Use fake test set data on O to graph input sample requirements.

[benwulfe]

We need a better understanding of how the number of input samples affects final precision. Using the lower right quadrant partitioned as a fixed size test set, lets test varieties of input samples to see how precision changes. This will present as evidence of our due diligence in determining the smallest amount for MVP.

There are likely two aspects to measure on a graph.

1. Number of geographically dispersed input samples (ensure all inputs are geographically dispersed) vs precision/loss against test set
2. Given a smaller set of geographically dispersed sites, varying the number of samples collected at each site vs precision/loss against test set

[benwulfe]

i think this will naturally come out as we do the validation and t tests.