Migrate from pyautogen to ag2 Library

06장/utils.py

@@ -0,0 +1,65 @@
+import json
+import pandas as pd
+
+
+def change_jsonl_to_csv(input_file, output_file, prompt_column="prompt", response_column="response"):
+    prompts = []
+    responses = []
+    with open(input_file, 'r') as json_file:
+        for data in json_file:
+            prompts.append(json.loads(data)[0]['messages'][0]['content'])
+            responses.append(json.loads(data)[1]['choices'][0]['message']['content'])
+
+    df = pd.DataFrame({prompt_column: prompts, response_column: responses})
+    df.to_csv(output_file, index=False)
+    return df
+
+
+def merge_gt_and_gen_result(df_gt, df_gen):
+    results = []
+    for idx, row in df_gen.iterrows():
+        with_sql_gt = df_gt.loc[df_gt['without_sql'] == row['without_sql']] 
+        gt_sql = with_sql_gt['sql'].values[0]
+        gen_sql = row['gen_sql']
+        results.append((with_sql_gt['ddl'].values[0], with_sql_gt['request'].values[0], gt_sql, gen_sql))
+    df_result = pd.DataFrame(results, columns=["ddl", "request", "gt_sql", "gen_sql"])
+    return df_result
+
+def make_evaluation_requests(df, filename, model="gpt-4-1106-preview"):
+    prompts = []
+    for idx, row in df.iterrows():
+        prompts.append(f"""Based on provided ddl, request, gen_sql, ground_truth_sql if gen_sql eqauls to ground_truth_sql, output "yes" else "no"
+DDL:
+{row['ddl']}
+Request:
+{row['request']}
+ground_truth_sql:
+{row['gt_sql']}
+gen_sql:
+{row['gen_sql']}
+
+Answer:""")
+
+    jobs = [{"model": model, "messages": [{"role": "system", "content": prompt}]} for prompt in prompts]
+    with open(filename, "w") as f:
+        for job in jobs:
+            json_string = json.dumps(job)
+            f.write(json_string + "\n")
+
+def make_prompt(ddl, request, sql=""):
+    prompt = f"""당신은 SQL을 생성하는 SQL 봇입니다. DDL과 요청사항을 바탕으로 적절한 SQL 쿼리를 생성하세요.
+
+DDL:
+{ddl}
+
+요청사항:
+{request}
+
+SQL:
+{sql}"""
+    return prompt
+
+
+if __name__ == '__main__':
+    df = pd.read_csv('./nl2sql_validation.csv')
+    df.sample(100).to_csv('nl2sql_validation_sample.csv', index=False)