Skip to content

README.md

Latest commit

 

History

History
109 lines (79 loc) · 2.79 KB

README.md

File metadata and controls

109 lines (79 loc) · 2.79 KB

Primitivo Model - Probabilistic Prescribing

Probabilistic prescribing model for IV-to-Oral antibiotic switching predictions. Uses Neural Processes (ConvGNP), GBDT/Logistic Regression, and baseline models trained on MIMIC-IV healthcare time series data.

Setup

Prerequisites

This project uses uv for Python package management.

curl -LsSf https://astral.sh/uv/install.sh | sh

Install

git clone https://github.com/SAFEHR-data/ivos-model.git
cd ivos-model
uv sync --all-extras
uv run pre-commit install

Usage

# Show CLI help
uv run primitivo-model --help

# Train models
uv run primitivo-model nps train --route simple-charts-dev --data-source mimic4
uv run primitivo-model tabular train-gbdt --route simple-charts-dev --data-source mimic4
uv run primitivo-model baseline repeat-last --route simple-charts-dev --data-source mimic4

# Process MIMIC data
uv run primitivo-model mimic process --smoke-test

# Show configuration
uv run primitivo-model config

Running Tests

uv run pytest
uv run pytest tests/test_criteria.py -v

Linting and Formatting

uv run ruff check .
uv run ruff format .
uv run pre-commit run --all-files

MLflow Integration

All experiments log to a local mlruns/ directory.

mlflow ui --port 5000

Data

MIMIC-IV

The models are trained on MIMIC-IV, which requires credentialed access via PhysioNet. To prepare the database:

  1. Download the CSV files (requires a PhysioNet account with signed data use agreement):

    wget -r -N -c -np --user YOUR_USERNAME --ask-password \
    https://physionet.org/files/mimiciv/3.1/

  2. Build the DuckDB database using mimic-code:

    git clone https://github.com/MIT-LCP/mimic-code
# Follow the instructions at:
# https://github.com/MIT-LCP/mimic-code/tree/main/mimic-iv/buildmimic/duckdb

    Note for v3.1: CSV parsing requires a small fix — on line 115 of the build script, change:

    COPY $TABLE_NAME FROM '$FILE' (HEADER);

    to:

    COPY $TABLE_NAME FROM '$FILE' (DELIMITER ',', HEADER, ESCAPE '"');

    See mimic-code#1881 for details.

  3. Validate the database:

    duckdb mimic4.db < mimic-code/mimic-iv/postgres/validate.sql

  4. Process into model format, pointing DATA_ROOT at the directory containing mimic4/mimic4.db:

    export DATA_ROOT=/path/to/your/data
uv run primitivo-model mimic process

By default DATA_ROOT is the data/ directory in the project root.

Radix

The Radix dataset used in this project is not publicly available.