This repository contains the code accompanying the paper "Learning Informative Representations of Biomedical Relations with Latent Variable Models", Harshil Shah and Julien Fauqueur, EMNLP SustaiNLP 2020.
Large corpora like PubMed abstracts are assumed to be tagged with entity types. Can you describe how this is done? If this is done by dictionary matching, how many sentences are satisfied when the requirements are 1) each sentence should have at least 2 entities, and 2) one is a gene type and the other is a disease type?
In the make_memmap code description, there is the following statement: '''entities_x : np.memmap The first entity in a sentence. The first column contains the index to the UUIDS; the second contains the index to the entity type.'''
which my understanding is that entities_x.mmap is a 2d array with two columns. While in the readme you describe it as a one-dimensional array. Also, the code is a little bit confusing when you define the shape of mm as num_data, 1, but also use mm[:, 0]=