Missing train/val/test data

[j-rausch]

Thanks for putting up this repo!

I was wondering whether there might be some missing files.
Specifically, when trying to run the first preprocessing step that is described in the readme file, the script cannot load the files train.sml etc. that are specified in pre_process_predicate_classifier.py:
base_path='data/new_data/multi-turn-with-clarification-predicate/' trainx, trainy=read_file(base_path + "train_sml.txt") validx, validy=read_file(base_path + "dev_sml.txt") testx, testy=read_file(base_path + "test_sml.txt")

Are those files perhaps missing, or obtainable elswhere?

[jingjingxupku]

Thanks for your question. The data was uploaded to the "data" folder. you need to change the file path as the code required. I will add more data path details on readme as soon as possible. You can refer to readme file to put the data into appropriate paths and run the code.

[j-rausch]

Thank you very much for the reply and updates to the code!

I am now able to run a larger fraction of the codes, but some seem to still be under development.
For instance, generate_input_file.py crashes for task 3.1/Transformer, because there's no 'label' key in the dictinoary.

Would it perhaps be possible to have some sort of overview about which scripts/tasks have already been checked and are up-to-date?

I was also wondering what knowledge base the entities and subgraphs are sourced from. In the paper, this information was not yet included to preserve anonymity during the review process.

Thanks!

[jingjingxupku]

Thanks for your reply! In the last few days, I updated codes by adding argument "file_path" into running commands. Due to the updates of codes, I re-check all running commands. The checking of task1 and task2 is finished and the checking of task3 is on-going. The whole checking process would not last too much time. I will reply you after checking all running commands.

[jingjingxupku]

Thank you very much for the reply and updates to the code!

I am now able to run a larger fraction of the codes, but some seem to still be under development.
For instance, generate_input_file.py crashes for task 3.1/Transformer, because there's no 'label' key in the dictinoary.

Would it perhaps be possible to have some sort of overview about which scripts/tasks have already been checked and are up-to-date?

I was also wondering what knowledge base the entities and subgraphs are sourced from. In the paper, this information was not yet included to preserve anonymity during the review process.

Thanks!

The checking of commands is finished. Please download the latest code and data for running. The dataset is based on knowledge base Satori. For more details, please refer to the latest paper in this repo. The keyword "predicate" is the target label of task 3.1.

[j-rausch]

Thanks a lot for the update and information!