250 lines
8.0 KiB
Python
Raw Normal View History

2023-06-06 09:03:13 -07:00
import os
import pathlib
from dataclasses import dataclass
from typing import Any, Dict
2023-06-06 09:03:13 -07:00
import pytest
from unstructured.documents.elements import DataSourceMetadata
2023-06-06 09:03:13 -07:00
from unstructured.ingest.interfaces import (
BaseConnectorConfig,
BaseIngestDoc,
PartitionConfig,
ReadConfig,
2023-06-06 09:03:13 -07:00
)
from unstructured.partition.auto import partition
from unstructured.staging.base import convert_to_dict
DIRECTORY = pathlib.Path(__file__).parent.resolve()
EXAMPLE_DOCS_DIRECTORY = os.path.join(DIRECTORY, "../..", "example-docs")
TEST_DOWNLOAD_DIR = "/tmp"
TEST_OUTPUT_DIR = "/tmp"
TEST_ID = "test"
TEST_FILE_PATH = os.path.join(EXAMPLE_DOCS_DIRECTORY, "book-war-and-peace-1p.txt")
2023-06-06 09:03:13 -07:00
@dataclass
class TestConfig(BaseConnectorConfig):
id: str
path: str
TEST_CONFIG = TestConfig(id=TEST_ID, path=TEST_FILE_PATH)
TEST_SOURCE_URL = "test-source-url"
TEST_VERSION = "1.1.1"
TEST_RECORD_LOCATOR = {"id": "data-source-id"}
TEST_DATE_CREATED = "2021-01-01T00:00:00"
TEST_DATE_MODIFIED = "2021-01-02T00:00:00"
TEST_DATE_PROCESSSED = "2022-12-13T15:44:08"
2023-06-06 09:03:13 -07:00
2023-06-06 09:03:13 -07:00
@dataclass
class TestIngestDoc(BaseIngestDoc):
connector_config: TestConfig
2023-06-06 09:03:13 -07:00
@property
def filename(self):
return TEST_FILE_PATH
2023-06-26 13:12:05 -07:00
@property
def _output_filename(self):
return TEST_FILE_PATH + ".json"
@property
def source_url(self) -> str:
return TEST_SOURCE_URL
@property
def version(self) -> str:
return TEST_VERSION
@property
def record_locator(self) -> Dict[str, Any]:
return TEST_RECORD_LOCATOR
@property
def date_created(self) -> str:
return TEST_DATE_CREATED
@property
def date_modified(self) -> str:
return TEST_DATE_MODIFIED
@property
def exists(self) -> bool:
return True
2023-06-06 09:03:13 -07:00
def cleanup_file(self):
pass
def get_file(self):
pass
def has_output(self):
return True
2023-06-06 09:03:13 -07:00
def write_result(self, result):
pass
2023-06-06 09:03:13 -07:00
@pytest.fixture()
def partition_test_results():
# Reusable partition test results, calculated only once
result = partition(
filename=str(TEST_FILE_PATH),
data_source_metadata=DataSourceMetadata(
url=TEST_SOURCE_URL,
version=TEST_VERSION,
record_locator=TEST_RECORD_LOCATOR,
date_created=TEST_DATE_CREATED,
date_modified=TEST_DATE_MODIFIED,
date_processed=TEST_DATE_PROCESSSED,
),
)
return result
2023-06-06 09:03:13 -07:00
2023-06-06 09:03:13 -07:00
@pytest.fixture()
def partition_file_test_results(partition_test_results):
# Reusable partition_file test results, calculated only once
return convert_to_dict(partition_test_results)
def test_partition_file():
"""Validate partition_file returns a list of dictionaries with the expected keys,
metadatakeys, and data source metadata values."""
test_ingest_doc = TestIngestDoc(
connector_config=TEST_CONFIG,
read_config=ReadConfig(download_dir=TEST_DOWNLOAD_DIR),
partition_config=PartitionConfig(output_dir=TEST_OUTPUT_DIR),
)
test_ingest_doc._date_processed = TEST_DATE_PROCESSSED
isd_elems = test_ingest_doc.partition_file()
assert len(isd_elems)
expected_keys = {
"element_id",
"text",
"type",
"metadata",
}
# The document in TEST_FILE_PATH does not have elements with coordinates so
# partition is not expected to return coordinates metadata.
expected_metadata_keys = {
"data_source",
"filename",
"file_directory",
"filetype",
"last_modified",
}
for elem in isd_elems:
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
# Parent IDs are non-deterministic - remove them from the test
elem["metadata"].pop("parent_id", None)
assert expected_keys == set(elem.keys())
assert expected_metadata_keys == set(elem["metadata"].keys())
data_source_metadata = elem["metadata"]["data_source"]
assert data_source_metadata["url"] == TEST_SOURCE_URL
assert data_source_metadata["version"] == TEST_VERSION
assert data_source_metadata["record_locator"] == TEST_RECORD_LOCATOR
assert data_source_metadata["date_created"] == TEST_DATE_CREATED
assert data_source_metadata["date_modified"] == TEST_DATE_MODIFIED
assert data_source_metadata["date_processed"] == TEST_DATE_PROCESSSED
2023-06-06 09:03:13 -07:00
def test_process_file_fields_include_default(mocker, partition_test_results):
"""Validate when metadata_include and metadata_exclude are not set, all fields:
("element_id", "text", "type", "metadata") are included"""
mock_partition = mocker.patch(
2023-06-06 09:03:13 -07:00
"unstructured.ingest.interfaces.partition",
return_value=partition_test_results,
)
test_ingest_doc = TestIngestDoc(
connector_config=TEST_CONFIG,
read_config=ReadConfig(download_dir=TEST_DOWNLOAD_DIR),
partition_config=PartitionConfig(output_dir=TEST_OUTPUT_DIR),
2023-06-06 09:03:13 -07:00
)
isd_elems = test_ingest_doc.process_file()
assert len(isd_elems)
assert mock_partition.call_count == 1
2023-06-06 09:03:13 -07:00
for elem in isd_elems:
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
# Parent IDs are non-deterministic - remove them from the test
elem["metadata"].pop("parent_id", None)
2023-06-06 09:03:13 -07:00
assert {"element_id", "text", "type", "metadata"} == set(elem.keys())
data_source_metadata = elem["metadata"]["data_source"]
assert data_source_metadata["url"] == TEST_SOURCE_URL
assert data_source_metadata["version"] == TEST_VERSION
assert data_source_metadata["record_locator"] == TEST_RECORD_LOCATOR
assert data_source_metadata["date_created"] == TEST_DATE_CREATED
assert data_source_metadata["date_modified"] == TEST_DATE_MODIFIED
assert data_source_metadata["date_processed"] == TEST_DATE_PROCESSSED
2023-06-06 09:03:13 -07:00
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_process_file_metadata_includes_filename_and_filetype(
mocker,
partition_test_results,
):
"""Validate when metadata_include is set to "filename,filetype",
2023-06-06 09:03:13 -07:00
only filename is included in metadata"""
mocker.patch(
"unstructured.ingest.interfaces.partition",
return_value=partition_test_results,
)
test_ingest_doc = TestIngestDoc(
connector_config=TEST_CONFIG,
read_config=ReadConfig(download_dir=TEST_DOWNLOAD_DIR),
partition_config=PartitionConfig(
2023-06-06 09:03:13 -07:00
output_dir=TEST_OUTPUT_DIR,
metadata_include=["filename", "filetype"],
2023-06-06 09:03:13 -07:00
),
)
isd_elems = test_ingest_doc.process_file()
assert len(isd_elems)
for elem in isd_elems:
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
# Parent IDs are non-deterministic - remove them from the test
elem["metadata"].pop("parent_id", None)
assert set(elem["metadata"].keys()) == {"filename", "filetype"}
2023-06-06 09:03:13 -07:00
2023-06-06 09:03:13 -07:00
def test_process_file_metadata_exclude_filename_pagenum(mocker, partition_test_results):
"""Validate when metadata_exclude is set to "filename,page_number",
neither filename nor page_number are included in metadata"""
mocker.patch(
"unstructured.ingest.interfaces.partition",
return_value=partition_test_results,
)
test_ingest_doc = TestIngestDoc(
connector_config=TEST_CONFIG,
read_config=ReadConfig(download_dir=TEST_DOWNLOAD_DIR),
partition_config=PartitionConfig(
2023-06-06 09:03:13 -07:00
output_dir=TEST_OUTPUT_DIR,
metadata_exclude=["filename", "page_number"],
2023-06-06 09:03:13 -07:00
),
)
isd_elems = test_ingest_doc.process_file()
assert len(isd_elems)
for elem in isd_elems:
assert "filename" not in elem["metadata"]
assert "page_number" not in elem["metadata"]
2023-06-06 09:03:13 -07:00
2023-06-06 09:03:13 -07:00
def test_process_file_flatten_metadata(mocker, partition_test_results):
mocker.patch(
"unstructured.ingest.interfaces.partition",
return_value=partition_test_results,
)
test_ingest_doc = TestIngestDoc(
connector_config=TEST_CONFIG,
read_config=ReadConfig(download_dir=TEST_DOWNLOAD_DIR),
partition_config=PartitionConfig(
2023-06-06 09:03:13 -07:00
output_dir=TEST_OUTPUT_DIR,
metadata_include=["filename", "data_source"],
2023-06-06 09:03:13 -07:00
flatten_metadata=True,
),
)
isd_elems = test_ingest_doc.process_file()
expected_keys = {"element_id", "text", "type", "filename", "data_source"}
2023-06-06 09:03:13 -07:00
for elem in isd_elems:
assert expected_keys == set(elem.keys())