v0.5.5: WandB / MLflow experiment tracker integration

What's new

Experiment tracker integration

After any audit command, push metrics directly to your experiment tracker:

# Weights & Biases
deltatau-audit audit-sb3 --model m.zip --algo ppo --env CartPole-v1 \
    --wandb --wandb-project my-project --wandb-run baseline

# MLflow
deltatau-audit audit-sb3 --model m.zip --algo ppo --env CartPole-v1 \
    --mlflow --mlflow-experiment my-experiment

Or from Python:

from deltatau_audit import run_full_audit
from deltatau_audit.tracker import log_to_wandb, log_to_mlflow

result = run_full_audit(adapter, env_factory)
log_to_wandb(result, project="my-project")
log_to_mlflow(result, experiment_name="my-experiment")

Logged data

• Scalars: deployment_score, stress_score, robustness_score, reliance_score, sensitivity_mean, per-scenario scenario/<name>/return_ratio
• Params: deployment_rating, stress_rating, reliance_rating, quadrant, _deltatau_version

Install

pip install "deltatau-audit[wandb]"    # wandb>=0.12
pip install "deltatau-audit[mlflow]"   # mlflow>=2.0

Tests

33 new tests (307 total). Graceful degradation: missing tracker package prints WARNING rather than crashing.

Full changelog: https://github.com/maruyamakoju/deltatau-audit/blob/main/CHANGELOG.md

Full Changelog: v0.5.4...v0.5.5