Closed DanielJWood closed 6 years ago
We've shared these results with research team and waiting on their feedback. If we only show data to some limited amt of precision, we can prob go ahead with no further action (maybe we need to programmatically generate a lot of results in JS and R to get an idea of the size of the error bars? TBD). But we might need to try to reduce error. @DanielJWood thinks it might have to do with tanh
inaccuracies.
See differences.xlsx for examples of several scenarios and results in R and JS with rounded and unrounded numbers. Not sure exactly why the unrounded numbers are not exactly the same as the R results but they're almost all <1% different, when calculated correctly.
@bchartoff please let me know what you think is next.