Wanted to show one interesting feature of FastAI library that I though at first is inconsistency.

It addresses creating, splitting, and labeling data.

The following code will import the PLANET_TINY dataset and print out the classes in use after every step.

import fastai
from fastai.vision import *
path_data = untar_data(URLs.PLANET_TINY); path_data.ls()

def bases(obj):    
    'Will provide obj class and base classes'
    print("...")
    print(type(obj))
    print(obj.__class__.__name__)
    for base in obj.__class__.__bases__:
        print(base)
        print (base.__name__)

        
# C:/Users/dj/.fastai/data/planet_tiny/train/labels.csv
c = ImageList.from_csv(path_data, 'labels.csv', folder='train-jpg', suffix='.jpg')
bases(c)   
c = c.split_by_rand_pct(0.2)
bases(c)   
c.label_from_df(label_delim=' ')
bases(c)

Out

...
<class 'fastai.vision.data.ImageList'>
ImageList
<class 'fastai.data_block.ItemList'>
ItemList
...
<class 'fastai.data_block.ItemLists'>
ItemLists
<class 'object'>
object
...
<class 'fastai.data_block.LabelLists'>
LabelLists
<class 'fastai.data_block.ItemLists'>
ItemLists

As you can see, we first create ImageList (base in ItemList), and then after the splitting we get ItemLists, and after the labeling we get LabelLists (based in ItemLists).

All classes do implement __call__ so we can write like this as well:

c = (ImageList.from_csv(path_data, 'labels.csv', folder='train-jpg', suffix='.jpg')
        .split_by_rand_pct(0.2)
        .label_from_df(label_delim=' '))