SWALP_LR

SWALP for logistic regression on inception V3 features of ImageNet

References

We use the SWALP as starter template.

Dependencies

• CUDA 9.0
• PyTorch version 1.0
• torchvision
• tensorflow to use tensorboard

To install other requirements through $ pip install -r requirements.txt.

Dataset

• DOGFISH After downloading, we need to copy it to ./features/

Usage

We provide scripts to run Small-block Block Floating Point experiments on inception v3 features of DOGFISH with Logistic Regression. Following are scripts to reproduce experimental results.

seed=100                                      # Specify experiment seed.
bash exp/block_lr_swa.sh DOGFISH ${seed}     # SWALP training on logistic regression with Small-block BFP in DOGFISH