387 lines
12 KiB
Python
Raw Normal View History

import pytest
from PIL import Image
from unstructured_inference.inference import layout
from unstructured_inference.inference.layout import LayoutElement
from unstructured_inference.inference.layoutelement import LocationlessLayoutElement
from unstructured.documents.coordinates import PixelSpace
from unstructured.documents.elements import (
CheckBox,
Feat: Create a naive hierarchy for elements (#1268) ## **Summary** By adding hierarchy to unstructured elements, users will have more information for implementing vector db/LLM chunking strategies. For example, text elements could be queried by their preceding title element. The hierarchy is implemented by a parent_id tag in the element's metadata. ### Features - Introduces a parent_id to ElementMetadata (The id of the parent element, not a pointer) - Creates a rule set for assigning hierarchies. Sensible default is assigned, with an optional override parameter - Sets element parent ids if there isn't an existing parent id or matches the ruleset ### How it works Hierarchies are assigned via a parent id field in element metadata. Elements are read sequentially and evaluated against a ruleset. For example take the following elements: 1. Title, "This is the Title" 2. Text, "this is the text" And the ruleset: `{"title": ["text"]}`. When evaluated, the parent_id of 2 will be the id of 1. The algorithm for determining this is more complex and resolves several edge cases, so please read the code for further details. ### Schema Changes ``` @dataclass class ElementMetadata: coordinates: Optional[CoordinatesMetadata] = None data_source: Optional[DataSourceMetadata] = None filename: Optional[str] = None file_directory: Optional[str] = None last_modified: Optional[str] = None filetype: Optional[str] = None attached_to_filename: Optional[str] = None + parent_id: Optional[Union[str, uuid.UUID, NoID, UUID]] = None + category_depth: Optional[int] = None ... ``` ### Testing ``` from unstructured.partition.auto import partition from typing import List elements = partition(filename="./unstructured/example-docs/fake-html.html", strategy="auto") for element in elements: print( f"Category: {getattr(element, 'category', '')}\n"\ f"Text: {getattr(element, 'text', '')}\n" f"ID: {element.id}\n" \ f"Parent ID: {element.metadata.parent_id}\n"\ f"Depth: {element.metadata.category_depth}\n" \ ) ``` ### Additional Notes Implementing this feature revealed a possibly undesired side-effect in how element metadata are processed. In `unstructured/partition/common.py` the `_add_element_metadata` is invoked as part of the `add_metadata_with_filetype` decorator for filetype partitioning. This method is intended to add additional information to the metadata generated with the element including filename and filetype, however the existing metadata is merged into a newly created metadata object rather than the other way around. Because of the way it's structured, new metadata fields can easily be forgotten and pose debugging challenges to developers. This likely warrants a new issue. I'm guessing that the implementation is done this way to avoid issues with deserializing elements, but could be wrong. --------- Co-authored-by: Benjamin Torres <benjats07@users.noreply.github.com>
2023-09-14 11:23:16 -04:00
ElementMetadata,
FigureCaption,
Feat: Create a naive hierarchy for elements (#1268) ## **Summary** By adding hierarchy to unstructured elements, users will have more information for implementing vector db/LLM chunking strategies. For example, text elements could be queried by their preceding title element. The hierarchy is implemented by a parent_id tag in the element's metadata. ### Features - Introduces a parent_id to ElementMetadata (The id of the parent element, not a pointer) - Creates a rule set for assigning hierarchies. Sensible default is assigned, with an optional override parameter - Sets element parent ids if there isn't an existing parent id or matches the ruleset ### How it works Hierarchies are assigned via a parent id field in element metadata. Elements are read sequentially and evaluated against a ruleset. For example take the following elements: 1. Title, "This is the Title" 2. Text, "this is the text" And the ruleset: `{"title": ["text"]}`. When evaluated, the parent_id of 2 will be the id of 1. The algorithm for determining this is more complex and resolves several edge cases, so please read the code for further details. ### Schema Changes ``` @dataclass class ElementMetadata: coordinates: Optional[CoordinatesMetadata] = None data_source: Optional[DataSourceMetadata] = None filename: Optional[str] = None file_directory: Optional[str] = None last_modified: Optional[str] = None filetype: Optional[str] = None attached_to_filename: Optional[str] = None + parent_id: Optional[Union[str, uuid.UUID, NoID, UUID]] = None + category_depth: Optional[int] = None ... ``` ### Testing ``` from unstructured.partition.auto import partition from typing import List elements = partition(filename="./unstructured/example-docs/fake-html.html", strategy="auto") for element in elements: print( f"Category: {getattr(element, 'category', '')}\n"\ f"Text: {getattr(element, 'text', '')}\n" f"ID: {element.id}\n" \ f"Parent ID: {element.metadata.parent_id}\n"\ f"Depth: {element.metadata.category_depth}\n" \ ) ``` ### Additional Notes Implementing this feature revealed a possibly undesired side-effect in how element metadata are processed. In `unstructured/partition/common.py` the `_add_element_metadata` is invoked as part of the `add_metadata_with_filetype` decorator for filetype partitioning. This method is intended to add additional information to the metadata generated with the element including filename and filetype, however the existing metadata is merged into a newly created metadata object rather than the other way around. Because of the way it's structured, new metadata fields can easily be forgotten and pose debugging challenges to developers. This likely warrants a new issue. I'm guessing that the implementation is done this way to avoid issues with deserializing elements, but could be wrong. --------- Co-authored-by: Benjamin Torres <benjats07@users.noreply.github.com>
2023-09-14 11:23:16 -04:00
Header,
ListItem,
NarrativeText,
Text,
Title,
)
from unstructured.partition import common
from unstructured.partition.common import (
_get_page_image_metadata,
contains_emoji,
document_to_element_list,
)
class MockPageLayout(layout.PageLayout):
def __init__(self, number: int, image: Image):
self.number = number
self.image = image
@property
def elements(self):
return [
LocationlessLayoutElement(
type="Headline",
text="Charlie Brown and the Great Pumpkin",
),
]
class MockDocumentLayout(layout.DocumentLayout):
@property
def pages(self):
return [
MockPageLayout(number=1, image=Image.new("1", (1, 1))),
]
def test_normalize_layout_element_dict():
layout_element = {
"type": "Title",
"coordinates": [[1, 2], [3, 4], [5, 6], [7, 8]],
"coordinate_system": None,
"text": "Some lovely text",
}
coordinate_system = PixelSpace(width=10, height=20)
Feat: Create a naive hierarchy for elements (#1268) ## **Summary** By adding hierarchy to unstructured elements, users will have more information for implementing vector db/LLM chunking strategies. For example, text elements could be queried by their preceding title element. The hierarchy is implemented by a parent_id tag in the element's metadata. ### Features - Introduces a parent_id to ElementMetadata (The id of the parent element, not a pointer) - Creates a rule set for assigning hierarchies. Sensible default is assigned, with an optional override parameter - Sets element parent ids if there isn't an existing parent id or matches the ruleset ### How it works Hierarchies are assigned via a parent id field in element metadata. Elements are read sequentially and evaluated against a ruleset. For example take the following elements: 1. Title, "This is the Title" 2. Text, "this is the text" And the ruleset: `{"title": ["text"]}`. When evaluated, the parent_id of 2 will be the id of 1. The algorithm for determining this is more complex and resolves several edge cases, so please read the code for further details. ### Schema Changes ``` @dataclass class ElementMetadata: coordinates: Optional[CoordinatesMetadata] = None data_source: Optional[DataSourceMetadata] = None filename: Optional[str] = None file_directory: Optional[str] = None last_modified: Optional[str] = None filetype: Optional[str] = None attached_to_filename: Optional[str] = None + parent_id: Optional[Union[str, uuid.UUID, NoID, UUID]] = None + category_depth: Optional[int] = None ... ``` ### Testing ``` from unstructured.partition.auto import partition from typing import List elements = partition(filename="./unstructured/example-docs/fake-html.html", strategy="auto") for element in elements: print( f"Category: {getattr(element, 'category', '')}\n"\ f"Text: {getattr(element, 'text', '')}\n" f"ID: {element.id}\n" \ f"Parent ID: {element.metadata.parent_id}\n"\ f"Depth: {element.metadata.category_depth}\n" \ ) ``` ### Additional Notes Implementing this feature revealed a possibly undesired side-effect in how element metadata are processed. In `unstructured/partition/common.py` the `_add_element_metadata` is invoked as part of the `add_metadata_with_filetype` decorator for filetype partitioning. This method is intended to add additional information to the metadata generated with the element including filename and filetype, however the existing metadata is merged into a newly created metadata object rather than the other way around. Because of the way it's structured, new metadata fields can easily be forgotten and pose debugging challenges to developers. This likely warrants a new issue. I'm guessing that the implementation is done this way to avoid issues with deserializing elements, but could be wrong. --------- Co-authored-by: Benjamin Torres <benjats07@users.noreply.github.com>
2023-09-14 11:23:16 -04:00
element = common.normalize_layout_element(
layout_element,
coordinate_system=coordinate_system,
)
assert element == Title(
text="Some lovely text",
coordinates=[[1, 2], [3, 4], [5, 6], [7, 8]],
coordinate_system=coordinate_system,
)
def test_normalize_layout_element_dict_caption():
layout_element = {
"type": "Figure",
"coordinates": [[1, 2], [3, 4], [5, 6], [7, 8]],
"text": "Some lovely text",
}
coordinate_system = PixelSpace(width=10, height=20)
Feat: Create a naive hierarchy for elements (#1268) ## **Summary** By adding hierarchy to unstructured elements, users will have more information for implementing vector db/LLM chunking strategies. For example, text elements could be queried by their preceding title element. The hierarchy is implemented by a parent_id tag in the element's metadata. ### Features - Introduces a parent_id to ElementMetadata (The id of the parent element, not a pointer) - Creates a rule set for assigning hierarchies. Sensible default is assigned, with an optional override parameter - Sets element parent ids if there isn't an existing parent id or matches the ruleset ### How it works Hierarchies are assigned via a parent id field in element metadata. Elements are read sequentially and evaluated against a ruleset. For example take the following elements: 1. Title, "This is the Title" 2. Text, "this is the text" And the ruleset: `{"title": ["text"]}`. When evaluated, the parent_id of 2 will be the id of 1. The algorithm for determining this is more complex and resolves several edge cases, so please read the code for further details. ### Schema Changes ``` @dataclass class ElementMetadata: coordinates: Optional[CoordinatesMetadata] = None data_source: Optional[DataSourceMetadata] = None filename: Optional[str] = None file_directory: Optional[str] = None last_modified: Optional[str] = None filetype: Optional[str] = None attached_to_filename: Optional[str] = None + parent_id: Optional[Union[str, uuid.UUID, NoID, UUID]] = None + category_depth: Optional[int] = None ... ``` ### Testing ``` from unstructured.partition.auto import partition from typing import List elements = partition(filename="./unstructured/example-docs/fake-html.html", strategy="auto") for element in elements: print( f"Category: {getattr(element, 'category', '')}\n"\ f"Text: {getattr(element, 'text', '')}\n" f"ID: {element.id}\n" \ f"Parent ID: {element.metadata.parent_id}\n"\ f"Depth: {element.metadata.category_depth}\n" \ ) ``` ### Additional Notes Implementing this feature revealed a possibly undesired side-effect in how element metadata are processed. In `unstructured/partition/common.py` the `_add_element_metadata` is invoked as part of the `add_metadata_with_filetype` decorator for filetype partitioning. This method is intended to add additional information to the metadata generated with the element including filename and filetype, however the existing metadata is merged into a newly created metadata object rather than the other way around. Because of the way it's structured, new metadata fields can easily be forgotten and pose debugging challenges to developers. This likely warrants a new issue. I'm guessing that the implementation is done this way to avoid issues with deserializing elements, but could be wrong. --------- Co-authored-by: Benjamin Torres <benjats07@users.noreply.github.com>
2023-09-14 11:23:16 -04:00
element = common.normalize_layout_element(
layout_element,
coordinate_system=coordinate_system,
)
assert element == FigureCaption(
text="Some lovely text",
coordinates=[[1, 2], [3, 4], [5, 6], [7, 8]],
coordinate_system=coordinate_system,
)
def test_normalize_layout_element_dict_figure_caption():
layout_element = {
"type": "FigureCaption",
"coordinates": [[1, 2], [3, 4], [5, 6], [7, 8]],
"text": "Some lovely text",
}
coordinate_system = PixelSpace(width=10, height=20)
Feat: Create a naive hierarchy for elements (#1268) ## **Summary** By adding hierarchy to unstructured elements, users will have more information for implementing vector db/LLM chunking strategies. For example, text elements could be queried by their preceding title element. The hierarchy is implemented by a parent_id tag in the element's metadata. ### Features - Introduces a parent_id to ElementMetadata (The id of the parent element, not a pointer) - Creates a rule set for assigning hierarchies. Sensible default is assigned, with an optional override parameter - Sets element parent ids if there isn't an existing parent id or matches the ruleset ### How it works Hierarchies are assigned via a parent id field in element metadata. Elements are read sequentially and evaluated against a ruleset. For example take the following elements: 1. Title, "This is the Title" 2. Text, "this is the text" And the ruleset: `{"title": ["text"]}`. When evaluated, the parent_id of 2 will be the id of 1. The algorithm for determining this is more complex and resolves several edge cases, so please read the code for further details. ### Schema Changes ``` @dataclass class ElementMetadata: coordinates: Optional[CoordinatesMetadata] = None data_source: Optional[DataSourceMetadata] = None filename: Optional[str] = None file_directory: Optional[str] = None last_modified: Optional[str] = None filetype: Optional[str] = None attached_to_filename: Optional[str] = None + parent_id: Optional[Union[str, uuid.UUID, NoID, UUID]] = None + category_depth: Optional[int] = None ... ``` ### Testing ``` from unstructured.partition.auto import partition from typing import List elements = partition(filename="./unstructured/example-docs/fake-html.html", strategy="auto") for element in elements: print( f"Category: {getattr(element, 'category', '')}\n"\ f"Text: {getattr(element, 'text', '')}\n" f"ID: {element.id}\n" \ f"Parent ID: {element.metadata.parent_id}\n"\ f"Depth: {element.metadata.category_depth}\n" \ ) ``` ### Additional Notes Implementing this feature revealed a possibly undesired side-effect in how element metadata are processed. In `unstructured/partition/common.py` the `_add_element_metadata` is invoked as part of the `add_metadata_with_filetype` decorator for filetype partitioning. This method is intended to add additional information to the metadata generated with the element including filename and filetype, however the existing metadata is merged into a newly created metadata object rather than the other way around. Because of the way it's structured, new metadata fields can easily be forgotten and pose debugging challenges to developers. This likely warrants a new issue. I'm guessing that the implementation is done this way to avoid issues with deserializing elements, but could be wrong. --------- Co-authored-by: Benjamin Torres <benjats07@users.noreply.github.com>
2023-09-14 11:23:16 -04:00
element = common.normalize_layout_element(
layout_element,
coordinate_system=coordinate_system,
)
assert element == FigureCaption(
text="Some lovely text",
coordinates=[[1, 2], [3, 4], [5, 6], [7, 8]],
coordinate_system=coordinate_system,
)
def test_normalize_layout_element_dict_misc():
layout_element = {
"type": "Misc",
"coordinates": [[1, 2], [3, 4], [5, 6], [7, 8]],
"text": "Some lovely text",
}
coordinate_system = PixelSpace(width=10, height=20)
Feat: Create a naive hierarchy for elements (#1268) ## **Summary** By adding hierarchy to unstructured elements, users will have more information for implementing vector db/LLM chunking strategies. For example, text elements could be queried by their preceding title element. The hierarchy is implemented by a parent_id tag in the element's metadata. ### Features - Introduces a parent_id to ElementMetadata (The id of the parent element, not a pointer) - Creates a rule set for assigning hierarchies. Sensible default is assigned, with an optional override parameter - Sets element parent ids if there isn't an existing parent id or matches the ruleset ### How it works Hierarchies are assigned via a parent id field in element metadata. Elements are read sequentially and evaluated against a ruleset. For example take the following elements: 1. Title, "This is the Title" 2. Text, "this is the text" And the ruleset: `{"title": ["text"]}`. When evaluated, the parent_id of 2 will be the id of 1. The algorithm for determining this is more complex and resolves several edge cases, so please read the code for further details. ### Schema Changes ``` @dataclass class ElementMetadata: coordinates: Optional[CoordinatesMetadata] = None data_source: Optional[DataSourceMetadata] = None filename: Optional[str] = None file_directory: Optional[str] = None last_modified: Optional[str] = None filetype: Optional[str] = None attached_to_filename: Optional[str] = None + parent_id: Optional[Union[str, uuid.UUID, NoID, UUID]] = None + category_depth: Optional[int] = None ... ``` ### Testing ``` from unstructured.partition.auto import partition from typing import List elements = partition(filename="./unstructured/example-docs/fake-html.html", strategy="auto") for element in elements: print( f"Category: {getattr(element, 'category', '')}\n"\ f"Text: {getattr(element, 'text', '')}\n" f"ID: {element.id}\n" \ f"Parent ID: {element.metadata.parent_id}\n"\ f"Depth: {element.metadata.category_depth}\n" \ ) ``` ### Additional Notes Implementing this feature revealed a possibly undesired side-effect in how element metadata are processed. In `unstructured/partition/common.py` the `_add_element_metadata` is invoked as part of the `add_metadata_with_filetype` decorator for filetype partitioning. This method is intended to add additional information to the metadata generated with the element including filename and filetype, however the existing metadata is merged into a newly created metadata object rather than the other way around. Because of the way it's structured, new metadata fields can easily be forgotten and pose debugging challenges to developers. This likely warrants a new issue. I'm guessing that the implementation is done this way to avoid issues with deserializing elements, but could be wrong. --------- Co-authored-by: Benjamin Torres <benjats07@users.noreply.github.com>
2023-09-14 11:23:16 -04:00
element = common.normalize_layout_element(
layout_element,
coordinate_system=coordinate_system,
)
assert element == Text(
text="Some lovely text",
coordinates=[[1, 2], [3, 4], [5, 6], [7, 8]],
coordinate_system=coordinate_system,
)
def test_normalize_layout_element_layout_element():
layout_element = LayoutElement(
type="Text",
2023-04-04 19:59:06 -07:00
x1=1,
y1=2,
x2=3,
y2=4,
text="Some lovely text",
)
coordinate_system = PixelSpace(width=10, height=20)
Feat: Create a naive hierarchy for elements (#1268) ## **Summary** By adding hierarchy to unstructured elements, users will have more information for implementing vector db/LLM chunking strategies. For example, text elements could be queried by their preceding title element. The hierarchy is implemented by a parent_id tag in the element's metadata. ### Features - Introduces a parent_id to ElementMetadata (The id of the parent element, not a pointer) - Creates a rule set for assigning hierarchies. Sensible default is assigned, with an optional override parameter - Sets element parent ids if there isn't an existing parent id or matches the ruleset ### How it works Hierarchies are assigned via a parent id field in element metadata. Elements are read sequentially and evaluated against a ruleset. For example take the following elements: 1. Title, "This is the Title" 2. Text, "this is the text" And the ruleset: `{"title": ["text"]}`. When evaluated, the parent_id of 2 will be the id of 1. The algorithm for determining this is more complex and resolves several edge cases, so please read the code for further details. ### Schema Changes ``` @dataclass class ElementMetadata: coordinates: Optional[CoordinatesMetadata] = None data_source: Optional[DataSourceMetadata] = None filename: Optional[str] = None file_directory: Optional[str] = None last_modified: Optional[str] = None filetype: Optional[str] = None attached_to_filename: Optional[str] = None + parent_id: Optional[Union[str, uuid.UUID, NoID, UUID]] = None + category_depth: Optional[int] = None ... ``` ### Testing ``` from unstructured.partition.auto import partition from typing import List elements = partition(filename="./unstructured/example-docs/fake-html.html", strategy="auto") for element in elements: print( f"Category: {getattr(element, 'category', '')}\n"\ f"Text: {getattr(element, 'text', '')}\n" f"ID: {element.id}\n" \ f"Parent ID: {element.metadata.parent_id}\n"\ f"Depth: {element.metadata.category_depth}\n" \ ) ``` ### Additional Notes Implementing this feature revealed a possibly undesired side-effect in how element metadata are processed. In `unstructured/partition/common.py` the `_add_element_metadata` is invoked as part of the `add_metadata_with_filetype` decorator for filetype partitioning. This method is intended to add additional information to the metadata generated with the element including filename and filetype, however the existing metadata is merged into a newly created metadata object rather than the other way around. Because of the way it's structured, new metadata fields can easily be forgotten and pose debugging challenges to developers. This likely warrants a new issue. I'm guessing that the implementation is done this way to avoid issues with deserializing elements, but could be wrong. --------- Co-authored-by: Benjamin Torres <benjats07@users.noreply.github.com>
2023-09-14 11:23:16 -04:00
element = common.normalize_layout_element(
layout_element,
coordinate_system=coordinate_system,
)
assert element == NarrativeText(
text="Some lovely text",
2023-04-04 19:59:06 -07:00
coordinates=((1, 2), (1, 4), (3, 4), (3, 2)),
coordinate_system=coordinate_system,
)
def test_normalize_layout_element_layout_element_narrative_text():
layout_element = LayoutElement(
type="NarrativeText",
2023-04-04 19:59:06 -07:00
x1=1,
y1=2,
x2=3,
y2=4,
text="Some lovely text",
)
coordinate_system = PixelSpace(width=10, height=20)
Feat: Create a naive hierarchy for elements (#1268) ## **Summary** By adding hierarchy to unstructured elements, users will have more information for implementing vector db/LLM chunking strategies. For example, text elements could be queried by their preceding title element. The hierarchy is implemented by a parent_id tag in the element's metadata. ### Features - Introduces a parent_id to ElementMetadata (The id of the parent element, not a pointer) - Creates a rule set for assigning hierarchies. Sensible default is assigned, with an optional override parameter - Sets element parent ids if there isn't an existing parent id or matches the ruleset ### How it works Hierarchies are assigned via a parent id field in element metadata. Elements are read sequentially and evaluated against a ruleset. For example take the following elements: 1. Title, "This is the Title" 2. Text, "this is the text" And the ruleset: `{"title": ["text"]}`. When evaluated, the parent_id of 2 will be the id of 1. The algorithm for determining this is more complex and resolves several edge cases, so please read the code for further details. ### Schema Changes ``` @dataclass class ElementMetadata: coordinates: Optional[CoordinatesMetadata] = None data_source: Optional[DataSourceMetadata] = None filename: Optional[str] = None file_directory: Optional[str] = None last_modified: Optional[str] = None filetype: Optional[str] = None attached_to_filename: Optional[str] = None + parent_id: Optional[Union[str, uuid.UUID, NoID, UUID]] = None + category_depth: Optional[int] = None ... ``` ### Testing ``` from unstructured.partition.auto import partition from typing import List elements = partition(filename="./unstructured/example-docs/fake-html.html", strategy="auto") for element in elements: print( f"Category: {getattr(element, 'category', '')}\n"\ f"Text: {getattr(element, 'text', '')}\n" f"ID: {element.id}\n" \ f"Parent ID: {element.metadata.parent_id}\n"\ f"Depth: {element.metadata.category_depth}\n" \ ) ``` ### Additional Notes Implementing this feature revealed a possibly undesired side-effect in how element metadata are processed. In `unstructured/partition/common.py` the `_add_element_metadata` is invoked as part of the `add_metadata_with_filetype` decorator for filetype partitioning. This method is intended to add additional information to the metadata generated with the element including filename and filetype, however the existing metadata is merged into a newly created metadata object rather than the other way around. Because of the way it's structured, new metadata fields can easily be forgotten and pose debugging challenges to developers. This likely warrants a new issue. I'm guessing that the implementation is done this way to avoid issues with deserializing elements, but could be wrong. --------- Co-authored-by: Benjamin Torres <benjats07@users.noreply.github.com>
2023-09-14 11:23:16 -04:00
element = common.normalize_layout_element(
layout_element,
coordinate_system=coordinate_system,
)
assert element == NarrativeText(
text="Some lovely text",
2023-04-04 19:59:06 -07:00
coordinates=((1, 2), (1, 4), (3, 4), (3, 2)),
coordinate_system=coordinate_system,
)
def test_normalize_layout_element_checked_box():
layout_element = LayoutElement(
type="Checked",
2023-04-04 19:59:06 -07:00
x1=1,
y1=2,
x2=3,
y2=4,
text="",
)
coordinate_system = PixelSpace(width=10, height=20)
Feat: Create a naive hierarchy for elements (#1268) ## **Summary** By adding hierarchy to unstructured elements, users will have more information for implementing vector db/LLM chunking strategies. For example, text elements could be queried by their preceding title element. The hierarchy is implemented by a parent_id tag in the element's metadata. ### Features - Introduces a parent_id to ElementMetadata (The id of the parent element, not a pointer) - Creates a rule set for assigning hierarchies. Sensible default is assigned, with an optional override parameter - Sets element parent ids if there isn't an existing parent id or matches the ruleset ### How it works Hierarchies are assigned via a parent id field in element metadata. Elements are read sequentially and evaluated against a ruleset. For example take the following elements: 1. Title, "This is the Title" 2. Text, "this is the text" And the ruleset: `{"title": ["text"]}`. When evaluated, the parent_id of 2 will be the id of 1. The algorithm for determining this is more complex and resolves several edge cases, so please read the code for further details. ### Schema Changes ``` @dataclass class ElementMetadata: coordinates: Optional[CoordinatesMetadata] = None data_source: Optional[DataSourceMetadata] = None filename: Optional[str] = None file_directory: Optional[str] = None last_modified: Optional[str] = None filetype: Optional[str] = None attached_to_filename: Optional[str] = None + parent_id: Optional[Union[str, uuid.UUID, NoID, UUID]] = None + category_depth: Optional[int] = None ... ``` ### Testing ``` from unstructured.partition.auto import partition from typing import List elements = partition(filename="./unstructured/example-docs/fake-html.html", strategy="auto") for element in elements: print( f"Category: {getattr(element, 'category', '')}\n"\ f"Text: {getattr(element, 'text', '')}\n" f"ID: {element.id}\n" \ f"Parent ID: {element.metadata.parent_id}\n"\ f"Depth: {element.metadata.category_depth}\n" \ ) ``` ### Additional Notes Implementing this feature revealed a possibly undesired side-effect in how element metadata are processed. In `unstructured/partition/common.py` the `_add_element_metadata` is invoked as part of the `add_metadata_with_filetype` decorator for filetype partitioning. This method is intended to add additional information to the metadata generated with the element including filename and filetype, however the existing metadata is merged into a newly created metadata object rather than the other way around. Because of the way it's structured, new metadata fields can easily be forgotten and pose debugging challenges to developers. This likely warrants a new issue. I'm guessing that the implementation is done this way to avoid issues with deserializing elements, but could be wrong. --------- Co-authored-by: Benjamin Torres <benjats07@users.noreply.github.com>
2023-09-14 11:23:16 -04:00
element = common.normalize_layout_element(
layout_element,
coordinate_system=coordinate_system,
)
assert element == CheckBox(
checked=True,
coordinates=((1, 2), (1, 4), (3, 4), (3, 2)),
coordinate_system=coordinate_system,
)
def test_normalize_layout_element_unchecked_box():
layout_element = LayoutElement(
type="Unchecked",
2023-04-04 19:59:06 -07:00
x1=1,
y1=2,
x2=3,
y2=4,
text="",
)
coordinate_system = PixelSpace(width=10, height=20)
Feat: Create a naive hierarchy for elements (#1268) ## **Summary** By adding hierarchy to unstructured elements, users will have more information for implementing vector db/LLM chunking strategies. For example, text elements could be queried by their preceding title element. The hierarchy is implemented by a parent_id tag in the element's metadata. ### Features - Introduces a parent_id to ElementMetadata (The id of the parent element, not a pointer) - Creates a rule set for assigning hierarchies. Sensible default is assigned, with an optional override parameter - Sets element parent ids if there isn't an existing parent id or matches the ruleset ### How it works Hierarchies are assigned via a parent id field in element metadata. Elements are read sequentially and evaluated against a ruleset. For example take the following elements: 1. Title, "This is the Title" 2. Text, "this is the text" And the ruleset: `{"title": ["text"]}`. When evaluated, the parent_id of 2 will be the id of 1. The algorithm for determining this is more complex and resolves several edge cases, so please read the code for further details. ### Schema Changes ``` @dataclass class ElementMetadata: coordinates: Optional[CoordinatesMetadata] = None data_source: Optional[DataSourceMetadata] = None filename: Optional[str] = None file_directory: Optional[str] = None last_modified: Optional[str] = None filetype: Optional[str] = None attached_to_filename: Optional[str] = None + parent_id: Optional[Union[str, uuid.UUID, NoID, UUID]] = None + category_depth: Optional[int] = None ... ``` ### Testing ``` from unstructured.partition.auto import partition from typing import List elements = partition(filename="./unstructured/example-docs/fake-html.html", strategy="auto") for element in elements: print( f"Category: {getattr(element, 'category', '')}\n"\ f"Text: {getattr(element, 'text', '')}\n" f"ID: {element.id}\n" \ f"Parent ID: {element.metadata.parent_id}\n"\ f"Depth: {element.metadata.category_depth}\n" \ ) ``` ### Additional Notes Implementing this feature revealed a possibly undesired side-effect in how element metadata are processed. In `unstructured/partition/common.py` the `_add_element_metadata` is invoked as part of the `add_metadata_with_filetype` decorator for filetype partitioning. This method is intended to add additional information to the metadata generated with the element including filename and filetype, however the existing metadata is merged into a newly created metadata object rather than the other way around. Because of the way it's structured, new metadata fields can easily be forgotten and pose debugging challenges to developers. This likely warrants a new issue. I'm guessing that the implementation is done this way to avoid issues with deserializing elements, but could be wrong. --------- Co-authored-by: Benjamin Torres <benjats07@users.noreply.github.com>
2023-09-14 11:23:16 -04:00
element = common.normalize_layout_element(
layout_element,
coordinate_system=coordinate_system,
)
assert element == CheckBox(
checked=False,
coordinates=((1, 2), (1, 4), (3, 4), (3, 2)),
coordinate_system=coordinate_system,
)
def test_normalize_layout_element_enumerated_list():
layout_element = LayoutElement(
type="List",
2023-04-04 19:59:06 -07:00
x1=1,
y1=2,
x2=3,
y2=4,
text="1. I'm so cool! 2. You're cool too. 3. We're all cool!",
)
coordinate_system = PixelSpace(width=10, height=20)
Feat: Create a naive hierarchy for elements (#1268) ## **Summary** By adding hierarchy to unstructured elements, users will have more information for implementing vector db/LLM chunking strategies. For example, text elements could be queried by their preceding title element. The hierarchy is implemented by a parent_id tag in the element's metadata. ### Features - Introduces a parent_id to ElementMetadata (The id of the parent element, not a pointer) - Creates a rule set for assigning hierarchies. Sensible default is assigned, with an optional override parameter - Sets element parent ids if there isn't an existing parent id or matches the ruleset ### How it works Hierarchies are assigned via a parent id field in element metadata. Elements are read sequentially and evaluated against a ruleset. For example take the following elements: 1. Title, "This is the Title" 2. Text, "this is the text" And the ruleset: `{"title": ["text"]}`. When evaluated, the parent_id of 2 will be the id of 1. The algorithm for determining this is more complex and resolves several edge cases, so please read the code for further details. ### Schema Changes ``` @dataclass class ElementMetadata: coordinates: Optional[CoordinatesMetadata] = None data_source: Optional[DataSourceMetadata] = None filename: Optional[str] = None file_directory: Optional[str] = None last_modified: Optional[str] = None filetype: Optional[str] = None attached_to_filename: Optional[str] = None + parent_id: Optional[Union[str, uuid.UUID, NoID, UUID]] = None + category_depth: Optional[int] = None ... ``` ### Testing ``` from unstructured.partition.auto import partition from typing import List elements = partition(filename="./unstructured/example-docs/fake-html.html", strategy="auto") for element in elements: print( f"Category: {getattr(element, 'category', '')}\n"\ f"Text: {getattr(element, 'text', '')}\n" f"ID: {element.id}\n" \ f"Parent ID: {element.metadata.parent_id}\n"\ f"Depth: {element.metadata.category_depth}\n" \ ) ``` ### Additional Notes Implementing this feature revealed a possibly undesired side-effect in how element metadata are processed. In `unstructured/partition/common.py` the `_add_element_metadata` is invoked as part of the `add_metadata_with_filetype` decorator for filetype partitioning. This method is intended to add additional information to the metadata generated with the element including filename and filetype, however the existing metadata is merged into a newly created metadata object rather than the other way around. Because of the way it's structured, new metadata fields can easily be forgotten and pose debugging challenges to developers. This likely warrants a new issue. I'm guessing that the implementation is done this way to avoid issues with deserializing elements, but could be wrong. --------- Co-authored-by: Benjamin Torres <benjats07@users.noreply.github.com>
2023-09-14 11:23:16 -04:00
elements = common.normalize_layout_element(
layout_element,
coordinate_system=coordinate_system,
)
assert elements == [
ListItem(
text="I'm so cool!",
coordinates=((1, 2), (1, 4), (3, 4), (3, 2)),
coordinate_system=coordinate_system,
),
ListItem(
text="You're cool too.",
coordinates=((1, 2), (1, 4), (3, 4), (3, 2)),
coordinate_system=coordinate_system,
),
ListItem(
text="We're all cool!",
coordinates=((1, 2), (1, 4), (3, 4), (3, 2)),
coordinate_system=coordinate_system,
),
]
def test_normalize_layout_element_bulleted_list():
layout_element = LayoutElement(
type="List",
2023-04-04 19:59:06 -07:00
x1=1,
y1=2,
x2=3,
y2=4,
text="* I'm so cool! * You're cool too. * We're all cool!",
)
coordinate_system = PixelSpace(width=10, height=20)
Feat: Create a naive hierarchy for elements (#1268) ## **Summary** By adding hierarchy to unstructured elements, users will have more information for implementing vector db/LLM chunking strategies. For example, text elements could be queried by their preceding title element. The hierarchy is implemented by a parent_id tag in the element's metadata. ### Features - Introduces a parent_id to ElementMetadata (The id of the parent element, not a pointer) - Creates a rule set for assigning hierarchies. Sensible default is assigned, with an optional override parameter - Sets element parent ids if there isn't an existing parent id or matches the ruleset ### How it works Hierarchies are assigned via a parent id field in element metadata. Elements are read sequentially and evaluated against a ruleset. For example take the following elements: 1. Title, "This is the Title" 2. Text, "this is the text" And the ruleset: `{"title": ["text"]}`. When evaluated, the parent_id of 2 will be the id of 1. The algorithm for determining this is more complex and resolves several edge cases, so please read the code for further details. ### Schema Changes ``` @dataclass class ElementMetadata: coordinates: Optional[CoordinatesMetadata] = None data_source: Optional[DataSourceMetadata] = None filename: Optional[str] = None file_directory: Optional[str] = None last_modified: Optional[str] = None filetype: Optional[str] = None attached_to_filename: Optional[str] = None + parent_id: Optional[Union[str, uuid.UUID, NoID, UUID]] = None + category_depth: Optional[int] = None ... ``` ### Testing ``` from unstructured.partition.auto import partition from typing import List elements = partition(filename="./unstructured/example-docs/fake-html.html", strategy="auto") for element in elements: print( f"Category: {getattr(element, 'category', '')}\n"\ f"Text: {getattr(element, 'text', '')}\n" f"ID: {element.id}\n" \ f"Parent ID: {element.metadata.parent_id}\n"\ f"Depth: {element.metadata.category_depth}\n" \ ) ``` ### Additional Notes Implementing this feature revealed a possibly undesired side-effect in how element metadata are processed. In `unstructured/partition/common.py` the `_add_element_metadata` is invoked as part of the `add_metadata_with_filetype` decorator for filetype partitioning. This method is intended to add additional information to the metadata generated with the element including filename and filetype, however the existing metadata is merged into a newly created metadata object rather than the other way around. Because of the way it's structured, new metadata fields can easily be forgotten and pose debugging challenges to developers. This likely warrants a new issue. I'm guessing that the implementation is done this way to avoid issues with deserializing elements, but could be wrong. --------- Co-authored-by: Benjamin Torres <benjats07@users.noreply.github.com>
2023-09-14 11:23:16 -04:00
elements = common.normalize_layout_element(
layout_element,
coordinate_system=coordinate_system,
)
assert elements == [
ListItem(
text="I'm so cool!",
coordinates=((1, 2), (1, 4), (3, 4), (3, 2)),
coordinate_system=coordinate_system,
),
ListItem(
text="You're cool too.",
coordinates=((1, 2), (1, 4), (3, 4), (3, 2)),
coordinate_system=coordinate_system,
),
ListItem(
text="We're all cool!",
coordinates=((1, 2), (1, 4), (3, 4), (3, 2)),
coordinate_system=coordinate_system,
),
]
class MockPopenWithError:
def __init__(self, *args, **kwargs):
pass
def communicate(self):
return b"", b"an error occurred"
def test_convert_office_doc_captures_errors(monkeypatch, caplog):
import subprocess
monkeypatch.setattr(subprocess, "Popen", MockPopenWithError)
common.convert_office_doc("no-real.docx", "fake-directory", target_format="docx")
assert "an error occurred" in caplog.text
class MockDocxEmptyTable:
def __init__(self):
self.rows = []
def test_convert_ms_office_table_to_text_works_with_empty_tables():
table = MockDocxEmptyTable()
assert common.convert_ms_office_table_to_text(table, as_html=True) == ""
assert common.convert_ms_office_table_to_text(table, as_html=False) == ""
@pytest.mark.parametrize(
("text", "expected"),
[
("<table><tbody><tr><td>👨\\U+1F3FB🔧</td></tr></tbody></table>", True),
("<table><tbody><tr><td>Hello!</td></tr></tbody></table>", False),
],
)
def test_contains_emoji(text, expected):
assert contains_emoji(text) is expected
def test_document_to_element_list_omits_coord_system_when_coord_points_absent():
layout_elem_absent_coordinates = MockDocumentLayout()
elements = document_to_element_list(layout_elem_absent_coordinates)
assert elements[0].metadata.coordinates is None
def test_get_page_image_metadata_and_coordinate_system():
doc = MockDocumentLayout()
metadata = _get_page_image_metadata(doc.pages[0])
assert isinstance(metadata, dict)
Feat: Create a naive hierarchy for elements (#1268) ## **Summary** By adding hierarchy to unstructured elements, users will have more information for implementing vector db/LLM chunking strategies. For example, text elements could be queried by their preceding title element. The hierarchy is implemented by a parent_id tag in the element's metadata. ### Features - Introduces a parent_id to ElementMetadata (The id of the parent element, not a pointer) - Creates a rule set for assigning hierarchies. Sensible default is assigned, with an optional override parameter - Sets element parent ids if there isn't an existing parent id or matches the ruleset ### How it works Hierarchies are assigned via a parent id field in element metadata. Elements are read sequentially and evaluated against a ruleset. For example take the following elements: 1. Title, "This is the Title" 2. Text, "this is the text" And the ruleset: `{"title": ["text"]}`. When evaluated, the parent_id of 2 will be the id of 1. The algorithm for determining this is more complex and resolves several edge cases, so please read the code for further details. ### Schema Changes ``` @dataclass class ElementMetadata: coordinates: Optional[CoordinatesMetadata] = None data_source: Optional[DataSourceMetadata] = None filename: Optional[str] = None file_directory: Optional[str] = None last_modified: Optional[str] = None filetype: Optional[str] = None attached_to_filename: Optional[str] = None + parent_id: Optional[Union[str, uuid.UUID, NoID, UUID]] = None + category_depth: Optional[int] = None ... ``` ### Testing ``` from unstructured.partition.auto import partition from typing import List elements = partition(filename="./unstructured/example-docs/fake-html.html", strategy="auto") for element in elements: print( f"Category: {getattr(element, 'category', '')}\n"\ f"Text: {getattr(element, 'text', '')}\n" f"ID: {element.id}\n" \ f"Parent ID: {element.metadata.parent_id}\n"\ f"Depth: {element.metadata.category_depth}\n" \ ) ``` ### Additional Notes Implementing this feature revealed a possibly undesired side-effect in how element metadata are processed. In `unstructured/partition/common.py` the `_add_element_metadata` is invoked as part of the `add_metadata_with_filetype` decorator for filetype partitioning. This method is intended to add additional information to the metadata generated with the element including filename and filetype, however the existing metadata is merged into a newly created metadata object rather than the other way around. Because of the way it's structured, new metadata fields can easily be forgotten and pose debugging challenges to developers. This likely warrants a new issue. I'm guessing that the implementation is done this way to avoid issues with deserializing elements, but could be wrong. --------- Co-authored-by: Benjamin Torres <benjats07@users.noreply.github.com>
2023-09-14 11:23:16 -04:00
def test_set_element_hierarchy():
elements_to_set = [
Title(text="Title"), # 0
NarrativeText(text="NarrativeText"), # 1
FigureCaption(text="FigureCaption"), # 2
ListItem(text="ListItem"), # 3
ListItem(text="ListItem", metadata=ElementMetadata(category_depth=1)), # 4
ListItem(text="ListItem", metadata=ElementMetadata(category_depth=1)), # 5
ListItem(text="ListItem"), # 6
CheckBox(element_id="some-id-1", checked=True), # 7
Title(text="Title 2"), # 8
ListItem(text="ListItem"), # 9
ListItem(text="ListItem"), # 10
Text(text="Text"), # 11
]
elements = common.set_element_hierarchy(elements_to_set)
assert (
elements[1].metadata.parent_id == elements[0].id
), "NarrativeText should be child of Title"
assert (
elements[2].metadata.parent_id == elements[0].id
), "FigureCaption should be child of Title"
assert elements[3].metadata.parent_id == elements[0].id, "ListItem should be child of Title"
assert elements[4].metadata.parent_id == elements[3].id, "ListItem should be child of Title"
assert elements[5].metadata.parent_id == elements[3].id, "ListItem should be child of Title"
assert elements[6].metadata.parent_id == elements[0].id, "ListItem should be child of Title"
assert (
elements[7].metadata.parent_id is None
), "CheckBox should be None, as it's not a Text based element"
assert elements[8].metadata.parent_id is None, "Title 2 should be child of None"
assert elements[9].metadata.parent_id == elements[8].id, "ListItem should be child of Title 2"
assert elements[10].metadata.parent_id == elements[8].id, "ListItem should be child of Title 2"
assert elements[11].metadata.parent_id == elements[8].id, "Text should be child of Title 2"
def test_set_element_hierarchy_custom_rule_set():
elements_to_set = [
Header(text="Header"), # 0
Title(text="Title"), # 1
NarrativeText(text="NarrativeText"), # 2
Text(text="Text"), # 3
Title(text="Title 2"), # 4
FigureCaption(text="FigureCaption"), # 5
]
custom_rule_set = {
"Header": ["Title", "Text"],
"Title": ["NarrativeText", "UncategorizedText", "FigureCaption"],
}
elements = common.set_element_hierarchy(
elements=elements_to_set,
ruleset=custom_rule_set,
)
assert elements[1].metadata.parent_id == elements[0].id, "Title should be child of Header"
assert (
elements[2].metadata.parent_id == elements[1].id
), "NarrativeText should be child of Title"
assert elements[3].metadata.parent_id == elements[1].id, "Text should be child of Title"
assert elements[4].metadata.parent_id == elements[0].id, "Title 2 should be child of Header"
assert (
elements[5].metadata.parent_id == elements[4].id
), "FigureCaption should be child of Title 2"