refactor: partition in elements

[qued]

Added the capability to partition granular-scale elements that have been identified (words, characters) by proximity using the word/character height as a reference. In many cases this does a good job of grouping text blocks. Also moved logic for extracting text from a region into the region itself, so in the future different logic can be used for embedded text and images.

Testing:

In a jupyter notebook, try the following code:

from unstructured_inference.inference.layout import load_pdf
from unstructured_inference.inference.elements import partition_groups_from_regions

words, images = load_pdf('sample-docs/layout-parser-paper.pdf')

colors = ["red", "green", "blue", "cyan", "yellow", "magenta", "purple"] * 8

from PIL.ImageDraw import ImageDraw
def annotate_regions(rects, image, color):
    img = image.copy()
    draw = ImageDraw(img)
    for rect in rects:
        x1, y1, x2, y2 = rect.x1, rect.y1, rect.x2, rect.y2
        draw.rectangle(((x1, y1), (x2, y2)), outline=color)
    return img

def annotate_groups(groups, image):
    out_image = image.copy()
    for i, group in enumerate(groups):
        out_image = annotate_regions(group, out_image, colors[i])
    return out_image
 
groups = partition_groups_from_regions(words[0])
annotate_groups(groups, images[0])

The output should be the first page of the layout parser paper with color-coded groupings of the words in the page, representing the partition produced.

[MthwRobinson]

Added a couple of inline question, but overall looks good. I also keep getting the following error message on the test code.

Also, I'm getting this error on the test code. I'm able to run the unstructured and unstructured-inference tests in the same environment though and the test code works in iPython so wondering if it's something related to my jupyter version.

ImportError: dlopen(/Users/mrobinson/.pyenv/versions/inference/lib/python3.8/site-packages/detectron2/_C.cpython-38-darwin.so, 0x0002): Symbol not found: __ZN2at4_ops10select_int4callERKNS_6TensorExx
  Referenced from: /Users/mrobinson/.pyenv/versions/3.8.13/envs/inference/lib/python3.8/site-packages/detectron2/_C.cpython-38-darwin.so
  Expected in: /Users/mrobinson/.pyenv/versions/3.8.13/envs/inference/lib/python3.8/site-packages/torch/lib/libtorch_cpu.dylib

[MthwRobinson]

LGTM pending a comment on the coefficients!