feat(openadapt.privacy.providers): add private_ai as a scrubbing provider (#486)

* ran black

* add blank file for priavte ai

* add class `PrivateAIScrubbingProvider`

* add pvt_ai api key in config

* ran `pre-commit run --all`

* add scrub_text function

* progress for image redaction

* format

* complete scrub_image

* start scrub_pdf

* add pdf redaction code

* add more wrapper for invalid reponse from priavate ai

* try to fix pytest

* try to fix pytest

* remove lst

* remove uncessary methods

* text scrubbing test all passses

* pdf_redaction test works

* add test_image_redaction test in pyetst

* add easy ocr to poetry

* pytest is fixed

* remove versino files

* remove un-neccasry files

* add code to remove uncessary files after pytest

* addressed #486 (comment)

* address comment #486 (comment)

* reduce line chars

* addressed comment: #486 (comment)

* fix flake8

* use f strings

* address comment: #486 (comment)

* address comment: #486 (comment)

* change to value error

* remove .keys()

* add constants

* fix flake8 erros

* use BytesIO

* address comment #486 (comment)

* rna black

* final commit

* remove unused code

* refactor typo

* rename `redact_file_path` to `redacted_file_path`

* use BytesIO wherever possible

* fix flake8

* add documentation links

* Apply suggestions from code review

* Update tests/openadapt/privacy/providers/test_private_ai_scrub.py

* fix poetry.lock

* poetry.lock

---------

Co-authored-by: Krish Patel <[email protected]>
Co-authored-by: Richard Abrich <[email protected]>
Co-authored-by: Richard Abrich <[email protected]>

Author: 3 people

assets/sample_llc_1.pdf

@@ -9,3 +9,4 @@ class ScrubProvider:  # pylint: disable=too-few-public-methods
 
     PRESIDIO = "PRESIDIO"
     COMPREHEND = "COMPREHEND"
+    PRIVATE_AI = "PRIVATE_AI"

openadapt/privacy/providers/__init__.py

@@ -0,0 +1,198 @@
+"""A Module for Private AI Scrubbing Provider."""
+
+from io import BytesIO
+from typing import List
+import base64
+
+from loguru import logger
+from PIL import Image
+import requests
+
+from openadapt import config
+from openadapt.privacy.base import Modality, ScrubbingProvider, TextScrubbingMixin
+from openadapt.privacy.providers import ScrubProvider
+
+BASE64_URL = "https://api.private-ai.com/deid/v3/process/files/base64"
+FILES_DIR = "assets/"
+HEADER_CONTENT_TYPE = "application/json"
+IMAGE_CONTENT_TYPE = "image/png"
+PDF_CONTENT_TYPE = "application/pdf"
+TEMP_IMAGEFILE_NAME = "temp_image_to_scrub.png"
+TEXT_URL = "https://api.private-ai.com/deid/v3/process/text"
+
+
+class PrivateAIScrubbingProvider(
+    ScrubProvider, ScrubbingProvider, TextScrubbingMixin
+):  # pylint: disable=abstract-method
+    """A Class for Private AI Scrubbing Provider."""
+
+    name: str = ScrubProvider.PRIVATE_AI
+    capabilities: List[Modality] = [Modality.TEXT, Modality.PIL_IMAGE, Modality.PDF]
+
+    def scrub_text(self, text: str, is_separated: bool = False) -> str:
+        """Scrub the text of all PII/PHI.
+
+        Args:
+            text (str): Text to be redacted
+            is_separated (bool): Whether the text is separated with special characters
+
+        Returns:
+            str: redacted text
+        """
+        payload = {
+            "text": [text],
+            "link_batch": False,
+            "entity_detection": {
+                "accuracy": "high",
+                "return_entity": True,
+            },
+            "processed_text": {
+                "type": "MARKER",
+                "pattern": "[UNIQUE_NUMBERED_ENTITY_TYPE]",
+            },
+        }
+
+        headers = {
+            "Content-Type": HEADER_CONTENT_TYPE,
+            "X-API-KEY": config.PRIVATE_AI_API_KEY,
+        }
+
+        response = requests.post(TEXT_URL, json=payload, headers=headers)
+        response.raise_for_status()
+        data = response.json()
+        logger.debug(f"{data=}")
+
+        # According to the PrivateAI API documentation,
+        # https://docs.private-ai.com/reference/latest/operation/process_text_v3_process_text_post/
+        # the response is a list of dicts when there is no error/issue in the request
+        # else it is a dict with a key "detail" containing the error message
+
+        if type(data) is dict and "detail" in data:
+            raise ValueError(data.get("detail"))
+
+        redacted_text = data[0].get("processed_text")
+        logger.debug(f"{redacted_text=}")
+
+        return redacted_text
+
+    def scrub_image(
+        self,
+        image: Image,
+        fill_color: int = config.SCRUB_FILL_COLOR,  # pylint: disable=no-member
+    ) -> Image:
+        """Scrub the image of all PII/PHI.
+
+        Args:
+            image (Image): A PIL.Image object to be redacted
+            fill_color (int): The color used to fill the redacted regions(BGR).
+
+        Returns:
+            Image: The redacted image with PII and PHI removed.
+        """
+        buffer = BytesIO()
+
+        image.save(buffer, format="PNG")
+        # Get the image data as bytes
+        image_data = buffer.getvalue()
+
+        file_data = base64.b64encode(image_data)
+        file_data = file_data.decode("ascii")
+
+        # Clean up by closing the BytesIO buffer
+        buffer.close()
+
+        payload = {
+            "file": {"data": file_data, "content_type": IMAGE_CONTENT_TYPE},
+            "entity_detection": {"accuracy": "high", "return_entity": True},
+            "pdf_options": {"density": 150, "max_resolution": 2000},
+            "audio_options": {"bleep_start_padding": 0, "bleep_end_padding": 0},
+        }
+
+        headers = {
+            "Content-Type": HEADER_CONTENT_TYPE,
+            "X-API-KEY": config.PRIVATE_AI_API_KEY,
+        }
+
+        response = requests.post(BASE64_URL, json=payload, headers=headers)
+        response = response.json()
+        logger.debug(f"{response=}")
+
+        # According to the PrivateAI API documentation,
+        # https://docs.private-ai.com/reference/latest/operation/process_files_base64_v3_process_files_base64_post/
+        # else it is a dict with a key "detail" containing the error message
+
+        if type(response) is dict and "detail" in response:
+            raise ValueError(response.get("detail"))
+
+        redacted_file_data = response.get("processed_file").encode("ascii")
+        redacted_file_data = base64.b64decode(redacted_file_data, validate=True)
+
+        # Use a BytesIO buffer to work with redacted_file_data
+        redacted_buffer = BytesIO(redacted_file_data)
+
+        redact_pil_image_data = Image.open(redacted_buffer)
+
+        return redact_pil_image_data
+
+    def scrub_pdf(self, path_to_pdf: str) -> str:
+        """Scrub the PDF of all PII/PHI.
+
+        Args:
+            path_to_pdf (str): Path to the PDF to be redacted
+
+        Returns:
+            str: Path to the redacted PDF
+        """
+        # Create a BytesIO buffer to read the PDF file
+        with open(path_to_pdf, "rb") as pdf_file:
+            pdf_buffer = BytesIO(pdf_file.read())
+
+        # Read PDF data from the BytesIO buffer
+        pdf_data = pdf_buffer.getvalue()
+        pdf_buffer.close()
+
+        # Encode PDF data as base64
+        pdf_base64 = base64.b64encode(pdf_data).decode("ascii")
+
+        payload = {
+            "file": {"data": pdf_base64, "content_type": PDF_CONTENT_TYPE},
+            "entity_detection": {"accuracy": "high", "return_entity": True},
+            "pdf_options": {"density": 150, "max_resolution": 2000},
+            "audio_options": {"bleep_start_padding": 0, "bleep_end_padding": 0},
+        }
+
+        headers = {
+            "Content-Type": HEADER_CONTENT_TYPE,
+            "X-API-KEY": config.PRIVATE_AI_API_KEY,
+        }
+
+        response = requests.post(BASE64_URL, json=payload, headers=headers)
+        response_data = response.json()
+
+        # According to the PrivateAI API documentation,
+        # https://docs.private-ai.com/reference/latest/operation/process_files_base64_v3_process_files_base64_post/
+        # the response is a list of dicts when there is no error/issue in the request
+        # else it is a dict with a key "detail" containing the error message
+
+        if isinstance(response_data, dict) and "details" in response_data:
+            raise ValueError(response_data.get("detail"))
+
+        logger.debug(f"{response_data.get('entities')=}")
+        logger.debug(f"{len(response_data.get('entities'))=}")
+
+        redacted_file_path = path_to_pdf.split(".")[0] + "_redacted.pdf"
+
+        # Create a BytesIO buffer to handle the redacted PDF data
+        redacted_buffer = BytesIO()
+
+        # Decode and write the redacted PDF data to the BytesIO buffer
+        processed_file = response_data.get("processed_file").encode("ascii")
+        processed_file = base64.b64decode(processed_file, validate=True)
+        redacted_buffer.write(processed_file)
+
+        # Write the redacted PDF data to a file
+        with open(redacted_file_path, "wb") as redacted_file:
+            redacted_buffer.seek(0)  # Move the buffer position to the beginning
+            redacted_file.write(redacted_buffer.read())
+
+        return redacted_file_path