fastai_track.py

# You should download and extract the data beforehand. Simply by doing this:
# wget https://github.com/AnelMusic/Arabic_MNIST_Character_Classification/blob/master/arabic_mnist_dataset.tar.gz?raw=true
# tar -xzf 'arabic_mnist_dataset.tar.gz?raw=true'

from fastai.vision.all import DataBlock, ImageBlock, CategoryBlock, accuracy
from fastai.vision.all import GrandparentSplitter, Resize, Normalize
from fastai.vision.all import get_image_files, aug_transforms, imagenet_stats
from fastai.vision.all import cnn_learner, resnet18, CrossEntropyLossFlat
import regex as re
from aim.fastai import AimCallback


def get_arabic_mnist_labels(file_path):
    regex = "label_(.+).png"
    label = re.search(regex, str(file_path)).group(1)
    return arabic_mnist_labels[int(label) - 1]


arabic_mnist_labels = ['alef', 'beh', 'teh', 'theh', 'jeem', 'hah', 'khah',
                       'dal', 'thal', 'reh', 'zain', 'seen', 'sheen', 'sad',
                       'dad', 'tah', 'zah', 'ain', 'ghain', 'feh', 'qaf',
                       'kaf', 'lam', 'meem', 'noon', 'heh', 'waw', 'yeh']
regex = "label_(.+).png"
arab_mnist = DataBlock(blocks=(ImageBlock, CategoryBlock),
                       get_items=get_image_files,
                       splitter=GrandparentSplitter(),
                       get_y=get_arabic_mnist_labels,
                       item_tfms=Resize(460),
                       batch_tfms=[
                           *aug_transforms(do_flip=False, size=224, min_scale=0.85),
                           Normalize.from_stats(*imagenet_stats)
])
# source
dls = arab_mnist.dataloaders('.')
learn = cnn_learner(dls, resnet18, pretrained=True,
                    loss_func=CrossEntropyLossFlat(),
                    metrics=accuracy, model_dir="/tmp/model/",
                    cbs=AimCallback(repo='.', experiment='fastai_test'))
learn.fit_one_cycle(1, lr_max=slice(10e-6, 1e-4))