Upload 12 files

LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2024 Nancy Otero
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

create_dataset_files.py

@@ -0,0 +1,21 @@
+# assume we are starting with data.json 
+# then we create several files per example:
+# name format as {mae}_{example_num}_question|incorrect_answer|correct_answer
+
+import pandas as pd 
+from tqdm import tqdm
+
+df = pd.read_json('data/data.json')
+
+for i, row in tqdm(df.iterrows()):
+    mae = row['Misconception ID']
+    num = row['Example Number']
+    correct_answer = row['Correct Answer']
+    incorrect_answer = row['Incorrect Answer']
+    question = row['Question'] 
+    with open(f'data/txt_files/{mae}_{num}_correct_answer.txt', 'a') as file:
+        file.write(correct_answer)
+    with open(f'data/txt_files/{mae}_{num}_incorrect_answer.txt', 'a') as file:
+        file.write(incorrect_answer)
+    with open(f'data/txt_files/{mae}_{num}_question.txt', 'a') as file:
+        file.write(question)                

create_exp1_outputs.py

@@ -0,0 +1,55 @@
+import openai
+import pandas as pd
+import pandas as pd
+import json
+import urllib
+import math
+import time
+import random
+import re
+from tqdm import tqdm
+from io import StringIO
+import exp_lib
+
+def experiment_1_trial(data_df, model_name):
+  x = data_df.sample(frac=1)
+  train_df = x.drop_duplicates('Misconception ID')
+  test_df = x.iloc[::-1].drop_duplicates('Misconception ID')
+  test_df = test_df.reset_index()
+  prompt = exp_lib.generate_prompt_test_batch(train_df.to_dict(orient='records'), test_df.to_dict(orient='records'))
+  response = exp_lib.get_gpt4_diagnosis(model_name, prompt)
+  response_df = pd.read_csv(StringIO(response), header=None, names=["test_example", "diagnosis"])
+  test_df["Predicted Diagnosis"] = response_df["diagnosis"].str.strip()
+  test_df["Model"] = model_name
+  return test_df[['Misconception ID', 'Example Number', 'Topic', 'Predicted Diagnosis', 'Model']]
+
+
+def experiment_1(input_file_path, model_name, num_iterations, output_file_path):
+  data_df = pd.read_json(input_file_path)
+  experiment_1_results_list = []
+  for i in tqdm(range(num_iterations)):
+    try:
+      trial_result = experiment_1_trial(data_df, model_name)
+      trial_result['Trial'] = i 
+      experiment_1_results_list.append(trial_result)
+    except Exception as e:
+      print(e)  
+  experiment_1_results_df = pd.concat(experiment_1_results_list)
+  experiment_1_results_df['Correct'] = (experiment_1_results_df['Misconception ID'] == experiment_1_results_df['Predicted Diagnosis'])
+  experiment_1_results_df.to_csv(output_file_path)
+
+
+if __name__ == '__main__':
+  experiment_name = 'experiment_1'
+  input_file_path = 'data/data.json' 
+  model_name = 'gpt-4-turbo'
+  num_iterations = 100
+  output_file_path = f'outputs/{experiment_name}_{model_name}_{num_iterations}iters.csv'
+
+  experiment_1(
+    input_file_path,
+    model_name,
+    num_iterations,
+    output_file_path
+  )
+

create_exp1_results.py

@@ -0,0 +1,32 @@
+import pandas as pd
+
+output_file_path = 'outputs/experiment_1_gpt-4-turbo_100iters.csv'
+outputs_df = pd.read_csv(output_file_path)
+
+print(outputs_df.head())
+
+
+
+def calculate_precision_recall(df):
+    # Calculate precision and recall
+    true_positives = df['Correct'].sum()
+    total_predicted = len(df)
+    total_actual = df['Misconception ID'].nunique()
+    
+    precision = true_positives / total_predicted if total_predicted else 0
+    recall = true_positives / total_actual if total_actual else 0
+    
+    return precision, recall
+
+# Overall precision and recall
+overall_precision, overall_recall = calculate_precision_recall(outputs_df)
+
+# Precision and recall per topic
+topic_precision_recall = outputs_df.groupby('Topic').apply(calculate_precision_recall).apply(pd.Series)
+topic_precision_recall.columns = ['Precision', 'Recall']
+
+# Display results
+print(f"Overall Precision: {overall_precision:.3f}")
+print(f"Overall Recall: {overall_recall:.3f}")
+print("\nPrecision and Recall per Topic:")
+print(topic_precision_recall)

create_exp2_outputs.py

@@ -0,0 +1,75 @@
+import openai
+import pandas as pd
+import pandas as pd
+import json
+import urllib
+import math
+import time
+import random
+import re
+from tqdm import tqdm
+from io import StringIO 
+import exp_lib
+
+
+
+def experiment_2_trial(data_df, model_name):
+  x = data_df.sample(frac=1)
+  train_df = x.drop_duplicates('Misconception ID')
+  test_df = x.iloc[::-1].drop_duplicates('Misconception ID')
+  test_df = test_df.reset_index()
+  topics = [
+       'Ratios and proportional reasoning', 
+       'Number Operations',
+       'Patterns, relationships, and functions', 
+       'Number sense',
+       'Algebraic representations',
+       'Variables, expressions, and operations',
+       'Equations and inequalities',
+       'Properties of number and operations'
+       ]
+  # now, iterate by topic and slice each topic data out of train_df, test_df
+  topic_test_dfs = []
+  for topic in topics:
+    topic_test_df = test_df[test_df['Topic'] == topic].copy()
+    topic_test_df = topic_test_df.reset_index()
+    topic_train_df = train_df[train_df['Topic'] == topic].copy()
+    prompt = exp_lib.generate_prompt_test_batch(topic_train_df.to_dict(orient='records'), topic_test_df.to_dict(orient='records'))
+    response = exp_lib.get_gpt4_diagnosis(model_name, prompt)
+    response_df = pd.read_csv(StringIO(response), header=None, names=["test_example", "diagnosis"])
+    topic_test_df["Predicted Diagnosis"] = response_df["diagnosis"].str.strip()
+    topic_test_df["Model"] = model_name
+    topic_test_dfs.append(topic_test_df)
+
+  topic_test_df2 = pd.concat(topic_test_dfs)  
+  return topic_test_df2[['Misconception ID', 'Example Number', 'Topic', 'Predicted Diagnosis', 'Model']]
+
+
+def experiment_2(input_file_path, model_name, num_iterations, output_file_path):
+  data_df = pd.read_json(input_file_path)
+  experiment_2_results_list = []
+  for i in tqdm(range(num_iterations)):
+    try:
+      trial_result = experiment_2_trial(data_df, model_name)
+      trial_result['Trial'] = i 
+      experiment_2_results_list.append(trial_result)
+    except Exception as e:
+      print(e)  
+  experiment_2_results_df = pd.concat(experiment_2_results_list)
+  experiment_2_results_df['Correct'] = (experiment_2_results_df['Misconception ID'] == experiment_2_results_df['Predicted Diagnosis'])
+  experiment_2_results_df.to_csv(output_file_path)
+
+
+if __name__ == '__main__':
+  experiment_name = 'experiment_2'
+  input_file_path = 'data/data.json' 
+  model_name = 'gpt-4-turbo'
+  num_iterations = 100
+  output_file_path = f'outputs/{experiment_name}_{model_name}_{num_iterations}iters.csv'
+
+  experiment_2(
+    input_file_path,
+    model_name,
+    num_iterations,
+    output_file_path
+  )

create_exp2_results.py

@@ -0,0 +1,32 @@
+import pandas as pd
+
+output_file_path = 'outputs/experiment_2_gpt-4-turbo_100iters.csv'
+outputs_df = pd.read_csv(output_file_path)
+
+print(outputs_df.head())
+
+
+
+def calculate_precision_recall(df):
+    # Calculate precision and recall
+    true_positives = df['Correct'].sum()
+    total_predicted = len(df)
+    total_actual = df['Misconception ID'].nunique()
+    
+    precision = true_positives / total_predicted if total_predicted else 0
+    recall = true_positives / total_actual if total_actual else 0
+    
+    return precision, recall
+
+# Overall precision and recall
+overall_precision, overall_recall = calculate_precision_recall(outputs_df)
+
+# Precision and recall per topic
+topic_precision_recall = outputs_df.groupby('Topic').apply(calculate_precision_recall).apply(pd.Series)
+topic_precision_recall.columns = ['Precision', 'Recall']
+
+# Display results
+print(f"Overall Precision: {overall_precision:.3f}")
+print(f"Overall Recall: {overall_recall:.3f}")
+print("\nPrecision and Recall per Topic:")
+print(topic_precision_recall)

exp_lib.py

@@ -0,0 +1,70 @@
+import openai
+import pandas as pd
+import pandas as pd
+import json
+import urllib
+import math
+import time
+import random
+import re
+from tqdm import tqdm
+from io import StringIO
+
+from tqdm import tqdm
+
+# In-context learning prompt template
+def generate_prompt_test_batch(train_examples, test_examples):
+    prompt = (
+        "You are an expert tutor on middle school math with years of experience understanding students' most common math mistakes. "
+        "You have identified a set of common mistakes called Misconceptions, and you use them to diagnose student's answers to math questions. "
+        "You have also developed a labeled dataset of question items, and diagnosed them with the appropriate misconception ID.\n"
+        "Using the set of misconceptions and the labeled dataset, your task today is to take some items of unlabeled data and provide a diagnosis for each unlabeled item.\n\n"
+        "Here is the list of misconceptions together with a brief description:\n"
+    )
+    # Add training examples
+    for i, example in enumerate(train_examples):
+        prompt += f"""
+Train Example {i+1}
+Question:
+{example['Question']}
+Answer:
+{example['Incorrect Answer']}
+Diagnosis: {example['Misconception ID']}
+Misconception Description: {example['Misconception']}
+Topic of Misconception: {example['Topic']}
+
+"""
+
+
+    prompt += """
+Below are the unlabeled Test Examples. For each Test Example, provide only the most likely Misconception ID for the Test Answer from the provided list.
+Don't write anything else but a sequence of lines of the format $Test_Example_Number, $Misconception_ID
+
+"""
+
+    for i, example in enumerate(test_examples):
+        prompt += f"""
+Test Example {i+1}:
+Question:
+{example['Question']}
+Test Answer:
+{example['Incorrect Answer']}
+
+"""
+
+    return prompt
+
+# GPT-4 API call
+def get_gpt4_diagnosis(model, prompt):
+    response = openai.ChatCompletion.create(
+        model=model,
+        messages=[
+            {"role": "system", "content": "You are a math expert specialized in diagnosing student misconceptions."},
+            {"role": "user", "content": prompt}
+        ],
+        temperature=0.2,
+        max_tokens=2000,
+        frequency_penalty=0.0,
+
+    )
+    return response.choices[0].message['content'].strip()

requirements.txt

@@ -0,0 +1,3 @@
+openai==0.28
+pandas
+tqdm