notebook.community

Edit and run



In [1]:

    
%matplotlib inline
%reload_ext autoreload
%autoreload 2



In [3]:

    
from fastai.conv_learner import *
# fastai/imagenet-fast/cifar10/models/ repo
from imagenet_fast_cifar_models.wideresnet import wrn_22
from torchvision import transforms, datasets

torch.backends.cudnn.benchmark = True
PATH = Path("data/cifar10")



In [4]:

    
classes = ('plane', 'car', 'bird', 'cat', 'deer', 'dog', 'frog', 'horse', 'ship', 'truck')
stats = (np.array([ 0.4914 ,  0.48216,  0.44653]), np.array([ 0.24703,  0.24349,  0.26159]))



In [14]:

    
# PATH = Path("data/cifar10_tmp")
PATH = Path("data/cifar10")



In [15]:

    
def get_loaders(bs, num_workers):
    traindir = str(PATH/'train')
    valdir   = str(PATH/'test')
    tfms     = [transforms.ToTensor(),
#                 transforms.Normalize(*stats)]
                transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010))]
    aug_tfms = transforms.Compose([
            transforms.RandomCrop(32, padding=4),
            transforms.RandomHorizontalFlip(),
        ] + tfms)
    
    train_dataset = datasets.ImageFolder(traindir, aug_tfms)
    val_dataset   = datasets.ImageFolder(valdir, transforms.Compose(tfms))
    aug_dataset   = datasets.ImageFolder(valdir, aug_tfms)
    
    train_loader = torch.utils.data.DataLoader(
        train_dataset, batch_size=bs, shuffle=True, num_workers=num_workers, pin_memory=False)
    val_loader   = torch.utils.data.DataLoader(
        val_dataset, batch_size=bs, shuffle=False, num_workers=num_workers, pin_memory=False)
    aug_loader   = torch.utils.data.DataLoader(
        aug_dataset, batch_size=bs, shuffle=False, num_workers=num_workers, pin_memory=False)
    
    return train_loader, val_loader, aug_loader



In [16]:

    
def get_data(bs, num_workers):
    trn_dl, val_dl, aug_dl = get_loaders(bs, num_workers)
    data = ModelData(PATH, trn_dl, val_dl)
    data.aug_dl = aug_dl
    data.sz = 32
    return data

def get_learner(arch, bs):
    learn = ConvLearner.from_model_data(arch.cuda(), get_data(bs, num_cpus()))
    learn.crit = nn.CrossEntropyLoss()
    learn.metrics = [accuracy]
    return learn

def get_TTA_accuracy(learn):
    preds, targs = learn.TTA()
    # combining the predictions across augmented and non augmented inputs
    preds = 0.6 * preds[0] + 0.4 * preds[1:].sum(0)
    return accuracy_np(preds, targs)



In [17]:

    
learner = get_learner(wrn_22(), 16)
learner.lr_find(wds=1e-4)
learner.sched.plot(n_skip_end=1)









    





 
 










    



100%|█████████▉| 3111/3125 [03:44<00:01, 13.86it/s, loss=8.2]



In [19]:

    
learner = get_learner(wrn_22(), 512)
learner.lr_find(wds=1e-4)
learner.sched.plot(n_skip_end=1)









    





 
 










    



epoch      trn_loss   val_loss   accuracy                 
    0      2.23063    24041.498281 0.1



In [20]:

    
torch.__version__









    Out[20]:





'0.3.1.post2'



In [22]:

    
%time learner.fit(1.5, 1, wds=1e-4, cycle_len=30, use_clr_beta=(15, 10, 0.95, 0.85));









    





 
 










    



epoch      trn_loss   val_loss   accuracy                 
    0      1.610085   1.626076   0.4395    
    1      1.223421   1.317524   0.5633                   
    2      0.940387   2.038279   0.4745                    
    3      0.762601   1.217451   0.6398                    
    4      0.654419   0.961155   0.7008                    
    5      0.565586   0.639681   0.7832                    
    6      0.50137    0.599894   0.7996                    
    7      0.460526   0.626692   0.7843                    
    8      0.425721   0.603371   0.7926                    
    9      0.393813   0.893688   0.749                     
    10     0.374892   0.699583   0.7713                    
    11     0.353577   0.630499   0.8066                    
    12     0.345882   0.576161   0.805                     
    13     0.327537   0.719755   0.781                     
    14     0.313273   0.639451   0.7996                    
    15     0.292252   0.559769   0.832                     
    16     0.282883   0.431539   0.856                     
    17     0.259445   0.644423   0.808                     
    18     0.243024   0.464651   0.8583                    
    19     0.232724   0.540036   0.8494                    
    20     0.217015   0.504534   0.8511                    
    21     0.207343   0.545313   0.8373                    
    22     0.18528    0.357806   0.8941                    
    23     0.163668   0.405116   0.8805                    
    24     0.140306   0.378086   0.8852                    
    25     0.112736   0.281033   0.9133                    
    26     0.074481   0.253059   0.9282                     
    27     0.045172   0.249973   0.9306                     
    28     0.027193   0.233644   0.9336                     
    29     0.019215   0.235123   0.9351                     
CPU times: user 1h 5min 31s, sys: 13min 45s, total: 1h 19min 16s
Wall time: 1h 19min 41s






    Out[22]:





[array([0.23512]), 0.9350999992370606]



In [24]:

    
get_TTA_accuracy(learner)



In [ ]: