Notebook

In [ ]:

%reload_ext autoreload
%autoreload 2

from fastai import *
from fastai.vision import *
from fastai.vision.models.wrn import wrn_22

torch.backends.cudnn.benchmark = True

In [ ]:

path = untar_data(URLs.CIFAR)
path

Out[ ]:

PosixPath('/home/ubuntu/.fastai/data/cifar10')

In [ ]:

ds_tfms = ([*rand_pad(4, 32), flip_lr(p=0.5)], [])
data = ImageDataBunch.from_folder(path, valid='test', ds_tfms=ds_tfms, bs=512).normalize(cifar_stats)

In [ ]:

learn = Learner(data, wrn_22(), metrics=accuracy).to_fp16()
learn.fit_one_cycle(30, 3e-3, wd=0.4, div_factor=10, pct_start=0.5)

VBox(children=(HBox(children=(IntProgress(value=0, max=30), HTML(value='0.00% [0/30 00:00<00:00]'))), HTML(val…

Total time: 07:22
epoch  train loss  valid loss  accuracy
0      1.395528    1.194204    0.564700  (00:23)
1      1.044369    0.964681    0.652700  (00:14)
2      0.859872    1.128236    0.631100  (00:14)
3      0.705556    0.816264    0.724100  (00:14)
4      0.608872    0.850188    0.717200  (00:14)
5      0.544764    0.768168    0.745700  (00:14)
6      0.501907    0.717987    0.758500  (00:14)
7      0.467273    0.690202    0.770600  (00:14)
8      0.438323    0.557466    0.822600  (00:14)
9      0.414395    0.535755    0.815700  (00:14)
10     0.390469    0.460306    0.846000  (00:14)
11     0.376003    0.528630    0.820200  (00:14)
12     0.355230    0.699035    0.782800  (00:14)
13     0.343304    0.442456    0.849400  (00:14)
14     0.337318    0.559917    0.815800  (00:14)
15     0.321129    0.733412    0.769400  (00:14)
16     0.303034    0.561098    0.810200  (00:14)
17     0.278229    0.402861    0.864800  (00:14)
18     0.257787    0.504758    0.845600  (00:14)
19     0.237806    0.401391    0.873000  (00:14)
20     0.209473    0.487110    0.854300  (00:14)
21     0.184482    0.434822    0.861900  (00:14)
22     0.154364    0.310643    0.899500  (00:14)
23     0.117682    0.278385    0.911500  (00:14)
24     0.091463    0.249205    0.922400  (00:14)
25     0.066957    0.249959    0.928400  (00:14)
26     0.043469    0.219788    0.936900  (00:14)
27     0.028288    0.213345    0.939300  (00:14)
28     0.020876    0.209119    0.942300  (00:14)
29     0.017776    0.210493    0.941400  (00:14)

With mixup

In [ ]:

learn = Learner(data, wrn_22(), metrics=accuracy).to_fp16().mixup()
learn.fit_one_cycle(24, 3e-3, wd=0.2, div_factor=10, pct_start=0.5)

VBox(children=(HBox(children=(IntProgress(value=0, max=24), HTML(value='0.00% [0/24 00:00<00:00]'))), HTML(val…

Total time: 05:52
epoch  train loss  valid loss  accuracy
0      1.754659    1.418296    0.510000  (00:14)
1      1.536941    1.274806    0.569400  (00:14)
2      1.398102    0.972869    0.675200  (00:14)
3      1.308902    1.026916    0.680600  (00:14)
4      1.252701    1.091272    0.649400  (00:14)
5      1.199010    0.735006    0.765200  (00:14)
6      1.181147    0.814292    0.750400  (00:14)
7      1.152908    0.697707    0.791000  (00:14)
8      1.132457    0.746398    0.768600  (00:14)
9      1.116421    0.715845    0.788800  (00:14)
10     1.100030    0.711368    0.792600  (00:14)
11     1.088852    0.572201    0.841300  (00:14)
12     1.075793    0.733747    0.790300  (00:14)
13     1.055172    0.555941    0.851700  (00:14)
14     1.036253    0.516668    0.866700  (00:14)
15     1.013288    0.522676    0.859600  (00:14)
16     0.996412    0.494712    0.866500  (00:14)
17     0.971465    0.406641    0.902100  (00:14)
18     0.943768    0.393006    0.906800  (00:14)
19     0.919369    0.337167    0.924500  (00:14)
20     0.900417    0.327883    0.930300  (00:14)
21     0.887146    0.313054    0.936300  (00:14)
22     0.872677    0.309915    0.937800  (00:14)
23     0.864825    0.305209    0.938900  (00:14)