contamination_standalone

Stand-alone contamination models with Docker images and scripts.

Images (Docker Hub, amd64)

• BMP runtime: docker pull nspies13/contamination-bmp:latest
• CBC runtime: docker pull nspies13/contamination-cbc:latest
• BMP training (no bundled models): docker pull nspies13/contamination-bmp-train:latest
• CBC training (no bundled models): docker pull nspies13/contamination-cbc-train:latest
• Web UI (Shiny): build locally with docker build -f Dockerfile.app -t contamination-app:latest .
• Apple Silicon: add --platform linux/amd64 to docker pull / docker run.

Batch CLI mode for Predictions on CSVs

• BMP: docker run --rm -v "$PWD/data:/data" nspies13/contamination-bmp:latest --mode batch --input-file /data/bmp_test_wide.csv --output-file /data/bmp_predictions.csv
• CBC: docker run --rm -v "$PWD/data:/data" nspies13/contamination-cbc:latest --mode batch --input-file /data/cbc_test_wide.csv --output-file /data/cbc_predictions.csv

Minimal JSON APIs

Load the bundled models and expose a REST API:

• BMP: docker run --rm -p 8000:8000 -e HOST=0.0.0.0 -e PORT=8000 nspies13/contamination-bmp:latest
• CBC: docker run --rm -p 8002:8000 -e HOST=0.0.0.0 -e PORT=8000 nspies13/contamination-cbc:latest

Example payloads

• BMP (wide JSON): see data/bmp_test_wide.csv for column names; send JSON array of objects with those columns.

  curl -s -X POST http://localhost:8000/predict \
    -H "Content-Type: application/json" \
    -d '[{"sodium":133,"chloride":94,"potassium_plas":3.7,"co2_totl":26,"bun":40,"creatinine":2.42,"calcium":8.9,"glucose":181,"sodium_prior":132,"chloride_prior":83,"potassium_plas_prior":4.5,"co2_totl_prior":24,"bun_prior":49,"creatinine_prior":3.62,"calcium_prior":10.3,"glucose_prior":135,"sodium_post":135,"chloride_post":94,"potassium_plas_post":3.4,"co2_totl_post":28,"bun_post":32,"creatinine_post":1.75,"calcium_post":9.0,"glucose_post":133}]'
  

• CBC (wide JSON): see data/cbc_test_wide.csv for column names.

  curl -s -X POST http://localhost:8002/predict \
    -H "Content-Type: application/json" \
    -d '[{"Hgb":12.0,"Plt":103,"WBC":0.35,"Hgb_prior":10.9,"WBC_prior":0.23,"Plt_prior":115,"Hgb_post":10.8,"WBC_post":0.6,"Plt_post":60}]'
  

Streaming endpoint

• Send newline-delimited JSON to /predict_stream to process multiple payloads in one request.

  printf '%s\n%s\n' '{"sodium":133,...}' '{"sodium":130,...}' | \
    curl -s -X POST http://localhost:8000/predict_stream -H "Content-Type: application/json" --data-binary @-
  

  Each line is parsed independently; the response is a JSON array of per-line results.

Responses are JSON with prediction probabilities/predicted classes (and mix ratios when I fix the versioning error).

Batch CLI mode (CSV -> CSV with predictions appended)

• BMP: docker run --rm -v "$PWD/data:/data" nspies13/contamination-bmp:latest --mode batch --input-file /data/bmp_test_wide.csv --output-file /data/bmp_predictions.csv
• CBC: docker run --rm -v "$PWD/data:/data" nspies13/contamination-cbc:latest --mode batch --input-file /data/cbc_test_wide.csv --output-file /data/cbc_predictions.csv
• Long-form batch input (BMP/CBC): add --input-format long and pass a long-form CSV with PATIENT_ID, DRAWN_DT_TM, TASK_ASSAY, RESULT_VALUE.
  • BMP example: docker run --rm -v "$PWD/data:/data" nspies13/contamination-bmp:latest --mode batch --input-format long --input-file /data/bmp_test_long.csv --output-file /data/bmp_predictions.csv
  • CBC example: docker run --rm -v "$PWD/data:/data" nspies13/contamination-cbc:latest --mode batch --input-format long --input-file /data/cbc_test_long.csv --output-file /data/cbc_predictions.csv

Training images

• BMP training: docker run --rm -v "$PWD/data:/data" -v "$PWD/tmp_outputs:/outputs" nspies13/contamination-bmp-train:latest /data/bmp_test_wide.csv /data/fluid_concentrations.tsv /outputs/bmp_models_combined.RDS
• CBC training: docker run --rm -v "$PWD/data:/data" -v "$PWD/tmp_outputs:/outputs" nspies13/contamination-cbc-train:latest /data/cbc_test_wide.csv /outputs/cbc_models_combined.RDS /outputs/cbc_mix_ratio_model.RDS

FluidFlagger app (Shiny)

• Build: docker build --platform linux/amd64 -f Dockerfile.app -t contamination-app:latest .
• Run: docker run --rm -p 8000:8000 contamination-app:latest
• Open: http://localhost:8000

Predict mode

• Choose BMP or CBC, upload a wide CSV/TSV, preview, and download predictions.
• BMP-only: filter which fluids to include.
• Optional: upload custom model RDS files (combined models, and mix-ratio models).

Train mode

• Upload a training template CSV/TSV and train models directly in the app.
• BMP-only: select fluids, add a new fluid with analyte concentrations.
• Download the trained model RDS after training completes (auto-download is enabled).
• Mix ratio models are skipped in this mode.

Review mode

• Upload a review CSV/TSV, label each row as Real/Contaminated/Equivocal.
• Labels are saved to results/<input_basename>_labels.csv with the original columns plus label, labeler, and timestamp.
• Download labels at any time from the UI.