Precision feedback pipeline

This is the pipeline service that implements the Precision Feedback Pipeline. The underlying model of precision feedback is captured in this conceptual model.

Read through our wiki pages for more detail on testing. Please note that this wiki might not be completely up to date.

The roadmap for the next version of the PFP can be found in the wiki.

Quick start

This is a Python software project and running the pipeline requires some familiarity with Python and virtual environments. This quick start gives directions using python's built in virtual environment tool venv and Poetry.

Clone the precision feedback pipeline

git clone https://github.com/Display-Lab/precision-feedback-pipeline.git

cd precision-feedback-pipeline

Setup a virtual environment and install dependencies

Using venv and pip

python --version # make sure you have python 3.11
python -m venv .venv

For Windows do this next to activate the virtual environment

.venv\Scripts\activate.bat

For Mac or Linux do this next to activate the virtual environment

source .venv/bin/activate

For Windows, Mac, and Linux, now complete the following two installs

pip install -r requirements.txt # this will take a while, so go get a cup of coffee
pip install uvicorn # not installed by default (needed for running locally)

Alternative: Using Poetry (for developers)

poetry env use 3.11 # optional, but makes sure you have python 3.11 available
poetry install # creates a virtual environment and install dependencies
poetry shell # activates the enviroment

Clone the knowledge base

Clone the knowledge base repository in a separate location

cd ..
git clone https://github.com/Display-Lab/knowledge-base.git 

Running the pipeline

Change back to the root of precision-feedback-pipeline

cd precision-feedback-pipeline

Update the .env.local file and change path/to/knowledge-base to point to the local knowledge base that you just checked out. (Don't remove the file:// for preferences and manifest.)

# .env.local
preferences=file:///Users/bob/knowledge-base/preferences.json 
mpm=/Users/bob/knowledge-base/prioritization_algorithms/motivational_potential_model.csv
manifest=file:///Users/bob/knowledge-base/mpog_local_manifest.yaml
...

Run the pipeline

ENV_PATH=.env.local uvicorn main:app
# Expect to see a server start message like this "INFO:     Uvicorn running on http://127.0.0.1:8000 (Press CTRL+C to quit)"

You can use Postman or your favorite tool to send a message and check the results. There is a sample input message file located at tests/test_cases/input_message.json. Here is a sample curl request:

curl --data "@tests/test_cases/input_message.json" http://localhost:8000/createprecisionfeedback/

Environment variables

Knowledge base settings

Local file path or URL (see .env.remote for github URL formats). All are required.

mpm: Path to the mpm csv file

preferences: Path to the preferences json file

manifest: Path to the manifest file that includes differend pieces of the base graph including (causal pathways, message templates, measures and comparators). See manifest configuration for more detail

Flags

display_window: Maximum number of month to be used to create visual displays (plots)

• default: 6

generate_image: If set to true and the display type is bar chart or line chart, then the pipeline will generate the images and include them as part of the response

• default: True

log_level: Sets the log level

• default: WARNING (this is the production defauslt)
• note: The PFP must be run with log_level=INFO in order to generate the candidate records in the output.

performance_month: If set, the pipeline will override the performance month in the input requests

• default: None

Scoring

These control the elements of the scoring algorithm.

use_coachiness: Switch to turn on and off coachiness

• default: True

use_history: Switch to turn on and off history

• default: True

use_mi: Switch to turn on and off motivating information

• default: True

use_preferences: Switch to turn on and off preferences

• default: True

manifest configuration

The manifest file includes all different pieces that should be loaded to the base graph including causal pathways, message templates, measures and comparators. It is a yaml file which specifies a directory structure containing JSON files for all those different categories.

Each entry consists of a key which is a URL (file:// or https:// or relative, see Uniform Resource Identifier (URI)) and a value which is a file path relative to the url. See manifest examples in the knowledge base.

If the key is a relative path, it must end with a '/'. In that case the key is going to be resolved towards the location of the manifest file by the pipeline.

examples

 ENV_PATH=/user/.../dev.env log_level=INFO use_preferences=True use_coachiness=True use_mi=True generate_image=False uvicorn main:app --workers=5

👉 uvicorn can be run with multiple workers. This is useful when testing with a client that can send multiple requests.