Reduce memory usage when pulling training events

[Lan2u]

We can be much more memory efficient if we are selective about what we pull back from google