import os import pandas as pd import shutil from random import shuffle import argparse import json DATA_DIR = "./data" DATASET_PRE_DIR = "./predata" DATASET_DIR = "./final" CHUNKSIZE = 10000 def generate_pre_dataset_src(course): """ Generate an ASSIST-like data set that can be used by the PRE model, remove the test questions with a score of 0, and do not divide and label the data set """ print(f"generate sourse dataset of course {course}") # create folders for course-related information course_dir = DATASET_PRE_DIR + f"{course}" if not os.path.exists(course_dir): os.mkdir(course_dir) print(f"make dir {course_dir}") submission_csv = DATA_DIR+f"{course}/response_log.csv" # output files dataset_csv = course_dir + "/dataset.csv" q_csv = course_dir + "/q.csv" # read knowledge concept information related files problem_knowledge_json = DATA_DIR+f"problem_knowledge.json" psp_full_score_json = DATA_DIR+f"psp_full_score.json" psp_problem_json = DATA_DIR+f"psp_problem.json" # read json dict with open(problem_knowledge_json,'r') as f: knowledge_dict = json.load(f) with open(psp_full_score_json,'r') as f: full_score_dict = json.load(f) with open(psp_problem_json,'r') as f: problem_dict = json.load(f) q_matrix = {} # generate response logs df_chunk = pd.read_csv(submission_csv, dtype=object, chunksize=CHUNKSIZE) total = 0 drop = 0 for chunk in df_chunk: content = [] for index, row in chunk.iterrows(): total += 1 user_id = row["user_id"] psp_id = row["problem_set_problem_id"] p_id = problem_dict[psp_id] submission_id = row["submission_id"] # calculating scoring rate final_score = 0 status = row["status"] score = float(row["score"]) full_score = float(full_score_dict[psp_id]) if status == "ACCEPTED": final_score = 1 elif status == "WRONG_ANSWER": final_score = 0 else: if full_score == 0: if float(score) == 0: pass else: drop+=1 print(f"problem_set_problem_id/submission_id: {psp_id}/{submission_id}") continue else: final_score = float(score)/float(full_score) # Q-matrix if p_id not in q_matrix.keys(): knowledge_id = knowledge_dict[p_id] q_matrix[p_id] = knowledge_id row = [user_id, p_id, final_score] content.append(row) # write problem info df = pd.DataFrame(content, columns=["user_id", "problem_id", "score"]) flag = True if os.path.exists(dataset_csv): flag = False df.to_csv(dataset_csv, mode='a', index=False, header=flag) print(f"till this chunk drop/total: {drop}/{total}") # generate Q-matrix content = [] for i in q_matrix.keys(): row = [i, q_matrix[i]] content.append(row) df = pd.DataFrame(content, columns=["item_id", "knowledge_code"]) df.to_csv(q_csv, index=False) print(f"generate dataset in {dataset_csv}, q_matrix in {q_csv} with {len(content)} items") def generate_final_dataset_with_index(course): """ Convert the data set of non-indexed values into indexed ones, and output them into the final data set """ print(f"generate final dataset with index of course {course}") # file paths dir_path = DATASET_DIR + f'{course}' if not os.path.exists(dir_path): os.makedirs(dir_path) # output files dst_dataset_path = dir_path + "/dataset.csv" dst_info_path = dir_path + "/info.json" # data source file dataset_file = DATASET_PRE_DIR + f"{course}/dataset.csv" info_dir = DATASET_PRE_DIR + f"{course}/info/" info_path = info_dir+"info.json" problem_list_file = info_dir+"problem_list.txt" student_list_file = info_dir+"student_list.txt" # read files with open(student_list_file, "r") as f: data = f.read() student_list = data.strip("\n").split(", ") with open(problem_list_file, "r") as f: data = f.read() problem_list = data.strip("\n").split(", ") dataset_data = pd.read_csv(dataset_file, dtype=object) # index mapping # replace item_id in dataset for i, e in enumerate(problem_list): dataset_data["problem_id"].replace(e, problem_list.index(e), inplace=True) if i % 100 == 0: print(f"replace item_id {i}/{len(problem_list)}") print(f"replace item_id({len(problem_list)}) in dataset") # replace user_id in dataset for i, e in enumerate(student_list): dataset_data["user_id"].replace(e, student_list.index(e), inplace=True) if i % 100 == 0: print(f"replace user_id {i}/{len(student_list)}") print(f"replace user_id({len(student_list)}) in dataset") dataset_data = dataset_data.rename(columns={"problem_id":"item_id"}) dataset_data.to_csv(dst_dataset_path, index=False) shutil.copyfile(info_path, dst_info_path) print(f"save files in {dir_path}") def generate_final_item_with_index(course): """ Convert the q matrix of non-indexed values into indexed ones, and output them to the final data set """ print(f"generate final q-matrix with index of course {course}") # file paths dir_path = DATASET_DIR + f'{course}' if not os.path.exists(dir_path): os.makedirs(dir_path) # output files dst_item_path = dir_path + "/item.csv" # info files info_dir = DATASET_PRE_DIR + f"{course}/info/" concept_list_file = info_dir+"concept_list.txt" problem_list_file = info_dir+"problem_list.txt" item_csv = DATASET_PRE_DIR + f"{course}/q.csv" # read files with open(concept_list_file, "r") as f: data = f.read() concept_list = data.strip("\n").split(", ") with open(problem_list_file, "r") as f: data = f.read() problem_list = data.strip("\n").split(", ") item_data = pd.read_csv(item_csv, dtype=object) # index mapping # replace knowledge_id in dataset for i in range(len(item_data)): knowledges = item_data.iloc[i]["knowledge_code"].strip("[").strip("]").split(", ") res = [] for j in knowledges: j = j.strip("\'") res.append(concept_list.index(j)) item_data.iloc[i]["knowledge_code"] = res # replace knowledge_id in q-matrix for i, e in enumerate(problem_list): item_data["item_id"].replace(e, problem_list.index(e), inplace=True) if i % 100 == 0: print(f"replace item_id {i}/{len(problem_list)}") print(f"replace item_id({len(problem_list)}) in q-matrix") item_data.to_csv(dst_item_path, index=False) print(f"save files in {dir_path}") def split_dataset_src(course): """ Split the dataset by train/valid/test set 7:1:2 """ print(f"split dataset of course {course}") # data source dataset_file = DATASET_DIR + f"{course}/dataset.csv" # output files train_file = DATASET_DIR + f"{course}/train.csv" test_file = DATASET_DIR + f"{course}/test.csv" valid_file = DATASET_DIR + f"{course}/valid.csv" # create new csv for i in [train_file, test_file, valid_file]: if not os.path.exists(i): content = [] df = pd.DataFrame(content, columns=['user_id', 'item_id','score']) df.to_csv(i, index=False) dataset_df = pd.read_csv(dataset_file) user_group = dataset_df.groupby(["user_id"]) indices = user_group.indices train_cnt = 0 test_cnt = 0 valid_cnt = 0 cnt = 0 total = len(user_group.groups) for group in user_group.groups: cnt += 1 index_list = indices[group] length = len(index_list) shuffle(index_list) # split data if cnt%2==0: train = index_list[:int(7*length/10)+1] test = index_list[int(4*length/5)+1:] valid = index_list[int(7*length/10)+1:int(4*length/5)+1] else: train = index_list[:int(7*length/10)] test = index_list[int(4*length/5):] valid = index_list[int(7*length/10):int(4*length/5)] train_cnt += len(train) test_cnt += len(test) valid_cnt += len(valid) # write according files train_df = dataset_df.iloc[train] train_df.to_csv(train_file, mode='a', header=False, index=False) test_df = dataset_df.iloc[test] test_df.to_csv(test_file, mode='a', header=False, index=False) valid_df = dataset_df.iloc[valid] valid_df.to_csv(valid_file, mode='a', header=False, index=False) print(f"for user {group} train/valid/test count: {len(train)}/{len(valid)}/{len(test)}") if cnt % 100 ==0: print(f"student cnt/total: {cnt}/{total}") print(f"for {course} train/valid/test count: {train_cnt}/{valid_cnt}/{test_cnt}") if __name__ == '__main__': parser = argparse.ArgumentParser() parser.add_argument('--course', '-c', type=str, default="Java_programming") args = parser.parse_args() course = args.course generate_pre_dataset_src(course) generate_final_dataset_with_index(course) generate_final_item_with_index(course) split_dataset_src(course)