Hyperparameter Tracker
The hyperparam_tracker module wraps GPy’s optimizer to record a snapshot of
all hyperparameters and the log-likelihood at each iteration. It detects
convergence issues (oscillation, NaN parameters, likelihood degradation) and
exports the full trajectory to a pandas DataFrame for offline analysis.
When to use: when you want to understand optimization dynamics — not just the final parameter values — or diagnose a model that isn’t converging cleanly.
import gpclarity
tracker = gpclarity.HyperparameterTracker(model)
tracker.wrapped_optimize(max_iters=200, capture_every=5)
report = tracker.get_convergence_report()
print(report["converged"], report["n_iterations"])
issues = tracker.detect_optimization_issues()
if issues["has_issues"]:
print(issues["summary"])
df = tracker.to_dataframe() # full history as a pandas DataFrame