About TempoQR result on TimeQuestions

[zhenjia2017]

Thank you so much for sharing the experimental results on TimeQuestions. It is super helpful for us! Right now we are conducting some experiments on TimeQuestions and need the full ranked answer list so that we could compute the metrics ourselves. I wonder if you could provide the ranked answer list of TimeQuestions? Thank you so much!

[cmavro]

Hi!

Unfortunately, I have not kept the trained model or the log files. However, I have the commands for reproducing the experiments here, e.g.,

python ./train_qa_model.py --model tempoqr --supervision hard --dataset_name timequestions2 --tkbc_model_file tkbc_timeq3.ckpt --max_epochs 200 --lr 1e-4 --valid_freq 5 --batch_size 8 --valid_batch_size 32 --frozen 0

Let me know if that will work!

[zhenjia2017]

Thanks a lot!
When I run the script, it shows the error message:
hard_supervision_functions.py, line 89, in get_kg_facts_for_datapoint
ent = e['annotation']['tail']
KeyError: 'tail'

I debug and found there is one annotation whose key is none (keys =e['annotation'].keys())

I wonder how to deal with this case? Thank you!

[zhenjia2017]

After removing the item in the data without any triple, the problem was solved :)