Add BCB'21 paper "DeepNote-GNN: predicting hospital readmission using clinical notes and patient network"

[pat-jj]

Integrating the implementation of the paper "DeepNote-GNN: Predicting Hospital Readmission using Clinical Notes and Patient Network" into PyHealth.

[pat-jj]

1. I added a base dataset for clinical notes (base_note_dataset.py, mimic3_note.py)
2. I added the corresponding sample dataset (SampleNoteDataset)
3. I added the ‘readmission_prediction_mimic3_note_fn’ task function to use clinical notes for readmission prediction.
4. For the ‘split_by_patient’ function, I added an option to balance the positive/negative samples in the validation/test set.
5. I initialized the DeepNote model, which has a load_graph function now, to create a global graph for clinical notes.

I provided a jupyter notebook to play this pipeline (until model initialization): ‘/PyHealth/test.ipynb’

I have to start preparing for my midterm exam (next Wednesday) now, and will continue this work by then.

[zzachw]

The model looks good to me. I will combine the dataset part with #172 and merge later.