IberLEF2023

All requirements were installed in a Python 3.11 environment.

First run the following command to install the requirements:

pip install -r reqs.txt

Furthermore, run the following command to install PyTorch for your GPU:

conda install pytorch torchvision torchaudio pytorch-cuda=11.8 -c pytorch -c nvidia

Change the pytorch-cuda argument according to your CUDA version. You can check your CUDA version with the following command:

nvcc --version

Then, to run the training script on a tiny dataset with the following command (Omit the tiny argument to use the full dataset):

python trainer.py -a 'cpu' -b 2 -w 2 -e 3 -c 'simple' -n "test_logger" -tiny 

Argument description:

• -a: accelerator (cpu or gpu)
• -b: batch size
• -w: number of workers
• -e: number of epochs
• -c: model architecture (simple for simple classifier)
• -tiny: activates use of tiny dataset (for testing purposes)
• -practise: activates use of practise dataset (subset of full dataset)
• -lr: learning rate
• -n: name of the logger
• -hp_path: path to the hyperparameter file/folder

If none of the arguments are specified, the script will run with the following default values:

• -a: cpu
• -b: from yaml file
• -n: 2
• -e: from yaml file
• -c: simple
• -tiny: deactivated
• -practise: deactivated

For inference, run the following command:

python inference.py -cp 'path/to/checkpoint' -tdp 'path/to/test_data' -op 'results/results.csv' -a 'cpu'