In this extension repository of our base project Question Answering SQUAD we present 4 diferent approaches that we have tried to improve the performances, in terms of F1 score and EM of our model.

To quickly try out our experiments, clone this repository and install the necessary requirements by running

pip install -r requirements.txt

We recommend creating a separate python 3.6 environment.

To the script compute_answers.py, you just need to launch it:

• python3 compute_answers.py --extension name_of_the_extension --test test_dataset --weights path_to_ner_weights

Where:

• --extension with this parameter you can choose the extension you want to try. It can assume only 4 values [ner, multiple_choice, classifier, convolutional_classifier]
• --test this parameter is used to choose for which dataset you want to compute the predictions, so you can pass to it the path to the json file of your dataset, or "squad_v1" or "squad_v2" if you want to compute the predictions for respectively the squad 1.1 dev_set and the squad 2.0 dev_set.

Then if youy want to evaluate the results you can use the evaluate.py script to test the method you have chosen:

• python3 evaluate.py *path_to_ground_truth* *path_to_predictions_file*

We strongly reccomend you to use a GPU for running the train.py and the compute_answers.py scripts. To do so you can use the Nvidia graphic card of your machine, if it has one. In this case make sure that you have all the prerequisites (https://docs.nvidia.com/deeplearning/cudnn/install-guide/index.html#install-windows) and to have installed the pytorch version for the CUDA platform (https://pytorch.org/).

Simone Gayed Said - [email protected]
Alex Rossi - [email protected]
Jia Liang Zhou - [email protected]
Hanying Zhang - [email protected]

Hugging Face library - https://huggingface.co/transformers/
Our organization on the Hugging Face Hub - https://huggingface.co/nlpunibo