v1.5.1¶
v1.5.1 packages a more scientific public SDK workflow: stronger study tooling, clearer protocol bundles, and a more reproducible evidence path.
Highlights¶
- Added a scientific workflow layer for hypothesis-driven studies:
iints study-protocoldocs/SCIENTIFIC_WORKFLOW.md- protocol bundles with
STUDY_PROTOCOL.md,study_design.json, andstudy_matrix.csv - Added controlled corruption tooling for clean-vs-corrupted experiments:
iints data corrupt-for-study- timestamp shifts, missing blocks, duplicated rows, glucose spikes, dropped meal annotations, and unit-scale errors
- Added a fixed study-preset path for repeatable benchmarking
- Extended study analysis:
- descriptive statistics with standard deviation and 95% confidence intervals
- failure analysis for severe hypo, early terminations, and safety-heavy runs
- optional external plausibility comparison against imported CareLink metrics
- Extended study comparison:
- confidence intervals on differences in means
- Cohen's d effect estimates for key metrics
- Improved study posters:
- failure-analysis visibility
- external plausibility verdicts
New Scientific Commands¶
iints scenarios export-study-pack --output-dir scenarios/study_pack
iints study-protocol --output-dir results/study_protocol
iints data corrupt-for-study data/demo/diabetes_cgm.csv \
--output-csv data/demo/diabetes_cgm_corrupted.csv \
--mode timestamp_shift \
--mode missing_block \
--mode glucose_spikes
iints analyze results/study \
--output-json results/study_summary.json \
--output-markdown results/study_summary.md \
--output-csv results/evidence_table.csv \
--output-evidence-markdown results/evidence_table.md \
--carelink-metrics results/personal_carelink/carelink_metrics.json
iints compare-study results/study_clean results/study_corrupted \
--output-json results/study_comparison.json \
--output-markdown results/study_comparison.md
iints run-study \
--algo algorithms/example_algorithm.py \
--output-dir results/study_bundle
Why This Release Matters¶
Before v1.5.1, the SDK already had simulation, certification, AI review, and poster tooling, but the scientific story still depended too much on manually assembled study folders and ad-hoc experiment design.
With v1.5.1, the SDK can now:
- define the study before running it
- document the intended corruption operators
- keep a fixed clean/corrupted/supervisor-off matrix
- quantify uncertainty and effect size
- compare simulation behavior against imported real-world glucose metrics
That makes the project much easier to defend in front of a science jury, a public audience, or a technical reviewer.
Installation¶
Latest release:
python -m pip install -U "iints-sdk-python35[mdmp]"
Pinned:
python -m pip install -U "iints-sdk-python35[mdmp]==1.5.1"