"{current:.2f}", | |
f"{delta:+.2f}", | |
delta_color="normal" if delta >= 0 else "inverse" | |
) | |
# Sugerir herramientas de mejora | |
if delta < 0: | |
suggest_improvement_tools(dimension) | |
with graph_col: | |
display_radar_chart_comparison( | |
baseline_metrics, | |
current_metrics | |
) | |
def display_metrics_and_suggestions(metrics, text_type, title, show_suggestions=False): | |
""" | |
Muestra métricas y opcionalmente sugerencias de mejora. | |
Args: | |
metrics: Diccionario con las métricas analizadas | |
text_type: Tipo de texto seleccionado | |
title: Título para las métricas ("Base" o "Iteración") | |
show_suggestions: Booleano para mostrar sugerencias | |
""" | |
try: | |
thresholds = TEXT_TYPES[text_type]['thresholds'] | |
st.markdown(f"### Métricas {title}") | |
for dimension, values in metrics.items(): | |
score = values['normalized_score'] | |
target = thresholds[dimension]['target'] | |
min_val = thresholds[dimension]['min'] | |
# Determinar estado y color | |
if score < min_val: | |
status = "⚠️ Por mejorar" | |
color = "inverse" | |
elif score < target: | |
status = "📈 Aceptable" | |
color = "off" | |
else: | |
status = "✅ Óptimo" | |
color = "normal" | |
# Mostrar métrica | |
st.metric( | |
dimension.title(), | |
f"{score:.2f}", | |
f"{status} (Meta: {target:.2f})", | |
delta_color=color, | |
help=f"Meta: {target:.2f}, Mínimo: {min_val:.2f}" | |
) | |
# Mostrar sugerencias si es necesario | |
if show_suggestions and score < target: | |
suggest_improvement_tools(dimension) | |
# Agregar espacio entre métricas | |
st.markdown("<div style='margin-bottom: 0.5rem;'></div>", unsafe_allow_html=True) | |
except Exception as e: | |
logger.error(f"Error mostrando métricas: {str(e)}") | |
st.error("Error al mostrar métricas") | |
def display_radar_chart(metrics_config, thresholds, baseline_metrics=None): | |
""" | |
Muestra el gráfico radar con los resultados. | |
Args: | |
metrics_config: Configuración actual de métricas | |
thresholds: Umbrales para las métricas | |
baseline_metrics: Métricas de línea base (opcional) | |
""" | |
try: | |
# Preparar datos para el gráfico | |
categories = [m['label'] for m in metrics_config] | |
values_current = [m['value'] for m in metrics_config] | |
min_values = [m['thresholds']['min'] for m in metrics_config] | |
target_values = [m['thresholds']['target'] for m in metrics_config] | |
# Crear y configurar gráfico | |
fig = plt.figure(figsize=(8, 8)) | |
ax = fig.add_subplot(111, projection='polar') | |
# Configurar radar | |
angles = [n / float(len(categories)) * 2 * np.pi for n in range(len(categories))] | |
angles += angles[:1] | |
values_current += values_current[:1] | |
min_values += min_values[:1] | |
target_values += target_values[:1] | |
# Configurar ejes | |
ax.set_xticks(angles[:-1]) | |
ax.set_xticklabels(categories, fontsize=10) | |
circle_ticks = np.arange(0, 1.1, 0.2) | |
ax.set_yticks(circle_ticks) | |
ax.set_yticklabels([f'{tick:.1f}' for tick in circle_ticks], fontsize=8) | |
ax.set_ylim(0, 1) | |
# Dibujar áreas de umbrales | |
ax.plot(angles, min_values, '#e74c3c', linestyle='--', linewidth=1, | |
label='Mínimo', alpha=0.5) | |
ax.plot(angles, target_values, '#2ecc71', linestyle='--', linewidth=1, | |
label='Meta', alpha=0.5) | |
ax.fill_between(angles, target_values, [1]*len(angles), | |
color='#2ecc71', alpha=0.1) | |
ax.fill_between(angles, [0]*len(angles), min_values, | |
color='#e74c3c', alpha=0.1) | |
# Si hay línea base, dibujarla primero | |
if baseline_metrics is not None: | |
values_baseline = [baseline_metrics[m['key']]['normalized_score'] | |
for m in metrics_config] | |
values_baseline += values_baseline[:1] | |
ax.plot(angles, values_baseline, '#888888', linewidth=2, | |
label='Línea base', linestyle='--') | |
ax.fill(angles, values_baseline, '#888888', alpha=0.1) | |
# Dibujar valores actuales | |
label = 'Actual' if baseline_metrics else 'Tu escritura' | |
color = '#3498db' if baseline_metrics else '#3498db' | |
ax.plot(angles, values_current, color, linewidth=2, label=label) | |
ax.fill(angles, values_current, color, alpha=0.2) | |
# Ajustar leyenda | |
legend_handles = [] | |
if baseline_metrics: | |
legend_handles.extend([ | |
plt.Line2D([], [], color='#888888', linestyle='--', | |
label='Línea base'), | |
plt.Line2D([], [], color='#3498db', label='Actual') | |
]) | |
else: | |
legend_handles.extend([ | |
plt.Line2D([], [], color='#3498db', label='Tu escritura') | |
]) | |
legend_handles.extend([ | |
plt.Line2D([], [], color='#e74c3c', linestyle='--', label='Mínimo'), | |
plt.Line2D([], [], color='#2ecc71', linestyle='--', label='Meta') | |
]) | |
ax.legend( | |
handles=legend_handles, | |
loc='upper right', | |
bbox_to_anchor=(1.3, 1.1), | |
fontsize=10, | |
frameon=True, | |
facecolor='white', | |
edgecolor='none', | |
shadow=True | |
) | |
plt.tight_layout() | |
st.pyplot(fig) | |
plt.close() | |
except Exception as e: | |
logger.error(f"Error mostrando gráfico radar: {str(e)}") | |
st.error("Error al mostrar el gráfico") | |
#Funciones auxiliares ################################## | |
def suggest_improvement_tools(dimension): | |
"""Sugiere herramientas basadas en la dimensión""" | |
suggestions = [] | |
for analysis, mapping in ANALYSIS_DIMENSION_MAPPING.items(): | |
if dimension in mapping['primary']: | |
suggestions.extend(mapping['tools']) | |
st.info(f"Herramientas sugeridas para mejorar {dimension}:") | |
for tool in suggestions: | |
st.write(f"- {tool}") | |
def prepare_metrics_config(metrics, text_type='student_essay'): | |
""" | |
Prepara la configuración de métricas en el mismo formato que display_results. | |
Args: | |
metrics: Diccionario con las métricas analizadas | |
text_type: Tipo de texto para los umbrales | |
Returns: | |
list: Lista de configuraciones de métricas | |
""" | |
# Obtener umbrales según el tipo de texto | |
thresholds = TEXT_TYPES[text_type]['thresholds'] | |
# Usar la misma estructura que en display_results | |
return [ | |
{ | |
'label': "Vocabulario", | |
'key': 'vocabulary', | |
'value': metrics['vocabulary']['normalized_score'], | |
'help': "Riqueza y variedad del vocabulario", | |
'thresholds': thresholds['vocabulary'] | |
}, | |
{ | |
'label': "Estructura", | |
'key': 'structure', | |
'value': metrics['structure']['normalized_score'], | |
'help': "Organización y complejidad de oraciones", | |
'thresholds': thresholds['structure'] | |
}, | |
{ | |
'label': "Cohesión", | |
'key': 'cohesion', | |
'value': metrics['cohesion']['normalized_score'], | |
'help': "Conexión y fluidez entre ideas", | |
'thresholds': thresholds['cohesion'] | |
}, | |
{ | |
'label': "Claridad", | |
'key': 'clarity', | |
'value': metrics['clarity']['normalized_score'], | |
'help': "Facilidad de comprensión del texto", | |
'thresholds': thresholds['clarity'] | |
} | |
] | |