Can HDF5 tables be used by different versions of pandas than what they were created on?

Question

I have a pandas dataframe stored in a .h5 file. I wanted to work on the data on my home machine, so I copied it onto my laptop. To my surprise, I was unable to load the dataframe, though the table loaded without incident. Moreover, when I printed the .h5 file on the machine it was made on (with pandas 0.13.0), I got the following result:

File path: wavefunction_file_key_2000.wfn.npy.h5
/wavefunction            frame        (shape->[1478,10001])

But on the other machine (pandas 0.7.0), it was:

File path: wavefunction_file_key_2000.wfn.npy.h5
wavefunction     DataFrame

Is the shape information lost when the file is copied between computers? Or are different versions of HDF5 incompatible?

Answer 1

They should be compatible even across platforms / HDF versions.

You should be able to read even a very old version with a pandas versions >= 0.10.1 (when the newer format stabilized)

However it is not possible to read a newer version (your top print) with a pre-0.10.1 version of pandas (the bottom print).