print_channelgroup and write_all fail with `IndexError: invalid map<K, T> key`

[wildmanwizard]

This includes running the example scripts (with my own data, of course) - the only difference is having to unzip the *.tdm first in order to read it as it was encoded (perhaps a nice-to-have built-in feature).

I'm mostly looking to see if there's index data that might be written by the group functions, as otherwise I seem to be missing index data (which appears to have a pointer named root in jack.get_channelgroup_info(...), but has no native method to access it I can see).

[wildmanwizard]

OK, found the index data >.< Data-dependent issue. Top-line issue still exists.

[mario-fink]

Hi @wildmanwizard, first of all thanks for the issue.

Regarding your feature-request: If we support extracting zipped *.tdm files, we should consequently support a whole bunch of the most common compression algorithms. Furthermore, the data files may be compressed separately or in a single archive, including subdirectories or with flat directory structure, there may be multiple pairs of *.tdm, *.tdx datasets or a single pair, there may be different naming conventions to match tdm - tdx pairs regarding same or different prefixes, etc. . Henceforth, this adds a lot of complexity to the process of providing the correct datafiles. In contrast the user may simply provide matching data files while keeping the generic approach of the package.

My usual question is: Are you able to legitimately share your data files or at least a minimal stripped-down version of it in order to reproduce the issue?

The error that you encounter usually refers to a non-existing or invalid channel id. In any case, the corresponding exception should be caught at python level and transformed into a more meaningful error message. I'll have a look into it....

[wildmanwizard]

It's simple enough on user side to use ZipFile to unzip the tdm file and then point termite towards the output 'header.xml' file, but in any case a flag in the docs that this may be necessary would probably be enough if there are different protocols.

It seems the headline error happens mostly in categorical columns where there is a map from numbers to labels.