pypreprocess 1.4.3

Preprocess SDK

Preprocess SDK

Preprocess is an API service that splits any kind of document into optimal chunks of text for use in language model tasks. Given documents in input Preprocess splits them into chunks of text that respect the layout and semantics of the original document. We split the content by taking into account sections, paragraphs, lists, images, data tables, text tables, and slides, and following the content semantics for long texts. We support:

• PDFs
• Microsoft Office documents (Word, PowerPoint, Excel)
• OpenOffice documents (ods, odt, odp)
• HTML content (web pages, articles, emails)
• Plain text

Installation

Install the Python Preprocess library if it is not already present:

pip install pypreprocess

Alternatively, if you want to add it as a dependency with poetry:

poetry add pypreprocess
poetry install

You need a Preprocess API Key to use the SDK, to get one please reach out to [e-mail] asking for an API key.

Getting started

Get the chunks from a file for use in your language model tasks.

from pypreprocess import Preprocess

#init the SDK with a file
preprocess = Preprocess(api_key=YOUR_API_KEY, filepath="path/for/file")

#chunk the file
preprocess.chunk()
preprocess.wait()

#get the result
result = preprocess.result()
for i in result.data['chunks']:
    #use chunks

Initialise a connection

You can initialize the SDK in 3 ways.

1- Passing a local filepath

when you want to init the SDK to chunk a local file

from pypreprocess import Preprocess
preprocess = Preprocess(api_key=YOUR_API_KEY, filepath="path/for/file")

2- Passing a process_id

when the chunking process starts Preprocess generate a process_id, you can use it to instance the SDK

from pypreprocess import Preprocess
preprocess = Preprocess(api_key=YOUR_API_KEY, process_id="id_of_the_process")

3- Passing a PreprocessResponse Object

When you need to store the result of a chunking process permanently, you can later load them in the SDK via a PreprocessResponse object

import json
from pypreprocess import Preprocess, ProcessResponse
response = PreprocessResponse(**json.loads("The JSON result from calling chunk before."))
preprocess = Preprocess(api_key=YOUR_API_KEY, process=PreprocessResponse)

Chunking options

We support a few options you can configure to get the best result for your ingestion pipeline.

Preprocess tries to output chunks of less than 512 tokens. Sometimes longer chunks can be produced to preserve the content integrity. We are currently working to allow you to set an arbitrary chunk token length, stay tuned.

Parameter	Type	Default	Description
merge	bool	False	If True small paragraphs will be merged to maximize chunk length
repeat_title	bool	False	If True each chunk will start with the title of the section in which is contained
repeat_table_header	bool	False	If True each chunk will start with the header of the table in which is contained
table_output_format	enum ['text', 'markdown', 'html']	'text'	Return tables in the format you need for your ingestion pipelines
keep_header	bool	True	If set to False, the content of the headers will be removed. Headers may include page numbers, document titles, section titles, paragraph titles, and fixed layout elements.
smart_header	bool	True	If set to True, only relevant titles will be included in the chunks, while other information will be removed. Relevant titles are those that should be part of the body of the page as a title. If set to False, only the keep_header parameter will be considered. If keep_header is False, the smart_header parameter will be ignored.
keep_footer	bool	False	If set to True, the content of the footers will be included in the chunks. Footers may include page numbers, footnotes, and fixed layout elements.
image_text	bool	False	If set to True, the text contained in the images will be added to the chunks.

You can pass each parameter from those during the SDK initialization

preprocess = Preprocess(api_key=YOUR_API_KEY, filepath="path/for/file", merge=True, repeat_title=True, ...)

Or set them later with the set_options method using a dict

preprocess.set_options({"merge": True, "repeat_title": True, ...})

Chunk

After initing the SDK with a filepath you should call the chunk() method.

from pypreprocess import Preprocess
preprocess = Preprocess(api_key=YOUR_API_KEY, filepath="path/for/file")
response = preprocess.chunk()

the response contains the process_id and details about the success of the API call.

Getting the results

The conversion and chunking process may take a while. You can use the built-in wait() method to wait and get the result as the process finishes.

result = preprocess.wait()
print(result.data['chunks'])

In a more complex scenario, you can store the process_id after initiating the chunking process and then use it in a different flow.

#initing the chunking process
from pypreprocess import Preprocess
preprocess = Preprocess(api_key=YOUR_API_KEY, filepath="path/for/file")
preprocess.chunk()
preocess_id = preprocess.get_process_id()

#in a different flow
from pypreprocess import Preprocess
preprocess = Preprocess(api_key=YOUR_API_KEY, process_id=process_id)
result = preprocess.wait()
print(result.data['chunks'])

If you want to implement a different logic for getting the result you can use the method result() and check if the status if it's FINISHED.

result = preprocess.result()
if result.data['process']['status'] == "FINISHED": 
   print(result.data['chunks'])

Other methods

Here is a list of some methods inside the SDK that may help you

• set_filepath(path) setting the file path after initializing the object.
• set_process_id(id) setting the process_id param by id.
• set_process(PreprocessResponse) setting the process_id param by PreprocessResponse object.
• set_options(dict) set the parameters for configuring chunking options.
• to_json() returning a JSON str representing the current object.
• get_process_id() returning the current process_id.
• get_filepath() returning the filepath.
• get_options() returning the current chunking options.