Python 如何在循环数据帧行期间使用多处理收集结果

单线程代码运行良好，保存了3个csvfiles并包含内容

def func(df:DataFrame, file_name:str):
    pbar = tqdm(len(df))
    res = list()
    for row in df.itertuples():
        for path in random.sample(path_list, 8):
            # DO some opration, the following code is simplify edition
            A = random.sample(os.listdir(row[3]), 2)
            C = "abc"
            cur_files = os.listdir(path)
            random.shuffle(cur_files)
            for each in cur_files:
                B.append(each)
                if len(B) == 2: break
            for i in range(2):
                res.append((row[2], A[i], B[i], C))

        pbar.update(1)
    
    data = pd.DataFrame(res, columns["name", "A", "B", "C"])
    data = data.sample(frac=1).reset_index(drop=True)
    data.to_csv(file_name)

if __name__ == "__main__":
    path_list = os.listdir(....)
    train_data = pd.DataFrame(...)
    val_data = pd.DataFrame(...)
    test_data = pd.DataFrame(...)
    for df, tp in [(train_data, "train.csv"), (val_data, "val.csv"), (test_data, "test.csv")]:
        func(df, tp)

我在循环三个数据集期间使用了多处理，这段代码运行良好，保存了三个csvfiles并包含内容

def func(df:DataFrame, file_name:str):
    pbar = tqdm(len(df))
    res = list()
    for row in df.itertuples():
        for path in random.sample(path_list, 8):
            # DO some opration, the following code is simplify edition
            A = random.sample(os.listdir(row[3]), 2)
            C = "abc"
            cur_files = os.listdir(path)
            random.shuffle(cur_files)
            for each in cur_files:
                B.append(each)
                if len(B) == 2: break
            for i in range(2):
                res.append((row[2], A[i], B[i], C))

        pbar.update(1)
    
    data = pd.DataFrame(res, columns["name", "A", "B", "C"])
    data = data.sample(frac=1).reset_index(drop=True)
    data.to_csv(file_name)

if __name__ == "__main__":
    path_list = os.listdir(....)
    train_data = pd.DataFrame(...)
    val_data = pd.DataFrame(...)
    test_data = pd.DataFrame(...)
    for df, tp in [(train_data, "train.csv"), (val_data, "val.csv"), (test_data, "test.csv")]:
        func(df, tp)

如果名称=“\uuuuu main\uuuuuuuu”：
path_list=os.listdir（..）
列车数据=pd.数据帧（…）
val_data=pd.DataFrame（…）
测试数据=局部数据帧（…）
cpuCount=multiprocessing.cpu\u count（）
将multiprocessing.Pool（cpuCount）作为池：
对于df，tp在[（train_数据，“train.csv”），（val_数据，“val.csv”），（test_数据，“test.csv”）]中：
pool.apply_async（func，（df，tp））
pool.close（）
pool.join（）
打印（“子流程全部完成”）

但是速度很慢，因为train_数据有很多行，所以我想在func方法中多处理行， 以下代码不起作用，没有错误并立即完成，保存了3个CSV文件，但CSV文件内容只有标题

def func(row, pbar, res):
    for path in random.sample(path_list, 8):
        # DO some opration, the following code is simplify edition
        A = random.sample(os.listdir(row[3]), 2)
        C = "abc"
        cur_files = os.listdir(path)
        random.shuffle(cur_files)
        for each in cur_files:
            B.append(each)
            if len(B) == 2: break
        for i in range(2):
            res.append((row[2], A[i], B[i], C))

    pbar.update(1)

if __name__ == "__main__":
    path_list = os.listdir(....)
    train_data = pd.DataFrame(...)
    val_data = pd.DataFrame(...)
    test_data = pd.DataFrame(...)
    cpuCount = multiprocessing.cpu_count()
    with multiprocessing.Pool(cpuCount) as pool:
        for df, tp in [(train_data, "train.csv"), (val_data, "val.csv"), (test_data, "test.csv")]:
            pbar = tqdm(len(df))
            res = list()
            for row in df.itertuples():
                pool.apply_async(func, (row, pbar, res))
            data = pd.DataFrame(res, columns["name", "A", "B", "C"])
            data = data.sample(frac=1).reset_index(drop=True)
            data.to_csv(file_name)]
        pool.close()
        pool.join()
   print("Sub-process all done.")
        

如何优化最后的代码或更好的方法