amin-nejad/mimic-mednli

mimic-mednli

1. Download and compile word2vec:

git clone https://github.com/svn2github/word2vec.git
cd word2vec
make

2. Run preprocessing notebook
3. Run word2vec script
4. Download mednli repository

git clone https://github.com/jgc128/mednli

5. Download and extract the content of the mednli_data.zip from Physionet and save into a directory called ./mednli/data/mednli_1.0
6. Follow the instructions on https://github.com/jgc128/mednli carefully for setting up the environment and some preprocessing
7. Use the word embeddings and the training script provided to train the model

python train_model.py with model_class=PyTorchSimpleModel embeddings_filename=<pickled_embeddings_file>

8. Run the trained model and retrieve accuracy result

python predict.py < data/mednli_1.0/mli_test_v1.jsonl > data/test_input_probabilities.txt