SentClass

Sentence classification on the Sentihood dataset with a couple CRF models.

Requirements

Code is tested with torch 1.0.0, pyro 0.3.0, and torchtext 0.4.0, and spacy 2.0.18.

Data

Data was obtained from jack.

Training

Run the models with the following command:

python main.py --flat-data --model {lstmfinal | crflstmdiag | crflstmlstm | crfemblstm} {--save} {--devid #}

Add the --save option to save checkpoints along with their valid and test performance. Be sure to create the save/{modelname} directory before saving. Models are saved based on validation accuracy.

Use the --devid # option to utilize a GPU.