Fitting DGP throws ValueError with 2 training frames

[cellistigs]

When running the run_dgp_demo.py script with two training frames, fitting DGP with labeled and unlabeled frames throws the following error:
ValueError: all the input array dimensions except for the concatenation axis must match exactly

The error occurs if running in test mode or not:
python "demo/run_dgp_demo.py" --dlcpath "path/to/model" or python "demo/run_dgp_demo.py" --dlcpath "path/to/model" --test

It does not occur if I increase the number of training frames to 4 or greater. I am using a fork of the dgp repo, forked from commit 0c5d91b, running on Ubuntu 18.04.4 LTS with a Tesla V100 GPU on Amazon AWS. I've attached the full stdout and stderr logs below.
DATASET_NAME_raw_data.zip_STATUS (5).json.txt

[waq1129]

Hi

Thanks for this feedback. One question: are these two training frames labeled or not? Like one is labeled but one doesn't have a label? Or both have labels.

[cellistigs]

This is the case where the both have labels!

[waq1129]

This is a bit weird. Could you post the entire error message? I'd like to see where the error is generated.

[cellistigs]

Unfortunately all I have is the output in the file attached above (DATASET_NAME_raw_data.zip_STATUS (5).json.txt)- this separates stdout and stderr, but both should be listed in their entirety there!

[waq1129]

Usually that happens when the size of joint_loc_full is wrong. But it's weird that this happens when both frames are labeled. I think it might be helpful if you could print out the shape of joint_loc_full.