Model Deployer

Silent deployment of AIM2REDUCE models to predict undesirable cancer events in Princess Margaret hospital under the EPIC system.

Getting started

git clone https://github.com/ml4oncology/model-deployer
pip install -e ".[dev]"
playwright install chromium

# optional
pre-commit install
mypy --install-types

Running the dashboard pipeline

python src/main.py --start-date 20240904 --end-date 20250804 --model-anchor clinic
python src/monthly_model_eval.py --model-anchor clinic --monthly-pull-date 20250604 --start-date 20240904 --end-date 20250804 --disable-save-dashboard-png

Dashboard prerequisite:

• Dashboard generation uses the silent deployment baseline file. To create that baseline, first run python src/main.py --start-date [start_date] --end-date [end_date] --model-anchor clinic --run-on-silent-deployment True, where [start_date] and [end_date] are the silent deployment dates.
• If you are not generating that silent deployment baseline first, run src/main.py with --disable-save-dashboard-png to avoid dashboard generation errors.

Optional dashboard arguments:

• --dashboard-layout {portrait,landscape} controls the dashboard image layout. Default is portrait.
• --dashboard-font-scale FLOAT scales clinician-facing dashboard text and histogram text proportionally. Default is 1.0.
• --disable-save-dashboard-png skips dashboard PNG generation. By default, dashboard PNGs are generated.
• --subset-dashboard-patients limits dashboard generation to the selected patient subset. Default is True.
• --no-subset-dashboard-patients disables subset filtering and generates dashboards for all patients.
• --run-on-silent-deployment runs the code on silent deployment patients for baseline histogram in dashboard. Default is False.

Example:

python src/main.py --start-date 20240904 --end-date 20250804 --model-anchor clinic --dashboard-font-scale 1.5
python src/main.py --start-date 20240904 --end-date 20250804 --model-anchor clinic --no-subset-dashboard-patients

Project Organization

├── src             <- The source code
├── Data            <- The daily, weekly, and monthly data pulls
├── Models          <- ML models
└── Infos           <- Configuration files (regimen mapping, thresholds, etc)
    └── Prep

Filenames:

AIM2REDUCE_DATA_YYYYMMDD.csv

• patients and their historical records who had a treatment scheduled the following day of
• pulled daily

AIM2REDUCE_DATA_monthly_YYYYMMDD.csv

• patients and their historical records who had a treatment since March 2024 until the month before the date
• pulled monthly

AIM2REDUCE_DATA_weekly_YYYYMMDD.csv

• patients and their historical records who had a clinic visit on the day of and have treatment scheduled within the next 5 days
• pulled daily

AIM2REDUCE_DATA_weekly_monthly_YYYYMMDD.csv

• patients and their historical records who had a clinic visit since Sep 2024 until the month before the date
• pulled monthly