Examples of many applications

This notebook is a quick(ish) test of most of the main application people use, taken from fastbook.

Image single classification

set_seed(99, True)
path = untar_data(URLs.PETS)/'images'
dls = ImageDataLoaders.from_name_func(
    path, get_image_files(path), valid_pct=0.2,
    label_func=lambda x: x[0].isupper(), item_tfms=Resize(224))

learn = vision_learner(dls, resnet34, metrics=error_rate).to_fp16()
learn.fine_tune(1)

epoch	train_loss	valid_loss	error_rate	time
0	0.166802	0.026535	0.009472	00:13

epoch	train_loss	valid_loss	error_rate	time
0	0.046429	0.012175	0.004060	00:15

Should be around 0.01 or less.

img = PILImage.create('../images/cat.jpg')
print(f"Probability it's a cat: {learn.predict(img)[2][1].item():.6f}")

Probability it's a cat: 1.000000

Segmentation

path = untar_data(URLs.CAMVID_TINY)
dls = SegmentationDataLoaders.from_label_func(
    path, bs=8, fnames = get_image_files(path/"images"),
    label_func = lambda o: path/'labels'/f'{o.stem}_P{o.suffix}',
    codes = np.loadtxt(path/'codes.txt', dtype=str)
)

learn = unet_learner(dls, resnet34)

learn.fine_tune(8)

epoch	train_loss	valid_loss	time
0	2.969643	2.448999	00:03

epoch	train_loss	valid_loss	time
0	1.696013	1.509799	00:02
1	1.484652	1.214625	00:02
2	1.457932	1.400337	00:02
3	1.351992	1.104257	00:02
4	1.233100	0.925221	00:03
5	1.122062	0.854342	00:02
6	1.028064	0.832836	00:02
7	0.954791	0.818452	00:02

Should be a bit above 0.8.

learn.show_results(max_n=6, figsize=(7,8))

RHS pics should be similar to LHS pics.

Text classification

path = untar_data(URLs.IMDB_SAMPLE)
df = pd.read_csv(path/'texts.csv')
imdb_clas = DataBlock(blocks=(TextBlock.from_df('text', seq_len=72), CategoryBlock),
                      get_x=ColReader('text'), get_y=ColReader('label'), splitter=ColSplitter())
dls = imdb_clas.dataloaders(df, bs=64)

learn = text_classifier_learner(dls, AWD_LSTM, drop_mult=0.5, metrics=accuracy)
learn.fine_tune(4, 1e-2)

epoch	train_loss	valid_loss	accuracy	time
0	0.634987	0.621853	0.705000	00:03

epoch	train_loss	valid_loss	accuracy	time
0	0.435843	0.508709	0.805000	00:05
1	0.386479	0.446490	0.790000	00:05
2	0.320961	0.522036	0.770000	00:05
3	0.267238	0.572048	0.785000	00:05

Should be a bit under 0.8.

learn.predict("I really liked that movie!")

('positive', tensor(1), tensor([1.3686e-05, 9.9999e-01]))

Should be a bit very nearly 1.

Tabular

path = untar_data(URLs.ADULT_SAMPLE)

dls = TabularDataLoaders.from_csv(path/'adult.csv', path=path, y_names="salary",
    cat_names = ['workclass', 'education', 'marital-status', 'occupation',
                 'relationship', 'race'],
    cont_names = ['age', 'fnlwgt', 'education-num'],
    procs = [Categorify, FillMissing, Normalize])

learn = tabular_learner(dls, metrics=accuracy)
learn.fit_one_cycle(3)

epoch	train_loss	valid_loss	accuracy	time
0	0.365791	0.362030	0.824171	00:07
1	0.349386	0.356809	0.832310	00:07
2	0.340156	0.352458	0.831695	00:07

Should be around 0.83

Collab filtering

path = untar_data(URLs.ML_SAMPLE)
dls = CollabDataLoaders.from_csv(path/'ratings.csv')
learn = collab_learner(dls, y_range=(0.5,5.5))

learn.fine_tune(6)

epoch	train_loss	valid_loss	time
0	1.519063	1.388863	00:00

epoch	train_loss	valid_loss	time
0	1.353412	1.301542	00:00
1	1.173434	0.995304	00:00
2	0.894901	0.777776	00:00
3	0.735333	0.725698	00:00
4	0.670091	0.714968	00:00
5	0.648496	0.713515	00:00

Should be a bit over 0.7

learn.show_results(max_n=4)

	userId	movieId	rating	rating_pred
0	45.0	95.0	1.0	3.216705
1	6.0	42.0	3.5	3.474894
2	2.0	3.0	4.5	3.789701
3	15.0	34.0	2.5	3.857912

Keypoints

path = untar_data(URLs.BIWI_HEAD_POSE)
img_files = get_image_files(path)
def img2pose(x): return Path(f'{str(x)[:-7]}pose.txt')
cal = np.genfromtxt(path/'01'/'rgb.cal', skip_footer=6)
def get_ctr(f):
    ctr = np.genfromtxt(img2pose(f), skip_header=3)
    c1 = ctr[0] * cal[0][0]/ctr[2] + cal[0][2]
    c2 = ctr[1] * cal[1][1]/ctr[2] + cal[1][2]
    return tensor([c1,c2])

biwi = DataBlock(
    blocks=(ImageBlock, PointBlock),
    get_items=get_image_files,
    get_y=get_ctr,
    splitter=FuncSplitter(lambda o: o.parent.name=='13'),
    batch_tfms=[*aug_transforms(size=(240,320)), 
                Normalize.from_stats(*imagenet_stats)])

dls = biwi.dataloaders(path)
dls.show_batch(max_n=9, figsize=(8,6))

learn = vision_learner(dls, resnet18, y_range=(-1,1))

learn.lr_find()

SuggestedLRs(lr_min=0.006918309628963471, lr_steep=9.12010818865383e-07)

learn.fine_tune(1, 1e-2)

epoch	train_loss	valid_loss	time
0	0.048685	0.011717	00:37

epoch	train_loss	valid_loss	time
0	0.003574	0.000477	00:47

Should be around 0.0005

learn.show_results(ds_idx=1, nrows=3, figsize=(6,8))

Red dots should be close to noses.