CIFAR 10



In [ ]:

    
%matplotlib inline
%reload_ext autoreload
%autoreload 2



In [ ]:

    
from fastai.conv_learner import *
PATH = "data/cifar10/"
os.makedirs(PATH,exist_ok=True)



In [ ]:

    
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 [ ]:

    
def get_data(sz,bs):
    tfms = tfms_from_stats(stats, sz, aug_tfms=[RandomFlipXY()], pad=sz//8)
    return ImageClassifierData.from_paths(PATH, val_name='test', tfms=tfms, bs=bs)



In [ ]:

    
bs=128

Look at data



In [ ]:

    
data = get_data(32,4)



In [ ]:

    
x,y=next(iter(data.trn_dl))



In [ ]:

    
plt.imshow(data.trn_ds.denorm(x)[0]);



In [ ]:

    
plt.imshow(data.trn_ds.denorm(x)[1]);

Initial model



In [ ]:

    
from fastai.models.cifar10.resnext import resnext29_8_64

m = resnext29_8_64()
bm = BasicModel(m.cuda(), name='cifar10_rn29_8_64')



In [ ]:

    
data = get_data(8,bs*4)



In [ ]:

    
learn = ConvLearner(data, bm)
learn.unfreeze()



In [ ]:

    
lr=1e-2; wd=5e-4



In [ ]:

    
learn.lr_find()



In [ ]:

    
learn.sched.plot()



In [ ]:

    
%time learn.fit(lr, 1)



In [ ]:

    
learn.fit(lr, 2, cycle_len=1)



In [ ]:

    
learn.fit(lr, 3, cycle_len=1, cycle_mult=2, wds=wd)



In [ ]:

    
learn.save('8x8_8')

16x16



In [ ]:

    
learn.load('8x8_8')



In [ ]:

    
learn.set_data(get_data(16,bs*2))



In [ ]:

    
%time learn.fit(1e-3, 1, wds=wd)



In [ ]:

    
learn.unfreeze()



In [ ]:

    
learn.lr_find()



In [ ]:

    
learn.sched.plot()



In [ ]:

    
lr=1e-2



In [ ]:

    
learn.fit(lr, 2, cycle_len=1, wds=wd)



In [ ]:

    
learn.fit(lr, 3, cycle_len=1, cycle_mult=2, wds=wd)



In [ ]:

    
learn.save('16x16_8')

24x24



In [ ]:

    
learn.load('16x16_8')



In [ ]:

    
learn.set_data(get_data(24,bs))



In [ ]:

    
learn.fit(1e-2, 1, wds=wd)



In [ ]:

    
learn.unfreeze()



In [ ]:

    
learn.fit(lr, 1, cycle_len=1, wds=wd)



In [ ]:

    
learn.fit(lr, 3, cycle_len=1, cycle_mult=2, wds=wd)



In [ ]:

    
learn.save('24x24_8')



In [ ]:

    
log_preds,y = learn.TTA()
preds = np.mean(np.exp(log_preds),0)metrics.log_loss(y,preds), accuracy(preds,y)

32x32



In [ ]:

    
learn.load('24x24_8')



In [ ]:

    
learn.set_data(get_data(32,bs))



In [ ]:

    
learn.fit(1e-2, 1, wds=wd)



In [ ]:

    
learn.unfreeze()



In [ ]:

    
learn.fit(lr, 3, cycle_len=1, cycle_mult=2, wds=wd)



In [ ]:

    
learn.fit(lr, 3, cycle_len=4, wds=wd)



In [ ]:

    
log_preds,y = learn.TTA()
metrics.log_loss(y,np.exp(log_preds)), accuracy(log_preds,y)



In [ ]:

    
learn.save('32x32_8')



In [ ]: