script has to checkpoint ~10 times (to be defined)
Draft description
[x] get 10 checkpoints for for smokeplume relaxed for the best alpha in wang2022. Use rotation equivariance (we might as well also use the other datasets if we are reproducing figure 4 anyway) full and partial (thijs has already done full rotation equivariance)
[x] low priority: reproduce figure 4 fully with checkpoints at the end points such that we can use lie derivative to compare to their metric for equivariance learned
Important details:
Draft description