Files & Links | Description |
---|---|
pred.csv | Result Link of Span Prediction on the test.csv |
model | Our best model's link |
trainfinal.csv | Our training file provided by our TAs. |
devfinal.csv | Initial test file provided by our TAs, but currently we use it as validation dataset. |
test.csv | Our test file provided by our TAs. |
- The trainig script can be found here.
- Divide the trainfinal dataset into training and validation dataset and upload it to cloab notebook.
-
(Finally we are using trainfinal as training and devfinal as validation dataset.) Our model is being saved at "./drive/My Drive/best" so we also need to mount the drive. You can directly access our model
- The Prediction generation script can be found here.
-
Upload test.csv to cloab notebook. Download and directly access our model Save this model at "./drive/My Drive/Submission/best" by mounting your drive. Run all the cells of the colab file.
- The evaluation script can be found here.
-
Upload the actual csv file containg the ground truth labels and the prediction.csv file. Run all the cells in the colab file. The F1-score will be printed at the end.
We use the following libraries:
- ast
- csv
- random
- statistics
- sys
- itertools
- string
- pandas
- sklearn
- spacy
Install the spacy and sklearn along with models as (no need if you are running in colab) :
pip install spacy sklearn
python -m spacy download en_core_web_sm
- Official Toxic Spans Detection Semeval (Task 5) webpage & GitHub Repository.