Cleanup

src/testtest.py

@@ -1,53 +0,0 @@
-from robustness import model_utils, datasets
-import torch as ch
-from tqdm.auto import tqdm 
-from torchvision import models
-
-# ds = datasets.CIFAR('/tmp/')
-ds = datasets.ImageNet('/home/hasalman/datasets/IMAGENET/imagenet')
-
-for ARCH in ['mnasnet',
-            # 'shufflenet',
-            # 'vgg16_bn',
-            # 'densenet',
-            # 'resnext50_32x4d',
-            # 'mobilenet',
-            ]:
-    for EPS in [0, 3]:
-        print(f'ARCH: {ARCH}     |     EPS: {EPS}')
-        # model_path = '../notebooks/cifar_l2_0_25.pt'
-        # model_path = '../notebooks/cifar_l2_1_0.pt'
-        # model_path = '../notebooks/cifar_nat.pt'
-        # model_path = 'msnet.ckpt'
-        # model_path = 'checkpoint.pt.best'
-        model_path = f'/home/hasalman/azure-madrylab/imagenet_experiments/opensource_models/{ARCH}_l2_eps{EPS}.ckpt'
-        # model_path = None
-
-        pytorch_models = {
-            'alexnet': models.alexnet,
-            'vgg16': models.vgg16,
-            'vgg16_bn': models.vgg16_bn,
-            'squeezenet': models.squeezenet1_0,
-            'densenet': models.densenet161,
-            # 'inception': models.inception_v3,
-            # 'googlenet': models.googlenet,
-            'shufflenet': models.shufflenet_v2_x1_0,
-            'mobilenet': models.mobilenet_v2,
-            'resnext50_32x4d': models.resnext50_32x4d,
-            'mnasnet': models.mnasnet1_0,
-        }
-
-        model, checkpoint = model_utils.make_and_restore_model(arch=pytorch_models[ARCH](True), dataset=ds, 
-                                                            resume_path=model_path, add_custom_forward=True)
-        # model, checkpoint = model_utils.make_and_restore_model(arch='resnet50', dataset=ds, resume_path=model_path, add_custom_forward=False)
-        train_loader, val_loader = ds.make_loaders(batch_size=64, workers=4)
-
-        correct = 0
-        model.eval()
-        with ch.no_grad():
-            for X,y in tqdm(val_loader):
-                X,y = X.cuda(), y.cuda()
-                out = model(X, with_image=False)
-                _, pred = out.topk(1,1)
-                correct += (pred.squeeze()==y).detach().cpu().sum()
-        print(f'The clean accuracy is {1.*correct/len(val_loader.dataset)*100.}%')

src/whileloop.py

@@ -1,5 +0,0 @@
-import time
-
-while True:
-    print('Hey')
-    time.sleep(1000)