A GAN-based method for time-dependent cloud workload generation

PyTorch implementation of A GAN-based method for time-dependent cloud workload generation.

Download dataset

Alibaba trace:
https://drive.google.com/file/d/173vBtzX4S-gslVmcH9QXTHnx9NKFulZt/view?usp=share_link

Getting started

PlanetLab

train:
python gan.py --data="data/planetLab/train.npy" --use_GPU=True --g_num_layers=6 --g_num_filters=32 --g_kernel_size=7 --d_num_layers=3 --d_downsample True True True

generate samples:
python samples_generate.py --data="data/planetLab/train.npy" --model_path="model_saved/planetLab.pkl" --use_GPU=True --g_num_layers=6 --g_num_filters=32 --g_kernel_size=7

Alibaba machine

train:
python gan.py --data="data/alibaba_machine/train.npy" --use_GPU=True --batch_size=64 --g_num_layers=8 --g_num_filters=25 --g_kernel_size=7 --d_num_layers=4 --d_downsample True True True True

generate samples:
python samples_generate.py --data="data/alibaba_machine/train.npy" --model_path="model_saved/alibaba_machine.pkl" --use_GPU=True --g_num_layers=8 --g_num_filters=25 --g_kernel_size=7

Alibaba container

train:
python cgan.py --data="data/alibaba_container/train.npy" --label="data/alibaba_container/label.npy" --use_GPU=True --g_num_layers=8 --g_num_filters=25 --g_kernel_size=7 --d_num_layers=4 --d_downsample True True True True

generate samples:
python samples_generate.py --data="data/alibaba_container/train.npy" --n_classes=7 --model_path="model_saved/1000epochs.pkl" --use_GPU=True --g_num_layers=8 --g_num_filters=25 --g_kernel_size=7