This folder contains a couple models, described below:

• v4: MCB + Genome
• v4_glove: MCB + Genome + Glove
• multi_att_2_glove: MCB + Genome + Attention + GloVe

Models that use GloVe vectors have additional dependencies. You'll need to install spaCy and download the GloVe vectors:

pip install cymem
pip install thinc
pip install https://github.com/spacy-io/spaCy/zipball/master
python -m spacy.en.download
sputnik --name spacy install en_glove_cc_300_1m_vectors

To train a model, edit the corresponding config.py and qlstm_solver.prototxt files.

In config.py, the TRAIN_DATA_SPLITS, QUESTION_VOCAB_SPACE, and ANSWER_VOCAB_SPACE parameters take a + delimited string of data sources, which are specified in the DATA_PATHS dictionary. We recommend using train+val+genome for training.

In qlstm_solver.prototxt, set snapshot and snapshot_prefix correctly.

Now just run python train_xxx.py. Training can take some time. Snapshots are saved according to the settings in qlstm_solver.prototxt. To stop training, just hit Control + C.