Mahynlo
commited on
Commit
·
0abe794
1
Parent(s):
29ccd55
Soporte para archivos Excel y audio + mejoras en procesamiento de archivos
Browse files- agents.py +24 -7
- requirements.txt +1 -1
- test_answers.py +58 -0
- tools.py +60 -0
agents.py
CHANGED
|
@@ -163,14 +163,15 @@ IMPORTANT: GAIA uses exact string matching. Be precise!"""
|
|
| 163 |
if files and len(files) > 0:
|
| 164 |
context_parts.append("\n📁 FILES PROVIDED:")
|
| 165 |
|
| 166 |
-
from tools import read_image_text
|
| 167 |
import requests
|
| 168 |
|
| 169 |
for file_url in files:
|
| 170 |
try:
|
| 171 |
-
|
| 172 |
-
|
| 173 |
-
|
|
|
|
| 174 |
if self.verbose:
|
| 175 |
print(f" 📷 Procesando imagen: {file_url}")
|
| 176 |
text = read_image_text(file_url)
|
|
@@ -180,8 +181,24 @@ IMPORTANT: GAIA uses exact string matching. Be precise!"""
|
|
| 180 |
else:
|
| 181 |
context_parts.append(f"\n⚠️ Could not extract text from image: {file_url}")
|
| 182 |
|
| 183 |
-
|
| 184 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 185 |
if self.verbose:
|
| 186 |
print(f" 📄 Procesando archivo texto: {file_url}")
|
| 187 |
response = requests.get(file_url, timeout=30)
|
|
@@ -191,7 +208,7 @@ IMPORTANT: GAIA uses exact string matching. Be precise!"""
|
|
| 191 |
context_parts.append(f"{content}")
|
| 192 |
|
| 193 |
else:
|
| 194 |
-
#
|
| 195 |
context_parts.append(f"\n📎 File available: {file_url}")
|
| 196 |
|
| 197 |
except Exception as e:
|
|
|
|
| 163 |
if files and len(files) > 0:
|
| 164 |
context_parts.append("\n📁 FILES PROVIDED:")
|
| 165 |
|
| 166 |
+
from tools import read_image_text, read_excel_file, read_audio_file
|
| 167 |
import requests
|
| 168 |
|
| 169 |
for file_url in files:
|
| 170 |
try:
|
| 171 |
+
file_lower = file_url.lower()
|
| 172 |
+
|
| 173 |
+
# Procesar imágenes con OCR
|
| 174 |
+
if any(ext in file_lower for ext in ['.jpg', '.jpeg', '.png', '.gif', '.bmp']):
|
| 175 |
if self.verbose:
|
| 176 |
print(f" 📷 Procesando imagen: {file_url}")
|
| 177 |
text = read_image_text(file_url)
|
|
|
|
| 181 |
else:
|
| 182 |
context_parts.append(f"\n⚠️ Could not extract text from image: {file_url}")
|
| 183 |
|
| 184 |
+
# Procesar archivos Excel
|
| 185 |
+
elif any(ext in file_lower for ext in ['.xlsx', '.xls']):
|
| 186 |
+
if self.verbose:
|
| 187 |
+
print(f" 📊 Procesando Excel: {file_url}")
|
| 188 |
+
content = read_excel_file(file_url)
|
| 189 |
+
context_parts.append(f"\n📊 EXCEL DATA from {file_url}:")
|
| 190 |
+
context_parts.append(content)
|
| 191 |
+
|
| 192 |
+
# Procesar archivos de audio (limitado)
|
| 193 |
+
elif any(ext in file_lower for ext in ['.mp3', '.wav', '.ogg', '.m4a']):
|
| 194 |
+
if self.verbose:
|
| 195 |
+
print(f" 🎵 Detectado audio: {file_url}")
|
| 196 |
+
info = read_audio_file(file_url)
|
| 197 |
+
context_parts.append(f"\n🎵 AUDIO FILE:")
|
| 198 |
+
context_parts.append(info)
|
| 199 |
+
|
| 200 |
+
# Procesar archivos de texto
|
| 201 |
+
elif any(ext in file_lower for ext in ['.txt', '.csv', '.json', '.py', '.md']):
|
| 202 |
if self.verbose:
|
| 203 |
print(f" 📄 Procesando archivo texto: {file_url}")
|
| 204 |
response = requests.get(file_url, timeout=30)
|
|
|
|
| 208 |
context_parts.append(f"{content}")
|
| 209 |
|
| 210 |
else:
|
| 211 |
+
# Archivo de tipo desconocido
|
| 212 |
context_parts.append(f"\n📎 File available: {file_url}")
|
| 213 |
|
| 214 |
except Exception as e:
|
requirements.txt
CHANGED
|
@@ -7,6 +7,7 @@ google-generativeai>=0.8.0
|
|
| 7 |
|
| 8 |
# Procesamiento de datos
|
| 9 |
pandas>=2.0.0
|
|
|
|
| 10 |
requests>=2.31.0
|
| 11 |
|
| 12 |
# Procesamiento de imágenes (OCR)
|
|
@@ -15,4 +16,3 @@ pytesseract>=0.3.10
|
|
| 15 |
|
| 16 |
# Variables de entorno
|
| 17 |
python-dotenv>=1.0.0
|
| 18 |
-
|
|
|
|
| 7 |
|
| 8 |
# Procesamiento de datos
|
| 9 |
pandas>=2.0.0
|
| 10 |
+
openpyxl>=3.1.0 # Para leer archivos Excel
|
| 11 |
requests>=2.31.0
|
| 12 |
|
| 13 |
# Procesamiento de imágenes (OCR)
|
|
|
|
| 16 |
|
| 17 |
# Variables de entorno
|
| 18 |
python-dotenv>=1.0.0
|
|
|
test_answers.py
ADDED
|
@@ -0,0 +1,58 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
"""
|
| 2 |
+
Script para testear respuestas específicas y mejorar la precisión
|
| 3 |
+
"""
|
| 4 |
+
|
| 5 |
+
# Respuestas del último run (15%)
|
| 6 |
+
respuestas_obtenidas = {
|
| 7 |
+
1: "4", # Mercedes Sosa albums
|
| 8 |
+
2: "4", # Bird species video
|
| 9 |
+
3: "right", # Reversed text ✅ (probablemente correcta)
|
| 10 |
+
4: "Ra1#", # Chess move
|
| 11 |
+
5: "funkmonk", # Wikipedia dinosaur
|
| 12 |
+
6: "b,e", # Math table ✅ (probablemente correcta)
|
| 13 |
+
7: "indeed", # Teal'c quote ✅ (probablemente correcta)
|
| 14 |
+
8: "silver", # Equine veterinarian
|
| 15 |
+
9: "bell pepper,broccoli,celery,corn,green beans,lettuce,sweet potatoes,zucchini", # Botany list
|
| 16 |
+
10: "cornstarch,lemon juice,salt,strawberries,sugar", # Pie ingredients
|
| 17 |
+
11: "wojciech", # Polish actor
|
| 18 |
+
12: "55", # Python code output
|
| 19 |
+
13: "553", # Baseball at bats
|
| 20 |
+
14: "21,22,23,24,25,26,27,28,29,30,31,32,33,34,35", # Homework pages
|
| 21 |
+
15: "80GSFC21N0002", # NASA article
|
| 22 |
+
16: "None", # Vietnamese specimens
|
| 23 |
+
17: "HAI", # 1928 Olympics
|
| 24 |
+
18: "Fujiwara,Ogasawara", # Baseball pitchers
|
| 25 |
+
19: "$210.00", # Excel sales
|
| 26 |
+
20: "dmitrij", # Malko Competition
|
| 27 |
+
}
|
| 28 |
+
|
| 29 |
+
# Patrones observados de posibles problemas:
|
| 30 |
+
problemas_potenciales = {
|
| 31 |
+
"Mayúsculas/minúsculas": [5, 7, 11, 17, 18, 20], # funkmonk vs FunkMonk, indeed vs Indeed
|
| 32 |
+
"Formato numérico": [1, 12, 13], # Números que podrían estar incorrectos
|
| 33 |
+
"Listas de items": [9, 10, 14, 18], # Listas que podrían tener orden o items incorrectos
|
| 34 |
+
"Formato especial": [15, 19], # Códigos NASA, moneda
|
| 35 |
+
"Análisis de archivos": [8, 12, 14, 19, 20], # Requieren leer archivos
|
| 36 |
+
}
|
| 37 |
+
|
| 38 |
+
print("=" * 60)
|
| 39 |
+
print("🔍 ANÁLISIS DE RESPUESTAS DEL ÚLTIMO RUN")
|
| 40 |
+
print("=" * 60)
|
| 41 |
+
print(f"\n✅ Correctas confirmadas: 3/20 (15%)")
|
| 42 |
+
print(f"🎯 Objetivo: 6/20 (30%)")
|
| 43 |
+
print(f"❌ Necesitamos mejorar: +3 respuestas correctas\n")
|
| 44 |
+
|
| 45 |
+
print("📋 CATEGORÍAS DE PROBLEMAS POTENCIALES:\n")
|
| 46 |
+
|
| 47 |
+
for categoria, preguntas in problemas_potenciales.items():
|
| 48 |
+
print(f" • {categoria}:")
|
| 49 |
+
print(f" Preguntas: {', '.join(map(str, preguntas))}")
|
| 50 |
+
respuestas = [respuestas_obtenidas[p] for p in preguntas]
|
| 51 |
+
print(f" Respuestas: {respuestas[:3]}..." if len(respuestas) > 3 else f" Respuestas: {respuestas}")
|
| 52 |
+
print()
|
| 53 |
+
|
| 54 |
+
print("\n💡 RECOMENDACIONES:")
|
| 55 |
+
print(" 1. Verificar case-sensitivity (mayúsculas/minúsculas)")
|
| 56 |
+
print(" 2. Mejorar procesamiento de archivos (especialmente audio/Excel)")
|
| 57 |
+
print(" 3. Validar formato de listas (orden alfabético vs orden original)")
|
| 58 |
+
print(" 4. Revisar precisión numérica en cálculos")
|
tools.py
CHANGED
|
@@ -20,6 +20,66 @@ def read_image_text(url: str):
|
|
| 20 |
print(f"❌ Error al leer imagen {url}: {e}")
|
| 21 |
return ""
|
| 22 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 23 |
def web_search(query: str):
|
| 24 |
"""Simula búsqueda web (puedes mejorarla con Wikipedia API)."""
|
| 25 |
return f"(Simulación de búsqueda web para '{query}')"
|
|
|
|
| 20 |
print(f"❌ Error al leer imagen {url}: {e}")
|
| 21 |
return ""
|
| 22 |
|
| 23 |
+
|
| 24 |
+
def read_excel_file(url: str):
|
| 25 |
+
"""
|
| 26 |
+
Lee un archivo Excel y lo convierte a texto estructurado.
|
| 27 |
+
Útil para preguntas GAIA que requieren análisis de datos.
|
| 28 |
+
"""
|
| 29 |
+
try:
|
| 30 |
+
import pandas as pd
|
| 31 |
+
|
| 32 |
+
response = requests.get(url, timeout=30)
|
| 33 |
+
response.raise_for_status()
|
| 34 |
+
|
| 35 |
+
# Leer Excel con pandas
|
| 36 |
+
df = pd.read_excel(BytesIO(response.content), sheet_name=None) # Todas las hojas
|
| 37 |
+
|
| 38 |
+
result = []
|
| 39 |
+
for sheet_name, data in df.items():
|
| 40 |
+
result.append(f"\n📊 Sheet: {sheet_name}")
|
| 41 |
+
# Convertir a string con formato tabular (primeras 50 filas)
|
| 42 |
+
result.append(data.head(50).to_string(index=False))
|
| 43 |
+
|
| 44 |
+
# Si hay más filas, indicarlo
|
| 45 |
+
if len(data) > 50:
|
| 46 |
+
result.append(f"\n... ({len(data) - 50} more rows)")
|
| 47 |
+
|
| 48 |
+
return "\n".join(result)
|
| 49 |
+
except ImportError:
|
| 50 |
+
return "ERROR: pandas/openpyxl not installed for Excel processing"
|
| 51 |
+
except Exception as e:
|
| 52 |
+
print(f"❌ Error al leer Excel {url}: {e}")
|
| 53 |
+
return f"ERROR reading Excel: {str(e)}"
|
| 54 |
+
|
| 55 |
+
|
| 56 |
+
def read_audio_file(url: str):
|
| 57 |
+
"""
|
| 58 |
+
Intenta procesar archivos de audio.
|
| 59 |
+
Nota: Transcripción de audio es compleja y requiere APIs externas.
|
| 60 |
+
Por ahora, solo indicamos que el archivo existe.
|
| 61 |
+
"""
|
| 62 |
+
try:
|
| 63 |
+
response = requests.head(url, timeout=10)
|
| 64 |
+
response.raise_for_status()
|
| 65 |
+
|
| 66 |
+
# Obtener información del archivo
|
| 67 |
+
content_type = response.headers.get('Content-Type', 'unknown')
|
| 68 |
+
content_length = response.headers.get('Content-Length', 'unknown')
|
| 69 |
+
|
| 70 |
+
return f"""⚠️ AUDIO FILE DETECTED: {url}
|
| 71 |
+
Content-Type: {content_type}
|
| 72 |
+
Size: {content_length} bytes
|
| 73 |
+
|
| 74 |
+
NOTE: Audio transcription requires external API (e.g., Whisper, Google Speech-to-Text).
|
| 75 |
+
This agent cannot directly process audio content.
|
| 76 |
+
Try to answer based on the question context."""
|
| 77 |
+
|
| 78 |
+
except Exception as e:
|
| 79 |
+
print(f"❌ Error al verificar audio {url}: {e}")
|
| 80 |
+
return f"ERROR accessing audio file: {str(e)}"
|
| 81 |
+
|
| 82 |
+
|
| 83 |
def web_search(query: str):
|
| 84 |
"""Simula búsqueda web (puedes mejorarla con Wikipedia API)."""
|
| 85 |
return f"(Simulación de búsqueda web para '{query}')"
|