Skip to content

Latest commit

 

History

History
22 lines (15 loc) · 1.05 KB

README.md

File metadata and controls

22 lines (15 loc) · 1.05 KB

PreciseGAN

A research repo for studying different techniques towards making more precise GANs

Experiment reproduction

To reproduce experiment first you need data, that can be downloaded with get_csv.sh script. Then you should learn and dump MinMaxScaler with script init_scaler.py, and make train/test data split with init_data.py

Then run main.py with preferred parameters

Notebook usage example: notebook_usage.ipynb

Original paper reproduction

To reproduce results of original paper "DijetGAN: a Generative-Adversarial Network approach for the simulation of QCD dijet events at the LHC" you should initialize data with --train_split=0.8 for integral training data and --train_split=0.15 --task=tail for tail training data

then run main.py with CLI parameters --architecture=cnn --iterations=500000 --optim=adam -lr 1e-5 --adam_beta_1=0.5 --adam_beta_2=0.9 --batch_size=128 and --level=ptcl for particle level or --level=reco for reco level, also specify --task=integral for integral training or --task=tail for tail training.