FastAI2021-Intro#

%load_ext autoreload
%autoreload 2

Vision Example#

from aiking.visuals import make_flowchart
blocks = ['Imports', 'Data Download/Engineering','Create Dataloaders', 'Create Learner', 'Training & Fine Tuning']

layers = ['from fastai.vision.all import *\n\
from fastai.vision.widgets import *\n\
from aiking.data.external import *', 
          'untar_data \n url => path',
          'ImageDataLoaders.from_name_func \n (path, get_image_files, valid_pct, label_func, item_tfms) => Dataloaders',
          'cnn_learner \n (Dataloaders, architecture, metrics) => learner',
          'learner.fine_tune \n (epochs) \n Prints Table']

make_flowchart(blocks, layers)
../../_images/introduction_3_0.svg

Imports#

from fastai.vision.all import *
from fastai.vision.widgets import *
from fastai.text.all import *
from fastai.tabular.all import *
from fastai.collab import *
from aiking.data.external import * #We need to import this after fastai modules 
from fastdownload import FastDownload

2022-03-11 09:50:05.298431: W tensorflow/stream_executor/platform/default/dso_loader.cc:64] Could not load dynamic library 'libcudart.so.11.0'; dlerror: libcudart.so.11.0: cannot open shared object file: No such file or directory
2022-03-11 09:50:05.298470: I tensorflow/stream_executor/cuda/cudart_stub.cc:29] Ignore above cudart dlerror if you do not have a GPU set up on your machine.

Data Download and Engineering#

URLs.PETS

'https://s3.amazonaws.com/fast-ai-imageclas/oxford-iiit-pet.tgz'

os.environ['AIKING_HOME']

'/Landmark2/pdo/aiking'

d = FastDownload(base=os.environ['AIKING_HOME'], archive='archive', data='data')
d.get(URLs.PETS)

Path('/Landmark2/pdo/aiking/data/oxford-iiit-pet')

untar_data(URLs.PETS)

Path('/Landmark2/pdo/aiking/data/oxford-iiit-pet')

path = untar_data(URLs.PETS)/"images"; path

Path('/Landmark2/pdo/aiking/data/oxford-iiit-pet/images')

# !rm -rf  /content/drive/MyDrive/PPV/S_Personal_Study/aiking/archive

# !rm -rf /content/drive/MyDrive/PPV/S_Personal_Study/aiking/archive/oxford-iiit-pet.tgz

list_ds()

(#14) ['oxford-iiit-pet','DoppelGanger','mnist_sample','Bears','camvid_tiny','imdb','california-housing-prices','adult_sample','imdb_tok','ultra-mnist'...]

path.ls()

(#7393) [Path('/Landmark2/pdo/aiking/data/oxford-iiit-pet/images/Russian_Blue_124.jpg'),Path('/Landmark2/pdo/aiking/data/oxford-iiit-pet/images/miniature_pinscher_195.jpg'),Path('/Landmark2/pdo/aiking/data/oxford-iiit-pet/images/British_Shorthair_54.jpg'),Path('/Landmark2/pdo/aiking/data/oxford-iiit-pet/images/Persian_60.jpg'),Path('/Landmark2/pdo/aiking/data/oxford-iiit-pet/images/american_pit_bull_terrier_161.jpg'),Path('/Landmark2/pdo/aiking/data/oxford-iiit-pet/images/great_pyrenees_174.jpg'),Path('/Landmark2/pdo/aiking/data/oxford-iiit-pet/images/Ragdoll_197.jpg'),Path('/Landmark2/pdo/aiking/data/oxford-iiit-pet/images/saint_bernard_65.jpg'),Path('/Landmark2/pdo/aiking/data/oxford-iiit-pet/images/pug_99.jpg'),Path('/Landmark2/pdo/aiking/data/oxford-iiit-pet/images/Russian_Blue_40.jpg')...]

Create Dataloaders#

doc(get_image_files)

get_image_files[source]

get_image_files(path, recurse=True, folders=None)

Get image files in path recursively, only in folders, if specified.

Show in docs

def is_cat(x): return x[0].isupper()

imgname = path.ls()[0]
print(imgname,"|||", imgname.parts[-1])
imgname.parts[-1][0].isupper()

/Landmark2/pdo/aiking/data/oxford-iiit-pet/images/Russian_Blue_124.jpg ||| Russian_Blue_124.jpg

True

get_image_files(path), path.ls()

((#7390) [Path('/Landmark2/pdo/aiking/data/oxford-iiit-pet/images/Russian_Blue_124.jpg'),Path('/Landmark2/pdo/aiking/data/oxford-iiit-pet/images/miniature_pinscher_195.jpg'),Path('/Landmark2/pdo/aiking/data/oxford-iiit-pet/images/British_Shorthair_54.jpg'),Path('/Landmark2/pdo/aiking/data/oxford-iiit-pet/images/Persian_60.jpg'),Path('/Landmark2/pdo/aiking/data/oxford-iiit-pet/images/american_pit_bull_terrier_161.jpg'),Path('/Landmark2/pdo/aiking/data/oxford-iiit-pet/images/great_pyrenees_174.jpg'),Path('/Landmark2/pdo/aiking/data/oxford-iiit-pet/images/Ragdoll_197.jpg'),Path('/Landmark2/pdo/aiking/data/oxford-iiit-pet/images/saint_bernard_65.jpg'),Path('/Landmark2/pdo/aiking/data/oxford-iiit-pet/images/pug_99.jpg'),Path('/Landmark2/pdo/aiking/data/oxford-iiit-pet/images/Russian_Blue_40.jpg')...],
 (#7393) [Path('/Landmark2/pdo/aiking/data/oxford-iiit-pet/images/Russian_Blue_124.jpg'),Path('/Landmark2/pdo/aiking/data/oxford-iiit-pet/images/miniature_pinscher_195.jpg'),Path('/Landmark2/pdo/aiking/data/oxford-iiit-pet/images/British_Shorthair_54.jpg'),Path('/Landmark2/pdo/aiking/data/oxford-iiit-pet/images/Persian_60.jpg'),Path('/Landmark2/pdo/aiking/data/oxford-iiit-pet/images/american_pit_bull_terrier_161.jpg'),Path('/Landmark2/pdo/aiking/data/oxford-iiit-pet/images/great_pyrenees_174.jpg'),Path('/Landmark2/pdo/aiking/data/oxford-iiit-pet/images/Ragdoll_197.jpg'),Path('/Landmark2/pdo/aiking/data/oxford-iiit-pet/images/saint_bernard_65.jpg'),Path('/Landmark2/pdo/aiking/data/oxford-iiit-pet/images/pug_99.jpg'),Path('/Landmark2/pdo/aiking/data/oxford-iiit-pet/images/Russian_Blue_40.jpg')...])

doc(ImageDataLoaders.from_name_func)

ImageDataLoaders.from_name_func[source]

ImageDataLoaders.from_name_func(path, fnames, label_func, valid_pct=0.2, seed=None, item_tfms=None, batch_tfms=None, bs=64, val_bs=None, shuffle=True, device=None)

Create from the name attrs of fnames in paths with label_func

Show in docs

dls = ImageDataLoaders.from_name_func(path, 
                                      get_image_files(path), 
                                      valid_pct=0.2,
                                      seed=42,
                                      label_func=is_cat,
                                      item_tfms=Resize(224)); dls

<fastai.data.core.DataLoaders at 0x112bba2ef160>

Create Learner#

doc(cnn_learner)

cnn_learner[source]

cnn_learner(dls, arch, normalize=True, n_out=None, pretrained=True, config=None, loss_func=None, opt_func=Adam, lr=0.001, splitter=None, cbs=None, metrics=None, path=None, model_dir='models', wd=None, wd_bn_bias=False, train_bn=True, moms=(0.95, 0.85, 0.95), cut=None, n_in=3, init=kaiming_normal_, custom_head=None, concat_pool=True, lin_ftrs=None, ps=0.5, first_bn=True, bn_final=False, lin_first=False, y_range=None)

Build a convnet style learner from dls and arch

Show in docs

learn = cnn_learner(dls, resnet34, metrics=error_rate); learn

Downloading: "https://download.pytorch.org/models/resnet34-b627a593.pth" to /home/rahul.saraf/.cache/torch/hub/checkpoints/resnet34-b627a593.pth

<fastai.learner.Learner at 0x112bb9e22760>

Training and Fine Tuning#

doc(learn.fine_tune)

Learner.fine_tune[source]

Learner.fine_tune(epochs, base_lr=0.002, freeze_epochs=1, lr_mult=100, pct_start=0.3, div=5.0, lr_max=None, div_final=100000.0, wd=None, moms=None, cbs=None, reset_opt=False)

Fine tune with Learner.freeze for freeze_epochs, then with Learner.unfreeze for epochs, using discriminative LR.

Show in docs

learn.fine_tune(1)
epoch train_loss valid_loss error_rate time
0 0.176319 0.031685 0.011502 00:54
epoch train_loss valid_loss error_rate time
0 0.064987 0.027418 0.008796 01:09 
uploader = widgets.FileUpload(); uploader

# !cat ~/anaconda3/envs/aiking/lib/python3.9/site-packages/fastai/vision/all.py

img = PILImage.create(uploader.data[0])
img

---------------------------------------------------------------------------
IndexError                                Traceback (most recent call last)
/tmp/ipykernel_14311/450188109.py in <module>
----> 1 img = PILImage.create(uploader.data[0])
      2 img

IndexError: list index out of range

doc(learn.predict)

Learner.predict[source]

Learner.predict(item, rm_type_tfms=None, with_input=False)

Prediction on item, fully decoded, loss function decoded and probabilities

Show in docs

prediction,_, probs = learn.predict(img)
print(f"Is this a cat?: {prediction}.")
print(f"Probability it's a cat: {probs}")

Is this a cat?: True.
Probability it's a cat: tensor([2.8618e-18, 1.0000e+00])

prediction,_, probs = learn.predict(img)
print(f"Is this a cat?: {prediction}.")
print(f"Probability it's a cat: {probs[1].item():.6f}")

Is this a cat?: True.
Probability it's a cat: 1.000000

Segmentation Example#

Creating a model that can recognize the content of every individual pixel in an image is called segmentation

path = untar_data(URLs.CAMVID_TINY); path
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)
); dls

learn = unet_learner(dls, resnet34); learn
learn.fine_tune(8)
epoch train_loss valid_loss time
0 3.968191 2.788000 00:09
epoch train_loss valid_loss time
0 2.070896 1.643526 00:04
1 1.716042 1.232054 00:03
2 1.537476 1.075623 00:03
3 1.413292 1.040084 00:03
4 1.283577 0.900883 00:03
5 1.157600 0.799863 00:03
6 1.050192 0.761217 00:03
7 0.966571 0.760026 00:03 
(path/"labels").ls()[0].stem

'Seq05VD_f02370_P'

np.loadtxt(path/'codes.txt', dtype=str)

array(['Animal', 'Archway', 'Bicyclist', 'Bridge', 'Building', 'Car',
       'CartLuggagePram', 'Child', 'Column_Pole', 'Fence', 'LaneMkgsDriv',
       'LaneMkgsNonDriv', 'Misc_Text', 'MotorcycleScooter', 'OtherMoving',
       'ParkingBlock', 'Pedestrian', 'Road', 'RoadShoulder', 'Sidewalk',
       'SignSymbol', 'Sky', 'SUVPickupTruck', 'TrafficCone',
       'TrafficLight', 'Train', 'Tree', 'Truck_Bus', 'Tunnel',
       'VegetationMisc', 'Void', 'Wall'], dtype='<U17')

learn.show_results(max_n=6, figsize=(7,8))
../../_images/introduction_39_1.png 
path

Path('/Landmark2/pdo/aiking/data/camvid_tiny')

Text Example#

dls = TextDataLoaders.from_folder(untar_data(URLs.IMDB), valid='test'); dls
learn = text_classifier_learner(dls, AWD_LSTM, drop_mult=0.5, metrics=accuracy); learn
learn.fine_tune(4, 1e-2)
epoch train_loss valid_loss accuracy time
0 0.457429 0.427618 0.803440 06:59
epoch train_loss valid_loss accuracy time
0 0.314991 0.253363 0.898840 13:40
1 0.244463 0.206381 0.919240 13:41
2 0.184156 0.188984 0.927960 13:40
3 0.154114 0.198980 0.927840 13:39

Tabular Example#

doc(untar_data)

untar_data[source]

untar_data(url, archive=None, data=None, c_key='data', force_download=False)

Download url to fname if dest doesn't exist, and extract to folder dest

Show in docs

path  = untar_data(URLs.ADULT_SAMPLE); path.ls()
100.69% [974848/968212 00:00<00:00] 
(#3) [Path('/content/drive/MyDrive/PPV/S_Personal_Study/aiking/data/adult_sample/export.pkl'),Path('/content/drive/MyDrive/PPV/S_Personal_Study/aiking/data/adult_sample/adult.csv'),Path('/content/drive/MyDrive/PPV/S_Personal_Study/aiking/data/adult_sample/models')]

df = pd.read_csv(path/'adult.csv'); df.head()
age workclass fnlwgt education education-num marital-status occupation relationship race sex capital-gain capital-loss hours-per-week native-country salary
0 49 Private 101320 Assoc-acdm 12.0 Married-civ-spouse NaN Wife White Female 0 1902 40 United-States >=50k
1 44 Private 236746 Masters 14.0 Divorced Exec-managerial Not-in-family White Male 10520 0 45 United-States >=50k
2 38 Private 96185 HS-grad NaN Divorced NaN Unmarried Black Female 0 0 32 United-States <50k
3 38 Self-emp-inc 112847 Prof-school 15.0 Married-civ-spouse Prof-specialty Husband Asian-Pac-Islander Male 0 0 40 United-States >=50k
4 42 Self-emp-not-inc 82297 7th-8th NaN Married-civ-spouse Other-service Wife Black Female 0 0 50 United-States <50k 
df.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 32561 entries, 0 to 32560
Data columns (total 15 columns):
 #   Column          Non-Null Count  Dtype  
---  ------          --------------  -----  
 0   age             32561 non-null  int64  
 1   workclass       32561 non-null  object 
 2   fnlwgt          32561 non-null  int64  
 3   education       32561 non-null  object 
 4   education-num   32074 non-null  float64
 5   marital-status  32561 non-null  object 
 6   occupation      32049 non-null  object 
 7   relationship    32561 non-null  object 
 8   race            32561 non-null  object 
 9   sex             32561 non-null  object 
 10  capital-gain    32561 non-null  int64  
 11  capital-loss    32561 non-null  int64  
 12  hours-per-week  32561 non-null  int64  
 13  native-country  32561 non-null  object 
 14  salary          32561 non-null  object 
dtypes: float64(1), int64(5), object(9)
memory usage: 3.7+ MB

doc(TabularDataLoaders.from_csv)

TabularDataLoaders.from_csv[source]

TabularDataLoaders.from_csv(csv, skipinitialspace=True, path='.', procs=None, cat_names=None, cont_names=None, y_names=None, y_block=None, valid_idx=None, bs=64, shuffle_train=None, shuffle=True, val_shuffle=False, n=None, device=None, drop_last=None, val_bs=None)

Create from csv file in path using procs

Show in docs

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

<fastai.tabular.data.TabularDataLoaders at 0x7f901f566550>

learn = tabular_learner(dls, metrics=accuracy); learn

<fastai.tabular.learner.TabularLearner at 0x7f901f571210>

learn.fit_one_cycle(5)
epoch train_loss valid_loss accuracy time
0 0.485079 0.465481 0.767813 00:05
1 0.458598 0.448505 0.780098 00:05
2 0.439319 0.444874 0.787776 00:05
3 0.447785 0.442088 0.787776 00:05
4 0.438344 0.440791 0.787623 00:05

Colab Example#

path = untar_data(URLs.ML_SAMPLE); path.ls()
110.72% [57344/51790 00:00<00:00] 
(#1) [Path('/content/drive/MyDrive/PPV/S_Personal_Study/aiking/data/movie_lens_sample/ratings.csv')]

doc(CollabDataLoaders.from_csv)

CollabDataLoaders.from_csv[source]

CollabDataLoaders.from_csv(csv, valid_pct=0.2, user_name=None, item_name=None, rating_name=None, seed=None, path='.', bs=64, val_bs=None, shuffle=True, device=None)

Create a DataLoaders suitable for collaborative filtering from csv.

Show in docs

dls = CollabDataLoaders.from_csv(path/'ratings.csv')
learn = collab_learner(dls, y_range=(0.5, 5)) # y_range  : Range of y values
learn.fine_tune(10) # without a pretrained model
epoch train_loss valid_loss time
0 1.949988 1.816280 00:00
epoch train_loss valid_loss time
0 1.752750 1.726716 00:00
1 1.590970 1.451683 00:00
2 1.256610 0.987660 00:00
3 0.925884 0.752191 00:00
4 0.752691 0.684037 00:00
5 0.691877 0.662923 00:00
6 0.663190 0.653058 00:00
7 0.651265 0.648660 00:00
8 0.639195 0.647364 00:00
9 0.628522 0.647159 00:00 
learn.show_results()
userId movieId rating rating_pred
0 53.0 22.0 3.0 3.991087
1 14.0 6.0 4.0 3.812191
2 16.0 77.0 4.0 4.235012
3 94.0 89.0 4.0 3.715150
4 77.0 80.0 5.0 4.053984
5 41.0 5.0 5.0 3.524751
6 18.0 62.0 4.5 3.794177
7 87.0 62.0 5.0 4.103097
8 77.0 33.0 3.0 3.273965