haha_2019

Using Fastai library to classify Twitter jokes in Spanish

Code assocaited with 3rd place finish in F1 score.

1. Install conda with conda env create -f environment.yml
2. place twitter data in ./data/all_file.txt
3. source activate fastaiv1_dev
4. jupter notebook in the home directory, go to the LM Train in Notebook and run
5. Put haha_2019_train.csv and haha_2019_test.csv in ./data/ directory
6. Run Finetune LM notebook
7. $cd ./prod/' run $./mult_seed_run_fwd_finetune.sh | tee --append out_fwd_1.txt`
8. run run $./mult_seed_regr_finetune.sh | tee --append out_reg_1.txt Generate Submission entry:
9. Run the Ensemble 20 Seeds select best F1 0610.ipynb Notebook for the classification
10. Run the Ensemble 20 modesl select best MSE 0610.ipynb Notebook for the regression outputs on the test set

Note:

• Data is installed in the same directory in ./data/ directory (but not checked into this repo.)