prepare_dataset.py

def img_size_group(size):
  max_size = 50000
  #image size bins
  img_size_groups = [i for i in 
                      zip(
                        [i for i in range(0, max_size, 1000)], 
                        [i for i in range(1000, max_size, 1000)]
                      )
                    ]
  
  for lower, upper in img_size_groups:
    if size > max_size:
      return str(max_size)+"+"
    elif lower < size < upper:
      return "{}-{}".format(lower, upper)
    
    
img_counts['filesize_group'] = img_counts['filesize'].apply(img_size_group)

#convert image size bins to categories
onehot_img = img_counts[['url']].join(pd.get_dummies(img_counts['filesize_group'], drop_first=True))
onehot_img = onehot_img.groupby("url").sum().reset_index()
onehot_img = img_counts[["url"]].merge(onehot_img, on="url", how="left").drop_duplicates()

# Each row = URL
# Each column = imagesize bucket
# Value = Count of images from that URL with that imagesize
onehot_img.drop("url", axis=1).head()