Assess the Effect of Lossy Compression on Models

[WardLT]

Use a few proxy problems (e.g., computing the capacity, fitting an ECM) to determine how much a derived result is affected by compression.

[victorventuri]

I think that's a good idea! Mainly two comments:

1. when thinking about the ECM fit, we want to, in particular, test the cases where we have two (or more!) RC elements with very different time constants (orders of magnitude), as the ones with steeper voltage decays are likely the ones to be most affected.

2. it would also be good to check if there are any particular implications for the Physics-based models our collaborators may use. Their higher degree of sophistication (compared to ECMs) probably makes them more susceptible to floating point and other numerical errors (I think even some of the ECM fits for RPTs we have been working on have had some of these issues)

At any rate, I don't know what would be the most consistent way of testing that. It might be excessive to test on the whole data, but I don't know which examples would be most representative of (a) our current datasets and (b) the data we will receive in the future. Ultimately, it would be fantastic to have an idea of how much error this compression may induce in our analysis, but is there a good way of finding an upper bound on this error?