Lenet

1. First train the model

cd Lenet

python lenet_network_pruning.py --early_stopping 30

Then provide the name of the trained mode

2. To prune:

Example (choose your own model from step 1):

python lenet_network_pruning.py --method shapley --shap_method kernel --shap_sample_num 30 --path_checkpoint_load checkpoint/scratch/mnist/mnist_trainval_0.8_epo_376_acc_99.05 --load_file 0 --resume 1 --prune_bool 1

3. To prune and retrain:

python lenet_network_pruning.py --method shapley --shap_method kernel --shap_sample_num 30 --path_checkpoint_load checkpoint/scratch/mnist/mnist_trainval_0.8_epo_376_acc_99.05 --load_file 0 --resume 1 --prune_bool 1 --retrain 1

Notes:

The default early-stopping value is 500, for trying the code choose a lower value. --early_stopping 30 Choose an architecture to prune: --arch 8,9,36,17 Choose the number of shapley samples (the default is 3) --shap_sample_num 100 The previously computed Shapley samples can be reused with load_file --load_file 1

VGG

Please find all the available flags in the VGG/vgg_load.py

For basic run:

cd VGG

1. To train the base network run:

python vgg_load.py --resume 1 --prune_bool 1 --retrain_bool 0 --method shapley --shap_method kernel --shap_sample_num 3 --model checkpoint/ckpt_vgg16_94.34.t7

2. To prune:

python vgg_load.py --resume 1 --prune_bool 1 --retrain_bool 0 --method shapley --shap_method kernel --shap_sample_num 3

3. To prune and retrain:

For example, select fisher pruning:

python vgg_load.py --resume 1 --prune_bool 1 --retrain_bool 1 --method shapley --shap_method kernel --shap_sample_num 3

Choose an architecture to prune: --pruned_arch 34,34,60,60,70,101,97,88,95,85,86,67,61,55,55 The previously computed Shapley samples can be reused with load_file --load_file 1

Resnet 50

cd Resnet50

To train:

python resnet50_run3_prune.py --data <imagenet_data_path> --pretrained 0 --prune 0 --train_bool 1

2. To prune:

python resnet50_run3_prune.py --data <imagenet_data_path> --pretrained 1 --prune 1 --train_bool 1

3. To prune and retrain:

python resnet50_run3_prune.py --data <imagenet_data_path> --pretrained 1 --prune 1 --train_bool 1

The previously computed Shapley samples can be reused with load_file --load_file 1