pandas.read_pickle can take buffer object as well, not only str (original) (raw)

File path where the pickled object will be loaded.

read_pickle does not accept google storage URL (in form "gs://bucket-name/path/file.pkl") as input

While this code works fine:
with tf.io.gfile.GFile("gs://bucket-name/path/file.pkl", "rb") as infile: df = pd.read_pickle(infile, compression=None)

Might be reasonable to amend the documentation with this if the above is true for the recent version of pandas.