熊猫无法读取用 h5py 创建的 hdf5 文件

Question

I get pandas error when I try to read HDF5 format files that I have created with h5py. I wonder if I am just doing something wrong?

import h5py
import numpy as np
import pandas as pd
h5_file = h5py.File('test.h5', 'w')
h5_file.create_dataset('zeros', data=np.zeros(shape=(3, 5)), dtype='f')
h5_file.close()
pd_file = pd.read_hdf('test.h5', 'zeros')

gives an error: TypeError: cannot create a storer if the object is not existing nor a value are passed

I tried to specify key set to '/zeros' (as I would do it with h5py when reading the file) with no luck.

If I use pandas.HDFStore to read it in, I get an empty store back:

store = pd.HDFStore('test.h5')
>>> store
<class 'pandas.io.pytables.HDFStore'>
File path: test.h5
Empty

I have no troubles reading just created file back with h5py:

h5_back = h5py.File('test.h5', 'r')
h5_back['/zeros']
<HDF5 dataset "zeros": shape (3, 5), type "<f4">

Using these versions:

Python 3.4.3 (v3.4.3:9b73f1c3e601, Feb 23 2015, 02:52:03) 
[GCC 4.2.1 (Apple Inc. build 5666) (dot 3)] on darwin

pd.__version__
'0.16.2'
h5py.__version__
'2.5.0'

Many thanks in advance, Masha

Answer 1

I've worked a little on the pytables module in pandas.io and from what I know pandas interaction with HDF files is limited to specific structures that pandas understands. To see what these look like, you can try

import pandas as pd
import numpy as np
pd.Series(np.zeros((3,5),dtype=np.float32).to_hdf('test.h5','test')

If you open 'test.h5' in HDFView , you will see a path /test with 4 items that are needed to recreate the DataFrame .

So I think your only option for reading in NumPy arrays is to read them in directly and then convert these to Pandas objects.