Closed mleseach closed 3 months ago
To me, extrapolate=true is sufficient and the cost of maintaining some odd boundary behavior likely is not worth it. @sathvikbhagavan can have an opinion
Yes, I don't think it is useful supporting this in this package. extrapolate=true
should be sufficient.
Closing as not planned, but thank you for the input @mleseach as it is an interesting suggestion but we don't think it's the right idea at this time. As mentioned in the OP, we currently think extrapolate=true
should be sufficient for most users, but if this is a frequently asked for change we'd be willing to reconsider it.
Is your feature request related to a problem? Please describe.
When dealing with pre-computed interpolations but with a different time parameterization, mapping from one parameterization to another can cause precision errors, leading to an extrapolation error if
extrapolate=false
.Describe the solution you’d like
I think the best solution is to allow some tolerance when checking for extrapolation error (e.g. something like
t < itp.t[1] - eps(t)
instead oft < itp.t[1]
for the lower bound)Additionally, the tolerance could be parameterized, allowing users to specify a custom tolerance level if the default isn't ideal for their use case. But in this case maybe
extrapolate=true
could be sufficient.If you think this feature is within the scope of this package, I can create a draft PR and discuss the details.