imuops 0.3.1

IMU data QA, reliability scoring, and reproducibility tooling

imuops

imuops is a public alpha for IMU data QA, reliability scoring, and reproducibility.

It is built around one core question:

Is this IMU data trustworthy, why is it failing, and how does that affect baseline algorithms?

The product is intentionally tabular-first. The main path is customer-shaped csv, tsv, or parquet data plus a small YAML mapping file. Benchmark adapters stay in the repo for reproducibility demos, but they are secondary to the tabular workflow.

Status

• alpha / preview release
• main product path: tabular
• benchmark/demo adapters: ronin, oxiod, wisdm
• local-only contrib adapter: legacy_arduino

Benchmark adapters are validated on fixtures and demo flows. They are not guaranteed across every upstream packaging or layout variant.

Install

Requires Python 3.11+.

Primary public path:

python3.12 -m venv .venv
source .venv/bin/activate
pip install .

Use any Python 3.11+ interpreter available on your machine.

Optional uv path:

uv venv
source .venv/bin/activate
uv pip install .

imuops is prepared for PyPI publication, but this alpha does not claim a live PyPI package yet.

Quickstart

After install, the offline first-run path is exactly three commands:

imuops ingest tabular examples/sample_tabular_imu.csv --config examples/sample_tabular_config.yaml --out output/sample_tabular_demo
imuops audit output/sample_tabular_demo --summary-format markdown
imuops report output/sample_tabular_demo --out output/sample_tabular_demo/report.html

You can also use the bundled demo wrapper:

bash examples/run_tabular_demo.sh

What It Does

• normalizes messy IMU tables into one canonical session format
• audits timing, clipping, dropout, magnetic disturbance, bias drift, and related issues
• computes a versioned trust_score with explicit penalties, weights, and thresholds
• replays conservative baseline algorithms for orientation and PDR reproducibility
• benchmarks task-aware baselines where labels or trajectories exist
• compares clean vs corrupted or before vs after sessions
• batches QA over many sessions and writes machine-readable summaries for CI use

Core Commands

imuops ingest tabular /path/to/session.csv --config /path/to/mapping.yaml --out output/session_a
imuops audit output/session_a --fail-below 0.80 --summary-format markdown
imuops export output/session_a --profile qa_filtered --format parquet --out output/session_a_clean
imuops compare output/session_a output/session_b --out output/compare.html
imuops batch audit output --out output/batch_artifacts

Included Adapters

Market-facing default

• tabular: customer-shaped csv, tsv, and parquet sources with YAML mapping and unit conversion

Benchmark/demo adapters

• ronin: clean inertial odometry benchmark sessions
• oxiod: clean handheld / phone inertial odometry benchmark sessions
• wisdm: lightweight HAR benchmark sessions

Contrib/local regression

• legacy_arduino: historical Arduino/MPU9255 adapter kept only for local regression and examples

Trust Score

imuops publishes the trust-score contract directly into artifacts and reports:

• per-window formula
• session aggregation formula
• penalty totals
• weight profile
• thresholds

That is documented in docs/trustscore.md, and the current validation tranche is in docs/trustscore_validation.md.

Docs

• Architecture
• Datasets
• Trust Score
• Trust-Score Validation
• Schema Compatibility
• Release Checklist
• Contributing

Public Alpha Notes

• This repo is a truthful alpha, not a commercial deployment claim.
• Reports support --redact-source-path and --redact-subject-id for safer sharing.
• Output quality is strongest for tabular customer data and fixture/demo benchmark layouts.

License

MIT