Roman/expose dpi param (#966)

* Bump inference version

* Pass through the dpi param if available

* Update CHANGELOG

* Check dpi param passed in via unit test

* Bump inference version

* Fix unit test around file info to work on mac as well

CHANGELOG.md

@@ -5,12 +5,13 @@
 ### Features
 
 * Adds Outlook connector
+* Add support for dpi parameter in inference library
 
 ### Fixes
 
 * Fixes issue with email partitioning where From field was being assigned the To field value.
 
-## 0.8.2-dev7
+## 0.8.2
 
 ### Enhancements
 

requirements/local-inference.in

@@ -1,3 +1,3 @@
 -c constraints.in
 -c base.txt
-unstructured-inference==0.5.5
+unstructured-inference==0.5.7

requirements/local-inference.txt

@@ -215,7 +215,7 @@ tzdata==2023.3
     # via
     #   -c requirements/base.txt
     #   pandas
-unstructured-inference==0.5.5
+unstructured-inference==0.5.7
     # via -r requirements/local-inference.in
 urllib3==1.26.16
     # via

test_unstructured/file_utils/test_exploration.py

@@ -39,7 +39,7 @@ def test_get_directory_file_info(tmpdir):
     assert isinstance(file_info, pd.DataFrame)
     assert set(file_info["filename"].to_list()) == {"filename1.txt", "filename2.txt"}
 
-    means = file_info.groupby("filetype").mean()
+    means = file_info.groupby("filetype").mean(numeric_only=True)
     assert means.columns.to_list() == ["filesize"]
 
 

test_unstructured/partition/test_pdf.py

@@ -384,6 +384,20 @@ def test_partition_pdf_with_copy_protection():
     assert {element.metadata.page_number for element in elements} == {1, 2}
 
 
+def test_partition_pdf_with_dpi():
+    filename = os.path.join("example-docs", "copy-protected.pdf")
+    with mock.patch.object(layout, "process_file_with_model", mock.MagicMock()) as mock_process:
+        pdf.partition_pdf(filename=filename, strategy="hi_res", pdf_image_dpi=100)
+        mock_process.assert_called_once_with(
+            filename,
+            is_image=False,
+            ocr_languages="eng",
+            extract_tables=False,
+            model_name=None,
+            pdf_image_dpi=100,
+        )
+
+
 def test_partition_pdf_requiring_recursive_text_grab(filename="example-docs/reliance.pdf"):
     elements = pdf.partition_pdf(filename=filename, strategy="fast")
     assert len(elements) > 50

unstructured/partition/pdf.py

@@ -203,12 +203,18 @@ def _partition_pdf_or_image_local(
 
     model_name = model_name if model_name else os.environ.get("UNSTRUCTURED_HI_RES_MODEL_NAME")
     if file is None:
+        pdf_image_dpi = kwargs.pop("pdf_image_dpi", None)
+        process_file_with_model_kwargs = {
+            "is_image": is_image,
+            "ocr_languages": ocr_languages,
+            "extract_tables": infer_table_structure,
+            "model_name": model_name,
+        }
+        if pdf_image_dpi:
+            process_file_with_model_kwargs["pdf_image_dpi"] = pdf_image_dpi
         layout = process_file_with_model(
             filename,
-            is_image=is_image,
-            ocr_languages=ocr_languages,
-            extract_tables=infer_table_structure,
-            model_name=model_name,
+            **process_file_with_model_kwargs,
         )
     else:
         layout = process_data_with_model(