Skip to content

mitre/linkage-agent-tools

Repository files navigation

Linkage Agent Tools

Tools for the Childhood Obesity Data Initative (CODI) Linkage Agent to use to accept garbled input from data owners / partners, perform matching and generate network IDs. This can also be thought of as Semi-Trusted Third Party (STTP) tools.

These tool facilitate a Privacy Preserving Record Linkage (PPRL) process. They build on the open source anonlink software package.

Installation

anonlink-entity-service

The primary dependency of these tools is on the anonlink-entity-service. This software package provides a web service for accessing anonlink's matching capabilites. This software must be installed for the Linkage Agent Tools to work. Install instructions can be found on the anonlink-entity-service Deployment page

After following the install instructions in the anonlink-entity-service documentation, you can confirm it is working if the API call to /api/v1/status responds as described in the example at https://anonlink-entity-service.readthedocs.io/en/stable/local-deployment.html to clarify when entity service is up and running correctly.

MongoDB

Linkage Agent Tools uses MongoDB to store results obtained from the anonlink-entity-service. Install MongoDB by downloading the community version.

Dependency Overview

Linkage Agent Tools is a set of scripts designed to interact with the previously mentioned anonlink-entity-service. They were created and tested on Python 3.7.4. The tools rely on two libraries: Requests and pymongo.

Requests is a library that makes HTTP requests. This is used for the tools to communicate with the web service offered by the anonlink-entity-service.

pymongo is a Python client library for MongoDB.

Linkage Agent Tools contains a test suite, which was created using pytest.

Installing with an existing Python install

Cloning the Repository

Clone the project locally as a Git repository

git clone https://github.com/mitre/linkage-agent-tools.git

Or download as a zip file:

  1. Click this link to download the project as a zip or use the "Clone or download" button on GitHub.
  2. Unzip the file.

Set up a virtual environment (Optional, but recommended)

It can be helpful to set up a virtual environment to isolate project dependencies from system dependencies. There are a few libraries that can do this, but this documentation will stick with venv since that is included in the Python Standard Library.

# Navigate to the project folder
cd linkage-agent-tools/

# Create a virtual environment in a `venv/` folder
python -m venv venv/

# Activate the virtual environment
source venv/bin/activate

Installing package and dependencies

pip install -r requirements.txt
pip install -e .

The second invocation of pip is required in order for setup.py to be able to communicate with your local installation of python so that the included modules can be found easily.

Installing with Anaconda

  1. Install Anaconda by following the install instructions.
    1. Depending on user account permissions, Anaconda may not install the latest version or may not be available to all users. If that is the case, try running conda update -n base -c defaults conda
  2. Download the tools as a zip file using the "Clone or download" button on GitHub.
  3. Unzip the file.
  4. Open an Anaconda Powershell Prompt
  5. Go to the unzipped directory
  6. Run the following commands:
    1. conda create --name codi
    2. conda activate codi
    3. conda install pip
    4. pip install -r requirements.txt
    5. pip install -e .

Configuration

Linkage Agent Tools is driven by a central configuration file, which is a JSON document saved as config.json. An example is shown below:

{
  "systems": ["site_a", "site_b", "site_c", "site_d", "site_e", "site_f"],
  "projects": ["name-sex-dob-phone", "name-sex-dob-zip",
    "name-sex-dob-parents", "name-sex-dob-addr"],
  "schema_folder": "/CODI/data-owner-tools/example-schema",
  "inbox_folder": "/CODI/inbox",
  "matching_results_folder": "/CODI/results",
  "project_results_folder": "/CODI/project_results",
  "output_folder": "/CODI/output",
  "entity_service_url": "http://localhost:8851/api/v1",
  "matching_threshold": 0.75,
  "mongo_uri": "localhost:27017",
  "blocked": false,
  "blocking_schema": "/CODI/data-owner-tools/example-schema/blocking-schema/lambda.json",
  "household_match": true,
  "household_schema": "/CODI/data-owner-tools/example-schema/household-schema/fn-phone-addr-zip.json"
}

A description of the properties in the file:

  • systems - The set of data owners in this matching effort. These are short names for the participants. When data owners send zip files, it is expected that they will have the format of "data owner name".zip.
  • projects - The anonlink linkage projects that are going to be used in this matching effort. It assumes that the project names will have a corresponding anonlink schema file in the schema folder.
  • schema_folder - A folder containing anonlink schema files. The schema files should be named "project name".json.
  • inbox_folder - The folder where zip files recieved from data owners should be placed.
  • matching_results_folder - Folder where the CSV containing the complete mapping of LINK_IDs to all data owners project_results_folder - Folder where results from projects run with anonlink-entity-service are stored.
  • output_folder - Folder where CSV files are generated, one per data owner. These files contain LINK_IDs mapped to a single data owner.
  • entity_service_url - The RESTful service endpoint for the anonlink-entity-service.
  • matching_threshold - The threshold for considering a potential set of records a match when comparing in anonlink. This can either be a single number between 0 and 1 or a list of numbers between 0 and 1
  • mongo_uri - The URI to use when connecting to MongoDB to store or access results. For details on the URI structure, consult the Connection String URI Format documentation
  • blocked - A boolean value indicating whether the CLKs from the data owner in the inbox folder were generated via blocking
  • blocking_schema - The optional path to the file used by data owner tools for blocking
  • household_match - A boolean true / false value for running the household pprl and matching options. The matching process can only be run in individual or household mode; if this value is true, matching will only be performed on household data provided by the data owners
  • household_schema - The path to the file used during household PPRL

Input and Output Setup

Once you specify the paths outlined in the configuration section above, you need to put the zip files from each data owner into the inbox_folder specified, with file from either individuals or households from systems aka data owners [site_a, site_b, site_c]. Below is an example for individuals, corresponding to a configuration setting of false for household_match:

inbox/
  site_a.zip
  site_b.zip
  site_c.zip
  ...

Note that these file names exactly match the systems list in the configuration, with .zip at the end. This is required.

And an example for households, with household_match set to true:

inbox/
  site_a_households.zip
  site_b_households.zip
  site_c_households.zip
  ...

Note that the household file names in this example also start with system names from the systems configuration value, and end with _households.zip; this is also required.

After running the scripts in the order specified in the repository structure section below, the project will produce the following files in the output_folder specified in the config. The first example would be the output for individuals:

output/
  site_a.zip
  site_b.zip
  site_c.zip
  ...

And the second example, for households:

output/
  site_a_households.zip
  site_b_households.zip
  site_c_households.zip
  ...

Structure

This project is a set of python scripts driven by a central configuration file, config.json. It is expected to operate in the following order:

  1. Data owners transmit their garbled zip files to the Linkage Agent. These zip files should be placed into the configured inbox folder.
  2. Update config.json to enable or disable household_match, depending on the type of files received from data owners.
  3. Run validate.py which will ensure all of the necessary files are present.
  4. Run drop.py if you have done a previous matching run to clear old data in the database; this will drop all data for individuals and households, whether household_match is true or false
  5. When all data is present, run projects.py to run the projects with the anonlink-entity-service in preparation for matching. Results will be stored in the project_results_folder.
  6. Run match.py, which will perform pairwise matching of the garbled information sent by the data owners for either individuals or households, depending on the value of household_match. The matching information will be stored in MongoDB.
  7. After matching, run link_ids.py, which will take all of the resulting matching information and use it to generate LINK_IDs, which are written to a CSV file in the configured results folder.
  8. Once all LINK_IDs have been created, run data_owner_ids.py which will create one ZIP file per data owner. That file will contain a metadata file and a CSV file with only information on their LINK_IDs.

projects.py, match.py and link_ids.py will also generate JSON metadata files that contain information about the corresponding process.

Example system folder hierarchy:

The schema_folder in the example below is using the example config paths from above, with household_match set to true. The schemas used by the data-owner during garbling of the data needs to be the same schemas pointed to in the linkage-agent config.json.

/CODI/
  linkage-agent-tools/
    ...
  inbox/
    site_a_households.zip
    site_a_block.zip
    site_b_households.zip
    site_b_block.zip
  output/
    site_a_households.csv
    site_b_households.csv
  data-owner-tools/
    ...
    example-schema/
      name-dob-ex.json
      name-phone-ex.json
      blocking-schema/
        lambda.json
      household-schema/
        fn-phone-addr-zip.json

Running Tests

Linkage Agent Tools contains a unit test suite. Tests can be run with the following command:

python -m pytest

Formatting and Linting

This repository uses black, flake8, and isort to maintain consistent formatting and style. These tools can be run with the following command:

black .
isort .
flake8 .

[WIP] Jupyter Notebook

The Linkage and Blocking Tuning Tool Jupyter notebook is a work in progress meant for testing and tuning different configurations against the synthetic data set with Data Owner Tools and Linkage Agent Tools projects running on the same machine. It will currently run all necessary scripts to do end to end testing of the entire PPRL process but is still being improved and will include more documentation when finalized.

Notice

Copyright 2020-2022 The MITRE Corporation.

Approved for Public Release; Distribution Unlimited. Case Number 19-2008