ProtoBox

Before Training

Before training, please do the following.

1. Set specifying paths
2. Download WSD datasets
3. Prepare the training data, development data, and test data

1. Set specifying paths

You can set some important paths by editting the code/config.py. ROOT_DIR is the root directory in your environment. EXP_DIR is the directory where trained models are saved. DATA_DIR is the directory where datasets are stored.

2. Download WSD datasets

Following previous WSD researches, we use this WSD Framework. So, please download the datasets. After that, please move the downloaded datasets to DATA_DIR.

3. Prepare the training data, development data, and test data

We train the model with the instances that contain the hyponyms of the root sense. And we evaluate trained models by three tasks: Word Sense Disambiguation, New Sense Classification, and Hypernym Identification. So, please execute following command to prepare these datasets.

python -m code.prepare_data --root animal.n.01

You can change the root sense by editting the --root option.

Run Training

You can run the training process by the following command.

bash scripts/train.sh

Run Evaluation

You can run the evaluation process by the following command. You change the task by editting the EXPERIMENT value in the scripts/test.sh.

bash scripts/test.sh

Contact

If you have any questions or comments, please email Kohei Oda (s2420017@jaist.ac.jp).