data/my_results/evaluate_gemini.py


1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24

import pandas as pd
import json

# Define file paths for the CSV files
file_path1 = 'data/my_results/gemini_2.0_flash_kg_rag_based_mcq_0.csv'


# Load the CSV files into DataFrames
df1 = pd.read_csv(file_path1)

# Define a function to check if the correct answer is present in the LLM answer
def contains_correct_answer(row):
    try: 
        return row['correct_answer'] == json.loads(row['llm_answer'].replace('```', '').replace('\n', '').replace('json', '').replace('{{', '{').replace('}}', '}').split('}')[0] + '}')['answer']
    except:
        return False

# Apply the function to each row of the DataFrames
df1['is_correct'] = df1.apply(contains_correct_answer, axis=1)

# Calculate the percentage of correct answers
correct_rate1 = df1['is_correct'].mean() * 100
print(f"Correct Answer Rate for {file_path1}: {correct_rate1:.2f}%")