Python pickle: fix \r characters before loading

Question

I got a pickled object (a list with a few numpy arrays in it) that was created on Windows and apparently saved to a file loaded as text, not in binary mode (ie. with open(filename, 'w') instead of open(filename, 'wb')). Result is that now I can't unpickle it (not even on Windows) because it's infected with \r characters (and possibly more)? The main complaint is

ImportError: No module named multiarray

supposedly because it's looking for numpy.core.multiarray\r, which of course doesn't exist. Simply removing the \r characters didn't do the trick (tried both sed -e 's/\r//g' and, in python s = file.read().replace('\r', ''), but both break the file and yield a cPickle.UnpicklingError later on)

Problem is that I really need to get the data out of the objects. Any ideas how to fix the files?

Edit: On request, the first few hundred bytes of my file, Octal:

\x80\x02]q\x01(}q\x02(U\r\ntotal_timeq\x03G?\x90\x15r\xc9(s\x00U\rreaction_timeq\x04NU\x0ejump_directionq\x05cnumpy.core.multiarray\r\nscalar\r\nq\x06cnumpy\r\ndtype\r\nq\x07U\x02f8K\x00K\x01\x87Rq\x08(K\x03U\x01<NNNJ\xff\xff\xff\xffJ\xff\xff\xff\xffK\x00tbU\x08\x025\x9d\x13\xfc#\xc8?\x86Rq\tU\x14normalised_directionq\r\nh\x06h\x08U\x08\xf0\xf9,\x0eA\x18\xf8?\x86Rq\x0bU\rjump_distanceq\x0ch\x06h\x08U\x08\x13\x14\xea&\xb0\x9b\x1a@\x86Rq\rU\x04jumpq\x0ecnumpy.core.multiarray\r\n_reconstruct\r\nq\x0fcnumpy\r\nndarray\r\nq\x10K\x00\x85U\x01b\x87Rq\x11(K\x01K\x02\x85h\x08\x89U\x10\x87\x16\xdaEG\xf4\xf3?\x06`OC\xe7"\x1a@tbU\x0emovement_speedq\x12h\x06h\x08U\x08\\p\xf5[2\xc2\xef?\x86Rq\x13U\x0ctrial_lengthq\x14G@\t\x98\x87\xf8\x1a\xb4\xbaU\tconditionq\x15U\x0bhigh_mentalq\x16U\x07subjectq\x17K\x02U\x12movement_directionq\x18h\x06h\x08U\x08\xde\x06\xcf\x1c50\xfd?\x86Rq\x19U\x08positionq\x1ah\x0fh\x10K\x00\x85U\x01b\x87Rq\x1b(K\x01K\x02\x85h\x08\x89U\x10K\xb7\xb4\x07q=\x1e\xc0\xf2\xc2YI\xb7U&\xc0tbU\x04typeq\x1ch\x0eU\x08movementq\x1dh\x0fh\x10K\x00\x85U\x01b\x87Rq\x1e(K\x01K\x02\x85h\x08\x89U\x10\xad8\x9c9\x10\xb5\xee\xbf\xffa\xa2hWR\xcf?tbu}q\x1f(h\x03G@\t\xba\xbc\xb8\xad\xc8\x14h\x04G?\xd9\x99%]\xadV\x00h\x05h\x06h\x08U\x08\xe3X\xa9=\xc1\xb1\xeb?\x86Rq h\r\nh\x06h\x08U\x08\x88\xf7\xb9\xc1\t\xd6\xff?\x86Rq!h\x0ch\x06h\x08U\x08v\x7f\xeb\x11\xea5\r@\x86Rq"h\x0eh\x0fh\x10K\x00\x85U\x01b\x87Rq#(K\x01K\x02\x85h\x08\x89U\x10\xcd\xd9\x92\x9a\x94=\x06@]C\xaf\xef\xeb\xef\x02@tbh\x12h\x06h\x08U\x08-\x9c&\x185\xfd\xef?\x86Rq$h\x14G@\r\xb8W\xb2`V\xach\x15h\x16h\x17K\x02h\x18h\x06h\x08U\x08\x8e\x87\xd1\xc2

You may also download the whole file (22k).

Answer 1

Presuming that the file was created with the default protocol=0 ASCII-compatible method, you should be able to load it anywhere by using open('pickled_file', 'rU') i.e. universal newlines.

If this doesn't work, show us the first few hundred bytes: print repr(open('pickled_file', 'rb').read(200)) and paste the results into an edit of your question.

Update after file contents were published:

Your file starts with '\x80\x02'; it was dumped with protocol 2, the latest/best. Protocols 1 and 2 are binary protocols. Your file was written in text mode on Windows. This has resulted in each '\n' being converted to '\r\n' by the C runtime. Files should be opened in binary mode like this:

with open('result.pickle', 'wb') as f: # b for binary
    pickle.dump(obj, f, pickle.HIGHEST_PROTOCOL)

with open('result.pickle', 'rb') as f: # b for binary
    obj = pickle.load(f)

Docs are here. This code will work portably on both Windows and non-Windows systems.

You can recover the original pickle image by reading the file in binary mode and then reversing the damage by replacing all occurrences of '\r\n' by '\n'. Note: This recovery procedure is necessary whether you are trying to read it on Windows or not.