SciKG MIMO

semi-supervised MIMO, a construction of Scientific Knowledge Graph

ENV

torch/0.4.0, python/2.7.14

The versions of python in training and predicting shoud be the same!!! (the pre-trained model was in python2, so please run extractor with python2, or you will get wrong prediction.)

Quick Start: RUN Extractor

mkdir predictions

python MIMO_Extractor.py --cuda --udata ./self_train/udata/stmts-demo-unlabeled-small.tsv --out_file ./predictions/stmts-demo-small-prediction --language_model ./models/LM/model.pt --wordembed ./models/WE/pubmed-vectors=50.bin

TRAIN MODEL

mkdir models mkdie results

Supervised MIMO (single featrue with multi-input gates):

an example:

python train.py --cuda --config 000111000 --language_model ./models/LM/model.pt --wordembed ./models/WE/pubmed-vectors=50.bin

Supervised MIMO (multi-input gates, multi-input ensembles):

an example:

python train_ensemble.py --cuda --config 111 --language_model ./models/LM/model.pt --wordembed ./models/WE/pubmed-vectors=50.bin

Semi-supervised MIMO (single featrue with multi-input gates)

an example:

python self_train.py --cuda --language_model ./models/LM/model.pt --wordembed ./models/WE/pubmed-vectors=50.bin --check_point ../models/supervised_model_011000000.torch --AR --TC --SH --DEL

Semi-supervised MIMO (multi-input gates, multi-input ensembles):

an example:

python self_train_ensemble.py --cuda --language_model ./models/LM/model.pt --wordembed ./models/WE/pubmed-vectors=50.bin --check_point ../models/ensemble_supervised_model_111.torch --AR --TC --SH --DEL

DOWNLOAD

• The word embedding we use can be found here.

• The pre-trained language model we use can be found here.

• Pre-trained model: For a quick use of the proposed semi-supervised MIMO model, we put the pre-trained model files here.