Tables, and text_as_html

A table is one element, however many rows it has. Its text is useless and its metadata is not.

The refund policy has a two column table in it. Here is what the Table element says.

from unstructured.partition.auto import partition

table = partition("refunds.html")[3]
print(table.category)
print(table.text)

Every cell, in reading order, separated by single spaces. Nothing marks the end of a row and nothing marks the header. Card 5 Bank transfer 8 is the whole table and it is not something you can work with.

The shape is in the metadata

from unstructured.partition.auto import partition

table = partition("refunds.html")[3]
print(table.metadata.text_as_html)

The rows are back. This is the field to hand to anything downstream that needs the table as a table, and it is the reason a table element is worth keeping whole rather than splitting.

Look closely at the first row. It was <th> in the source file and it is <td> here. The library keeps the grid and drops the distinction between a header cell and a data cell, so a program that wants the column names has to take the first row and hope.

A CSV is a table

from unstructured.partition.auto import partition

elements = partition("orders.csv")
print(len(elements), elements[0].category)
print(elements[0].metadata.text_as_html)

One element for the whole file. The empty courier cell in the second row came back as an empty tag rather than being skipped, so the columns still line up.