You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The time points when we evaluate frequencies should not exceed the date range of the observed data unless the user has requested a specific start and/or end date. As @rneher notes in the conversation linked above:
Pivots for KDE frequencies should never extend more than the narrow width beyond the last time window were there is more or less representative data. They would be too heavily influenced by fluctuations in that case. This is less of an issue for the diffusion frequencies.
Instead of setting the lower and upper date bounds based on the pivot frequency like so:
The latest approach to calculating pivots would guarantee that the last pivot matches the latest data point. However, there is no guarantee that a lower date bound based on the earliest observation would match that observation. We could change the logic of the pivot calculations to ensure that the earliest observation (or start date) is always included, though. This isn't a large change conceptually, but it would require us to update the expected behavior of the frequencies API as represented in our unit tests.
The text was updated successfully, but these errors were encountered:
Context
This issue arose as part of a conversation about pivot intervals for frequencies and what reasonable start/end dates should be.
Description
The time points when we evaluate frequencies should not exceed the date range of the observed data unless the user has requested a specific start and/or end date. As @rneher notes in the conversation linked above:
Instead of setting the lower and upper date bounds based on the pivot frequency like so:
we should consider at least using a max date of the latest observation like so:
The latest approach to calculating pivots would guarantee that the last pivot matches the latest data point. However, there is no guarantee that a lower date bound based on the earliest observation would match that observation. We could change the logic of the pivot calculations to ensure that the earliest observation (or start date) is always included, though. This isn't a large change conceptually, but it would require us to update the expected behavior of the frequencies API as represented in our unit tests.
The text was updated successfully, but these errors were encountered: