MUGA LAB Undergraduate Research – Results Directory

This directory serves as the primary output location for all MLflow experiment runs, model artifacts, and logged metrics.

Each experiment stage (Baseline Tuning, Calibration, Distillation, etc.) automatically writes its logs to the MLflow backend under this directory.

Directory Overview

Folder / File	Description
`mlruns/`	Default MLflow tracking directory containing all experiment runs and models.
`README.md`	This documentation file.

Each subfolder in mlruns/ corresponds to an MLflow experiment:

MLflow Experiment	Script Source	Output Content
`Baseline_Tuning`	`01_baseline_tuning.py`	Hyperparameter trials, validation metrics, trained models.
`Temperature_Scaling`	`02_temperature_scaling.py`	Optimal temperature (T^*), calibration metrics (ECE, MCS, NLL).
`Distillation_Experiment`	`03_distillation_experiment.py`	Teacher and student model metrics and artifacts.
`Cross_Architecture_Evaluation`	`04_cross_architecture_eval.py`	Comparative metrics across architectures.
`Calibration_Summary`	`05_calibration_summary.py`	Aggregated summaries and exported reports.
`Seed_Sensitivity`	`utils/seed_sensitivity_utils.py`	Multi-seed reproducibility and robustness logs.

To visualize results locally:

mlflow ui --backend-store-uri ./mlruns