BC03 3DoF, 1 Sphere, Compare Effect of #Demos

I ran PN++ naively with 10 demos to see what it would give us and to possibly see if we get better results with denser representations (sample efficiency).
These are with the v02 data, BTW, WITHOUT zero-padding.
Hmm ... seems like success is worse with 10 demos, but not that much worse.
Why are these susceptible to CUDA malloc errors on the cluster?
10 demos: I had to remove 2 of 5. Re-running 2 more on 05/21/2022.
100 demos: I had to remove 1 of 5. Re-running 1 more on 05/21/2022.
Update: the above should now be accounted for and resolved in reporting results.

PN++ 10 and 100 Demo Comparisons