kuielab / lg_dacon_competition_template

A hydra integrated template for LG-Dacon Competetion

lg_dacon_competition_template

This repository shows the example of the pytorch-lightning and hydra on LG-Dacon competition.

Setup Environment

Conda

conda env create -f environment.yaml -n lg
conda activate lg

.env file

you must create .env file by copying .env.sample to set environmental variables.

wandb_api_key=[Your Key] # "xxxxxxxxxxxxxxxxxxxxxxxx"
data_dir=[Your Path] # "/home/kuielab/lg_dacon_data_dir"

• about wandb_api_key
  • we currently only support wandb for logging.
  • for wandb_api_key, visit wandb, go to setting, and then copy your api key
• about data_dir
  • the absolute path where datasets are stored

Train script example

python main.py run_name=lg_example 

If you want to change the option of this code, you need to modify the YAML files on the config folder or put the other options in the training script. For example, you want to change the batch size, put options as follows:

python main.py run_name=lg_example dataloader.datasets.batch_size=64

Training result example

https://wandb.ai/ielab/LG_hydra_example/reports/LG_Dacon_Competition--Vmlldzo4NjU2NzI

Acknowledgement

lightning-hydra-template