search

paninski-lab / deepgraphpose

DeepGraphPose
GNU Lesser General Public License v3.0
32 stars 9 forks source link

Fitting DGP throws ValueError with 2 training frames #9

Open cellistigs opened 3 years ago

cellistigs commented 3 years ago

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 0c5d91b8ef76d04a1438cfb4c7c391fc584d4f84, 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 commented 3 years ago

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 commented 3 years ago

This is the case where the both have labels!

waq1129 commented 3 years ago

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

cellistigs commented 3 years ago

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 commented 3 years ago

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.