存储包含数据帧的字典的最有效方法

Question

我有一本包含数据框的字典。

dictionary = {"key1": df1,
              "key2": df2, and so on...}

很少有 Whosebug 帖子和 reddit 建议 Json 模块和 pickle 模块。

什么是最有效的方法，为什么？

当我将小字典转换为 pickle 时，它的内存小于 0kb，并且呈现 EOFError: Ran out of input，这在此处进行了解释 Why do I get "Pickle - EOFError: Ran out of input" reading an empty file?

Answer 1

如果您喜欢紧凑的文件格式，我建议您使用 pickle。

# import packages
import pandas as pd
import numpy as np
import pickle
import os

# create dictionary of dataframes
nrows, ncols, ndataframes = 1_000, 50, 100
my_dict = {k:v for (k,v) in [[f'df_{n}', pd.DataFrame(np.random.rand(nrows, ncols))] for n in range(ndataframes)]}

# save dictionary as pickle file
pickle_out = open('my_dict.pickle', 'wb')
pickle.dump(my_dict, pickle_out)
pickle_out.close()

# create new dictionary from pickle file
pickle_in = open('my_dict.pickle', 'rb')
new_dict = pickle.load(pickle_in)

# print file size
print('File size pickle file is', round(os.path.getsize('my_dict.pickle') / (1024**2), 1), 'MB')

# sample
new_dict['df_10'].iloc[:5, :5]

结果：

File size pickle file is 38.2 MB

          0         1         2         3         4
0  0.338838  0.501158  0.406240  0.693233  0.567305
1  0.092142  0.569312  0.952694  0.083705  0.006950
2  0.684314  0.373091  0.550300  0.391419  0.877889
3  0.117929  0.597653  0.726894  0.763094  0.466603
4  0.530755  0.472033  0.553457  0.863435  0.906389

Answer 2

另一种选择可能是 HDFStore，它是一个类似 dict 的对象，它使用高性能 HDF5 格式读写 pandas，更多细节在这里：http://pandas-docs.github.io/pandas-docs-travis/user_guide/io.html#hdf5-pytables

存储包含数据帧的字典的最有效方法

Most efficient way of storing dictionary consisting dataframes

python

pickle

pandas