Code for training the CORE model

[NtaylorOX]

Great work - although I am struggling to tell if there is the code for the pre-training of the CORE model?

I.e. the training/knowledge integration steps described in the paper?

Any help would be much appreciated.

Thanks

[bvanaken]

Hi Niall,

Yes, the code is included in this repo. You would first need to generate the pre-training data from the scripts you can find in the tasks folders adm_dis_match_.... If you want to create a combined pre-training from all the sources, you can combine the generated files with combine_pretrain_tasks.py.

Then you can use the code in outcome_pretraining.py to apply the pre-training to an existing model, e.g. BioBERT.

Hope that helps! Betty