iints-sdk-python35 0.1.21

A pre-clinical Edge-AI SDK for diabetes management validation.

IINTS-AF SDK

IINTS-AF is a safety-first simulation SDK for insulin algorithm research. It helps you test controllers, validate safety behavior, and generate audit-ready artifacts.

New here? Start with: docs/PLAIN_LANGUAGE_GUIDE.md

What You Get

• Virtual patient simulation for insulin strategies.
• Deterministic safety supervision (hard rules).
• Optional AI glucose predictor as advisory signal.
• Audit outputs: CSV, JSON, PDF reports.
• Validation and scorecard tooling for repeatable benchmarks.
• Transparent evidence manifest linked to peer-reviewed sources (iints sources).

Open Logic (Dual-Guard)

IINTS-AF is built as layered logic, not black-box control:

1. InputValidator
   Filters biologically implausible glucose input and applies fail-soft fallback.
2. LSTM Predictor (optional)
   Forecasts future glucose; advisory only.
3. IndependentSupervisor
   Deterministically validates every dose request and overrides unsafe actions.

Install

pip install iints-sdk-python35

Research extras:

pip install iints-sdk-python35[research]

Quick Start (CLI)

iints quickstart --project-name iints_quickstart
cd iints_quickstart
iints presets run --name baseline_t1d --algo algorithms/example_algorithm.py

One full run with artifacts:

iints run-full \
  --algo algorithms/example_algorithm.py \
  --scenario-path scenarios/clinic_safe_baseline.json \
  --output-dir results/run_full

Dual-Guard run with predictor:

iints run-full \
  --algo algorithms/example_algorithm.py \
  --predictor models/hupa_finetuned_v2/predictor.pt \
  --scenario-path scenarios/clinic_safe_baseline.json \
  --output-dir results/dual_guard

Quick Start (Python)

import iints
from iints.core.algorithms.pid_controller import PIDController

outputs = iints.run_simulation(
    algorithm=PIDController(),
    scenario="scenarios/example_scenario.json",
    patient_config="default_patient",
    duration_minutes=720,
    seed=42,
    output_dir="results/quick_run",
)

Most Used Commands

# Environment and validation checks
iints doctor --smoke-run
iints validation-profiles
iints validate-run --results-csv results/run_full/results.csv --profile research_default
iints contract-verify --output-json results/contract_report.json
iints replay-check --algo algorithms/example_algorithm.py --output-json results/replay_check.json
iints golden-benchmark --algo algorithms/example_algorithm.py --output-json results/golden_benchmark.json

# Scenario bank benchmark
iints scorecard --algo algorithms/example_algorithm.py --profile research_default --output-dir results/scorecard

# Research diagnostics
iints research audit-split --data data_packs/public/OhioT1DM/processed/ohio_merged.csv
iints research evaluate-forecast --input-csv results/dual_guard/results.csv
iints research evaluate-forecast --input-csv results/dual_guard/results.csv --gate-profile research_default
iints research registry-list --registry models/registry.json
iints research registry-promote --registry models/registry.json --run-id <run-id> --stage validated

Demos and Notebooks

• Demo scripts: examples/demos
• Notebook tutorials: examples/notebooks

Documentation

• Manual (PDF): docs/manuals/IINTS-AF_SDK_Manual.pdf
• Comprehensive guide: docs/COMPREHENSIVE_GUIDE.md
• Technical README: docs/TECHNICAL_README.md
• Plain-language guide: docs/PLAIN_LANGUAGE_GUIDE.md
• Evidence base (peer-reviewed sources): docs/EVIDENCE_BASE.md
• Research track: research/README.md

Evidence and Provenance

The SDK ships a source manifest grounded in peer-reviewed diabetes literature and standards.

iints sources
iints sources --category trial --output-json results/source_manifest.json

Safety Notice

This SDK is for research and validation only.
It is not a medical device and does not provide clinical dosing advice.