PyTorch code for generating particle number size distribution (PNSD) datasets.
For datasets, you can use the code get_datasets to download the measured PNSD for SMEAR I, II, and III from https://smear.avaa.csc.fi/.
You can also use the datasets we have already downloaded:
measured PNSD for SMEAR I, II, III
Put the datasets under the folder datasets and rename them with the names var, hyy, and kum for SMEAR I, II, and III, respectively.
Though we have observations for many years, we need more samples for training.
For data augmentation, please run (use hyy and kum for SMEAR II and III, respectively)
$ python3 datasets/obtain_csv.py --station var
We use the WGAN-GP as the default model, which can alleviate the mode collapse problem.
$ python3 demo.py --exp_name var --station var --epochs 20 --dataroot datasets --batch_size 64 --lr 0.0001 --lambda_gp 10
To visualize the generated results, please refer to the demo.ipynb file for more details.