Airkorea PM dataset: 2014.zip 2015.zip 2016.zip 2017.zip
Download and unzip files. You can find a few auxiliary functions to manage dataset in read_and_preprocess_pm.py, or just read pickled data df_pm.pickle to reduce loading times.
Just run seoul_main.py with some flags. ex) python seoul_main.py --model_dir=experiments/rnn --model=rnn