Loss is not Enough

This experiment compares two optimization trajectories against each other. For these two trajectories, we observe virtually the same loss curve, however, the actual movement in the loss landscape is dramatically different. We show that with the alpha distribution, we are able to distinguish both cases easily.

1. Run experiment: python run.py (creates a results directory with two __log files for the blue and orange optimization trajectory). Alternatively, extract the results.zip to use our original results: unzip results.zip.
2. Plot results: python plot.py. If you created the results yourself, you need to change the path to the __log files in the plot.py. Find the plot in the output directory.
3. Clean up or start over: bash clean.sh (removes results)