document_parser

langroid/parsing/document_parser.py

DocumentParser(source, config)

Bases: Parser

Abstract base class for extracting text from special types of docs such as PDFs or Docx.

Attributes:

Name	Type	Description
source	str	The source, either a URL or a file path.
doc_bytes	BytesIO	BytesIO object containing the doc data.

a path, a URL or a bytes object.

Source code in langroid/parsing/document_parser.py

def __init__(self, source: str | bytes, config: ParsingConfig):
    """
    Args:
        source (str|bytes): The source, which could be
        a path, a URL or a bytes object.
    """
    super().__init__(config)
    self.config = config
    if isinstance(source, bytes):
        self.source = "bytes"
        self.doc_bytes = BytesIO(source)
    else:
        self.source = source
        self.doc_bytes = self._load_doc_as_bytesio()

create(source, config, doc_type=None) classmethod

Create a DocumentParser instance based on source type and config..library specified.

Parameters:

Name	Type	Description	Default
source	str | bytes	The source, could be a URL, file path, or bytes object.	required
config	ParserConfig	The parser configuration.	required
doc_type	str | None	The type of document, if known	None

Returns:

Name	Type	Description
DocumentParser	DocumentParser	An instance of a DocumentParser subclass.

Source code in langroid/parsing/document_parser.py

@classmethod
def create(
    cls,
    source: str | bytes,
    config: ParsingConfig,
    doc_type: str | DocumentType | None = None,
) -> "DocumentParser":
    """
    Create a DocumentParser instance based on source type
        and config.<source_type>.library specified.

    Args:
        source (str|bytes): The source, could be a URL, file path,
            or bytes object.
        config (ParserConfig): The parser configuration.
        doc_type (str|None): The type of document, if known

    Returns:
        DocumentParser: An instance of a DocumentParser subclass.
    """
    if DocumentParser._document_type(source, doc_type) == DocumentType.PDF:
        if config.pdf.library == "fitz":
            return FitzPDFParser(source, config)
        elif config.pdf.library == "pypdf":
            return PyPDFParser(source, config)
        elif config.pdf.library == "pdfplumber":
            return PDFPlumberParser(source, config)
        elif config.pdf.library == "unstructured":
            return UnstructuredPDFParser(source, config)
        elif config.pdf.library == "pdf2image":
            return ImagePdfParser(source, config)
        else:
            raise ValueError(
                f"Unsupported PDF library specified: {config.pdf.library}"
            )
    elif DocumentParser._document_type(source, doc_type) == DocumentType.DOCX:
        if config.docx.library == "unstructured":
            return UnstructuredDocxParser(source, config)
        elif config.docx.library == "python-docx":
            return PythonDocxParser(source, config)
        else:
            raise ValueError(
                f"Unsupported DOCX library specified: {config.docx.library}"
            )
    elif DocumentParser._document_type(source, doc_type) == DocumentType.DOC:
        return UnstructuredDocParser(source, config)
    else:
        source_name = source if isinstance(source, str) else "bytes"
        raise ValueError(f"Unsupported document type: {source_name}")

chunks_from_path_or_bytes(source, parser, doc_type=None, lines=None) staticmethod

Get document chunks from a file path or bytes object. Args: source (str|bytes): The source, which could be a URL, path or bytes object. parser (Parser): The parser instance (for splitting the document). doc_type (str|DocumentType|None): The type of document, if known. lines (int|None): The number of lines to read from a plain text file. Returns: List[Document]: A list of Document objects, each containing a chunk of text, determined by the chunking and splitting settings in the parser config.

Source code in langroid/parsing/document_parser.py

@staticmethod
def chunks_from_path_or_bytes(
    source: str | bytes,
    parser: Parser,
    doc_type: str | DocumentType | None = None,
    lines: int | None = None,
) -> List[Document]:
    """
    Get document chunks from a file path or bytes object.
    Args:
        source (str|bytes): The source, which could be a URL, path or bytes object.
        parser (Parser): The parser instance (for splitting the document).
        doc_type (str|DocumentType|None): The type of document, if known.
        lines (int|None): The number of lines to read from a plain text file.
    Returns:
        List[Document]: A list of `Document` objects,
            each containing a chunk of text, determined by the
            chunking and splitting settings in the parser config.
    """
    dtype: DocumentType = DocumentParser._document_type(source, doc_type)
    if dtype in [DocumentType.PDF, DocumentType.DOC, DocumentType.DOCX]:
        doc_parser = DocumentParser.create(
            source,
            parser.config,
            doc_type=doc_type,
        )
        chunks = doc_parser.get_doc_chunks()
        if len(chunks) == 0 and dtype == DocumentType.PDF:
            doc_parser = ImagePdfParser(source, parser.config)
            chunks = doc_parser.get_doc_chunks()
        return chunks
    else:
        # try getting as plain text; these will be chunked downstream
        # -- could be a bytes object or a path
        if isinstance(source, bytes):
            content = source.decode()
            if lines is not None:
                file_lines = content.splitlines()[:lines]
                content = "\n".join(line.strip() for line in file_lines)
        else:
            with open(source, "r") as f:
                if lines is not None:
                    file_lines = list(itertools.islice(f, lines))
                    content = "\n".join(line.strip() for line in file_lines)
                else:
                    content = f.read()
        soup = BeautifulSoup(content, "html.parser")
        text = soup.get_text()
        source_name = source if isinstance(source, str) else "bytes"
        doc = Document(
            content=text,
            metadata=DocMetaData(source=str(source_name)),
        )
        return parser.split([doc])

iterate_pages()

Yield each page in the PDF.

Source code in langroid/parsing/document_parser.py

def iterate_pages(self) -> Generator[Tuple[int, Any], None, None]:
    """Yield each page in the PDF."""
    raise NotImplementedError

extract_text_from_page(page)

Extract text from a given page.

Source code in langroid/parsing/document_parser.py

def extract_text_from_page(self, page: Any) -> str:
    """Extract text from a given page."""
    raise NotImplementedError

fix_text(text)

Fix text extracted from a PDF.

Parameters:

Name	Type	Description	Default
text	str	The extracted text.	required

Returns:

Name	Type	Description
str	str	The fixed text.

Source code in langroid/parsing/document_parser.py

def fix_text(self, text: str) -> str:
    """
    Fix text extracted from a PDF.

    Args:
        text (str): The extracted text.

    Returns:
        str: The fixed text.
    """
    # Some pdf parsers introduce extra space before hyphen,
    # so use regular expression to replace 'space-hyphen' with just 'hyphen'
    return re.sub(r" +\-", "-", text)

get_doc()

Get entire text from source as a single document.

Returns:

Type	Description
Document	a Document object containing the content of the pdf file, and metadata containing source name (URL or path)

Source code in langroid/parsing/document_parser.py

def get_doc(self) -> Document:
    """
    Get entire text from source as a single document.

    Returns:
        a `Document` object containing the content of the pdf file,
            and metadata containing source name (URL or path)
    """

    text = "".join(
        [self.extract_text_from_page(page) for _, page in self.iterate_pages()]
    )
    return Document(content=text, metadata=DocMetaData(source=self.source))

get_doc_chunks()

Get document chunks from a pdf source, with page references in the document metadata.

Adapted from https://github.com/whitead/paper-qa/blob/main/paperqa/readers.py

Returns:

Type	Description
List[Document]	List[Document]: a list of Document objects, each containing a chunk of text

Source code in langroid/parsing/document_parser.py

def get_doc_chunks(self) -> List[Document]:
    """
    Get document chunks from a pdf source,
    with page references in the document metadata.

    Adapted from
    https://github.com/whitead/paper-qa/blob/main/paperqa/readers.py

    Returns:
        List[Document]: a list of `Document` objects,
            each containing a chunk of text
    """

    split = []  # tokens in curr split
    pages: List[str] = []
    docs: List[Document] = []
    for i, page in self.iterate_pages():
        page_text = self.extract_text_from_page(page)
        split += self.tokenizer.encode(page_text)
        pages.append(str(i + 1))
        # split could be so long it needs to be split
        # into multiple chunks. Or it could be so short
        # that it needs to be combined with the next chunk.
        while len(split) > self.config.chunk_size:
            # pretty formatting of pages (e.g. 1-3, 4, 5-7)
            pg = "-".join([pages[0], pages[-1]])
            text = self.tokenizer.decode(split[: self.config.chunk_size])
            docs.append(
                Document(
                    content=text,
                    metadata=DocMetaData(
                        source=f"{self.source} pages {pg}",
                        is_chunk=True,
                    ),
                )
            )
            split = split[self.config.chunk_size - self.config.overlap :]
            pages = [str(i + 1)]
    if len(split) > self.config.overlap:
        pg = "-".join([pages[0], pages[-1]])
        text = self.tokenizer.decode(split[: self.config.chunk_size])
        docs.append(
            Document(
                content=text,
                metadata=DocMetaData(
                    source=f"{self.source} pages {pg}",
                    is_chunk=True,
                ),
            )
        )
    self.add_window_ids(docs)
    return docs

FitzPDFParser(source, config)

Bases: DocumentParser

Parser for processing PDFs using the fitz library.

a path, a URL or a bytes object.

Source code in langroid/parsing/document_parser.py

def __init__(self, source: str | bytes, config: ParsingConfig):
    """
    Args:
        source (str|bytes): The source, which could be
        a path, a URL or a bytes object.
    """
    super().__init__(config)
    self.config = config
    if isinstance(source, bytes):
        self.source = "bytes"
        self.doc_bytes = BytesIO(source)
    else:
        self.source = source
        self.doc_bytes = self._load_doc_as_bytesio()

iterate_pages()

Yield each page in the PDF using fitz.

Returns:

Type	Description
Generator[Tuple[int, Page], None, None]	Generator[fitz.Page]: Generator yielding each page.

Source code in langroid/parsing/document_parser.py

def iterate_pages(self) -> Generator[Tuple[int, fitz.Page], None, None]:
    """
    Yield each page in the PDF using `fitz`.

    Returns:
        Generator[fitz.Page]: Generator yielding each page.
    """
    doc = fitz.open(stream=self.doc_bytes, filetype="pdf")
    for i, page in enumerate(doc):
        yield i, page
    doc.close()

extract_text_from_page(page)

Extract text from a given fitz page.

Parameters:

Name	Type	Description	Default
page	Page	The fitz page object.	required

Returns:

Name	Type	Description
str	str	Extracted text from the page.

Source code in langroid/parsing/document_parser.py

def extract_text_from_page(self, page: fitz.Page) -> str:
    """
    Extract text from a given `fitz` page.

    Args:
        page (fitz.Page): The `fitz` page object.

    Returns:
        str: Extracted text from the page.
    """
    return self.fix_text(page.get_text())

PyPDFParser(source, config)

Bases: DocumentParser

Parser for processing PDFs using the pypdf library.

a path, a URL or a bytes object.

Source code in langroid/parsing/document_parser.py

def __init__(self, source: str | bytes, config: ParsingConfig):
    """
    Args:
        source (str|bytes): The source, which could be
        a path, a URL or a bytes object.
    """
    super().__init__(config)
    self.config = config
    if isinstance(source, bytes):
        self.source = "bytes"
        self.doc_bytes = BytesIO(source)
    else:
        self.source = source
        self.doc_bytes = self._load_doc_as_bytesio()

iterate_pages()

Yield each page in the PDF using pypdf.

Returns:

Type	Description
Generator[Tuple[int, PageObject], None, None]	Generator[pypdf.pdf.PageObject]: Generator yielding each page.

Source code in langroid/parsing/document_parser.py

def iterate_pages(self) -> Generator[Tuple[int, pypdf.PageObject], None, None]:
    """
    Yield each page in the PDF using `pypdf`.

    Returns:
        Generator[pypdf.pdf.PageObject]: Generator yielding each page.
    """
    reader = pypdf.PdfReader(self.doc_bytes)
    for i, page in enumerate(reader.pages):
        yield i, page

extract_text_from_page(page)

Extract text from a given pypdf page.

Parameters:

Name	Type	Description	Default
page	PageObject	The pypdf page object.	required

Returns:

Name	Type	Description
str	str	Extracted text from the page.

Source code in langroid/parsing/document_parser.py

def extract_text_from_page(self, page: pypdf.PageObject) -> str:
    """
    Extract text from a given `pypdf` page.

    Args:
        page (pypdf.pdf.PageObject): The `pypdf` page object.

    Returns:
        str: Extracted text from the page.
    """
    return self.fix_text(page.extract_text())

PDFPlumberParser(source, config)

Bases: DocumentParser

Parser for processing PDFs using the pdfplumber library.

a path, a URL or a bytes object.

Source code in langroid/parsing/document_parser.py

def __init__(self, source: str | bytes, config: ParsingConfig):
    """
    Args:
        source (str|bytes): The source, which could be
        a path, a URL or a bytes object.
    """
    super().__init__(config)
    self.config = config
    if isinstance(source, bytes):
        self.source = "bytes"
        self.doc_bytes = BytesIO(source)
    else:
        self.source = source
        self.doc_bytes = self._load_doc_as_bytesio()

iterate_pages()

Yield each page in the PDF using pdfplumber.

Returns:

Type	Description
Generator[Tuple[int, Page], None, None]	Generator[pdfplumber.Page]: Generator yielding each page.

Source code in langroid/parsing/document_parser.py

def iterate_pages(
    self,
) -> (Generator)[Tuple[int, pdfplumber.pdf.Page], None, None]:  # type: ignore
    """
    Yield each page in the PDF using `pdfplumber`.

    Returns:
        Generator[pdfplumber.Page]: Generator yielding each page.
    """
    with pdfplumber.open(self.doc_bytes) as pdf:
        for i, page in enumerate(pdf.pages):
            yield i, page

extract_text_from_page(page)

Extract text from a given pdfplumber page.

Parameters:

Name	Type	Description	Default
page	Page	The pdfplumber page object.	required

Returns:

Name	Type	Description
str	str	Extracted text from the page.

Source code in langroid/parsing/document_parser.py

def extract_text_from_page(self, page: pdfplumber.pdf.Page) -> str:  # type: ignore
    """
    Extract text from a given `pdfplumber` page.

    Args:
        page (pdfplumber.Page): The `pdfplumber` page object.

    Returns:
        str: Extracted text from the page.
    """
    return self.fix_text(page.extract_text())

ImagePdfParser(source, config)

Bases: DocumentParser

Parser for processing PDFs that are images, i.e. not "true" PDFs.

a path, a URL or a bytes object.

Source code in langroid/parsing/document_parser.py

def __init__(self, source: str | bytes, config: ParsingConfig):
    """
    Args:
        source (str|bytes): The source, which could be
        a path, a URL or a bytes object.
    """
    super().__init__(config)
    self.config = config
    if isinstance(source, bytes):
        self.source = "bytes"
        self.doc_bytes = BytesIO(source)
    else:
        self.source = source
        self.doc_bytes = self._load_doc_as_bytesio()

extract_text_from_page(page)

Extract text from a given pdf2image page.

Parameters:

Name	Type	Description	Default
page	Image	The PIL Image object.	required

Returns:

Name	Type	Description
str	str	Extracted text from the image.

Source code in langroid/parsing/document_parser.py

def extract_text_from_page(self, page: "Image") -> str:  # type: ignore
    """
    Extract text from a given `pdf2image` page.

    Args:
        page (Image): The PIL Image object.

    Returns:
        str: Extracted text from the image.
    """
    import pytesseract

    text = pytesseract.image_to_string(page)
    return self.fix_text(text)

UnstructuredPDFParser(source, config)

Bases: DocumentParser

Parser for processing PDF files using the unstructured library.

a path, a URL or a bytes object.

Source code in langroid/parsing/document_parser.py

def __init__(self, source: str | bytes, config: ParsingConfig):
    """
    Args:
        source (str|bytes): The source, which could be
        a path, a URL or a bytes object.
    """
    super().__init__(config)
    self.config = config
    if isinstance(source, bytes):
        self.source = "bytes"
        self.doc_bytes = BytesIO(source)
    else:
        self.source = source
        self.doc_bytes = self._load_doc_as_bytesio()

extract_text_from_page(page)

Extract text from a given unstructured element.

Parameters:

Name	Type	Description	Default
page	unstructured element	The unstructured element object.	required

Returns:

Name	Type	Description
str	str	Extracted text from the element.

Source code in langroid/parsing/document_parser.py

def extract_text_from_page(self, page: Any) -> str:
    """
    Extract text from a given `unstructured` element.

    Args:
        page (unstructured element): The `unstructured` element object.

    Returns:
        str: Extracted text from the element.
    """
    text = " ".join(el.text for el in page)
    return self.fix_text(text)

UnstructuredDocxParser(source, config)

Bases: DocumentParser

Parser for processing DOCX files using the unstructured library.

a path, a URL or a bytes object.

Source code in langroid/parsing/document_parser.py

def __init__(self, source: str | bytes, config: ParsingConfig):
    """
    Args:
        source (str|bytes): The source, which could be
        a path, a URL or a bytes object.
    """
    super().__init__(config)
    self.config = config
    if isinstance(source, bytes):
        self.source = "bytes"
        self.doc_bytes = BytesIO(source)
    else:
        self.source = source
        self.doc_bytes = self._load_doc_as_bytesio()

extract_text_from_page(page)

Extract text from a given unstructured element.

Note

The concept of "pages" doesn't actually exist in the .docx file format in the same way it does in formats like .pdf. A .docx file is made up of a series of elements like paragraphs and tables, but the division into pages is done dynamically based on the rendering settings (like the page size, margin size, font size, etc.).

Parameters:

Name	Type	Description	Default
page	unstructured element	The unstructured element object.	required

Returns:

Name	Type	Description
str	str	Extracted text from the element.

Source code in langroid/parsing/document_parser.py

def extract_text_from_page(self, page: Any) -> str:
    """
    Extract text from a given `unstructured` element.

    Note:
        The concept of "pages" doesn't actually exist in the .docx file format in
        the same way it does in formats like .pdf. A .docx file is made up of a
        series of elements like paragraphs and tables, but the division into
        pages is done dynamically based on the rendering settings (like the page
        size, margin size, font size, etc.).

    Args:
        page (unstructured element): The `unstructured` element object.

    Returns:
        str: Extracted text from the element.
    """
    text = " ".join(el.text for el in page)
    return self.fix_text(text)

PythonDocxParser(source, config)

Bases: DocumentParser

Parser for processing DOCX files using the python-docx library.

a path, a URL or a bytes object.

Source code in langroid/parsing/document_parser.py

def __init__(self, source: str | bytes, config: ParsingConfig):
    """
    Args:
        source (str|bytes): The source, which could be
        a path, a URL or a bytes object.
    """
    super().__init__(config)
    self.config = config
    if isinstance(source, bytes):
        self.source = "bytes"
        self.doc_bytes = BytesIO(source)
    else:
        self.source = source
        self.doc_bytes = self._load_doc_as_bytesio()

iterate_pages()

Simulate iterating through pages. In a DOCX file, pages are not explicitly defined, so we consider each paragraph as a separate 'page' for simplicity.

Source code in langroid/parsing/document_parser.py

def iterate_pages(self) -> Generator[Tuple[int, Any], None, None]:
    """
    Simulate iterating through pages.
    In a DOCX file, pages are not explicitly defined,
    so we consider each paragraph as a separate 'page' for simplicity.
    """
    import docx

    doc = docx.Document(self.doc_bytes)
    for i, para in enumerate(doc.paragraphs, start=1):
        yield i, [para]

extract_text_from_page(page)

Extract text from a given 'page', which in this case is a single paragraph.

Parameters:

Name	Type	Description	Default
page	list	A list containing a single Paragraph object.	required

Returns:

Name	Type	Description
str	str	Extracted text from the paragraph.

Source code in langroid/parsing/document_parser.py

def extract_text_from_page(self, page: Any) -> str:
    """
    Extract text from a given 'page', which in this case is a single paragraph.

    Args:
        page (list): A list containing a single Paragraph object.

    Returns:
        str: Extracted text from the paragraph.
    """
    paragraph = page[0]
    return self.fix_text(paragraph.text)

find_last_full_char(possible_unicode)

Find the index of the last full character in a byte string. Args: possible_unicode (bytes): The bytes to check. Returns: int: The index of the last full unicode character.

Source code in langroid/parsing/document_parser.py

def find_last_full_char(possible_unicode: bytes) -> int:
    """
    Find the index of the last full character in a byte string.
    Args:
        possible_unicode (bytes): The bytes to check.
    Returns:
        int: The index of the last full unicode character.
    """

    for i in range(len(possible_unicode) - 1, 0, -1):
        if (possible_unicode[i] & 0xC0) != 0x80:
            return i
    return 0

is_plain_text(path_or_bytes)

Check if a file is plain text by attempting to decode it as UTF-8. Args: path_or_bytes (str|bytes): The file path or bytes object. Returns: bool: True if the file is plain text, False otherwise.

Source code in langroid/parsing/document_parser.py

def is_plain_text(path_or_bytes: str | bytes) -> bool:
    """
    Check if a file is plain text by attempting to decode it as UTF-8.
    Args:
        path_or_bytes (str|bytes): The file path or bytes object.
    Returns:
        bool: True if the file is plain text, False otherwise.
    """
    if isinstance(path_or_bytes, str):
        if path_or_bytes.startswith(("http://", "https://")):
            response = requests.get(path_or_bytes)
            response.raise_for_status()
            content = response.content[:1024]
        else:
            with open(path_or_bytes, "rb") as f:
                content = f.read(1024)
    else:
        content = path_or_bytes[:1024]
    try:
        # Attempt to decode the content as UTF-8
        content = content[: find_last_full_char(content)]

        _ = content.decode("utf-8")
        # Additional checks can go here, e.g., to verify that the content
        # doesn't contain too many unusual characters for it to be considered text
        return True
    except UnicodeDecodeError:
        # If decoding fails, it's likely not plain text (or not encoded in UTF-8)
        return False