|
|
import os |
|
|
import io |
|
|
import pandas as pd |
|
|
import requests |
|
|
from openai import OpenAI |
|
|
|
|
|
class GaiaAgent: |
|
|
def __init__(self): |
|
|
self.client = OpenAI(api_key=os.getenv("OPENAI_API_KEY")) |
|
|
self.instructions = ( |
|
|
"You are a reasoning assistant solving GAIA benchmark questions. " |
|
|
"If data is provided, analyze it logically and extract the relevant facts. " |
|
|
"Think step by step. Output only the final answer." |
|
|
) |
|
|
self.api_url = "https://agents-course-unit4-scoring.hf.space" |
|
|
|
|
|
def analyze_csv(self, csv_text: str, question: str) -> str: |
|
|
try: |
|
|
df = pd.read_csv(io.StringIO(csv_text)) |
|
|
question_lower = question.lower() |
|
|
|
|
|
if any(k in question_lower for k in ["lowest", "cheapest", "minimum"]): |
|
|
col = self._detect_column(df, ["price", "cost", "amount"]) |
|
|
if col: |
|
|
row = df.sort_values(by=col).iloc[0].to_dict() |
|
|
return f"Lowest {col}: {row}" |
|
|
elif any(k in question_lower for k in ["highest", "most expensive", "maximum"]): |
|
|
col = self._detect_column(df, ["price", "score", "rating"]) |
|
|
if col: |
|
|
row = df.sort_values(by=col, ascending=False).iloc[0].to_dict() |
|
|
return f"Highest {col}: {row}" |
|
|
elif "how many" in question_lower: |
|
|
return f"Total rows: {len(df)}" |
|
|
|
|
|
sample = df.iloc[0].to_dict() |
|
|
return f"Sample row: {sample}" |
|
|
|
|
|
except Exception as e: |
|
|
return f"[CSV parsing failed: {e}]" |
|
|
|
|
|
def _detect_column(self, df, candidates): |
|
|
for col in df.columns: |
|
|
for name in candidates: |
|
|
if name in col.lower(): |
|
|
return col |
|
|
return None |
|
|
|
|
|
def fetch_file_context(self, task_id: str, question: str) -> str: |
|
|
try: |
|
|
url = f"{self.api_url}/files/{task_id}" |
|
|
response = requests.get(url, timeout=10) |
|
|
response.raise_for_status() |
|
|
content_type = response.headers.get("Content-Type", "") |
|
|
|
|
|
if "csv" in content_type or url.endswith(".csv"): |
|
|
return self.analyze_csv(response.text, question) |
|
|
elif "json" in content_type: |
|
|
return f"JSON Preview: {response.text[:1000]}" |
|
|
elif "text/plain" in content_type: |
|
|
return f"Text Sample: {response.text[:1000]}" |
|
|
elif "pdf" in content_type: |
|
|
return "[PDF detected. OCR not supported.]" |
|
|
else: |
|
|
return f"[Unsupported file type: {content_type}]" |
|
|
|
|
|
except Exception as e: |
|
|
return f"[Error fetching file: {e}]" |
|
|
|
|
|
def __call__(self, question: str, task_id: str = None) -> str: |
|
|
file_fact = "" |
|
|
if task_id: |
|
|
file_fact = self.fetch_file_context(task_id, question) |
|
|
file_fact = f"FILE INSIGHTS:\n{file_fact}\n" |
|
|
|
|
|
prompt = f"{self.instructions}\n\n{file_fact}QUESTION: {question}\nANSWER:" |
|
|
|
|
|
response = self.client.chat.completions.create( |
|
|
model="gpt-4-turbo", |
|
|
messages=[ |
|
|
{"role": "system", "content": self.instructions}, |
|
|
{"role": "user", "content": prompt} |
|
|
], |
|
|
temperature=0.0, |
|
|
) |
|
|
|
|
|
return response.choices[0].message.content.strip() |
|
|
|
