the epochs were created with a version prior to 1.8, and
we never updated the read_epochs code path to convert birthday tuples to datetime objects, and
when the TFR is created from the Epochs, info["subject_info"] is copied over untouched
the TFR writer is still writing tuples because itâs writing to HDF5 (not FIFF) and so the safeguards added in #12720 arenât kicking in.
@virvw Says the epochs and TFR files they sent to me are the exact same ones they are computing ITC from and trying to load, respectively, and when I read in both, I see birthday as a datetime.
Pinging also @larsoner for any possible insight since they were involved in some of these changes.
I went ahead and produced a set of epochs from the same raw data (same participant etc). I performed a quick TFR and the outputs are now readable in Spyder.
The epochs may have originally been created w/ mne-python 1.7
I am going to double check this, and restart the pipeline from there!
Thanks a bunch for these insights.
Let me know if I can help on the resolution of version compatibility issues.
@drammock, do you think read_epochs should be updated to convert birthday tuples to datetime? Should there also be checks in place for reading/writing TFR objects?
@tsbinns yeah it would be good if we could maintain backward compatibility for the old objects we created. So once the data is read, if itâs tuple then convert to date etc.
