trendbreaker

trendbreaker copied to clipboard

asmodee() shouldn't incorporate lm()

1

As I understand trendbreaker is designed for the analysis of count time series. Thus `lm` (with its assumption of normal distribution, with support `[-Inf, Inf]`) shouldn't be used.

stephaneghozzi

While looking at a special case, I found outlier detection is not deterministic. In the code below, `asmodee` is applied 100 times to the same time series, but only a...

stephaneghozzi

Investigate integration with `incidence` package

1

Maybe useful, but not a priority at the moment

dirkschumacher

Add print/format functions to class epichange_model_fit

1

dirkschumacher

Check via requireNamespace if optional packages are loaded

3

Before using them. Otherwise trigger an error. For `brms` and `MASS`.

dirkschumacher

model output for unquoted family argument (cosmetic)

3

Currently if an unquoted argument is given for the `family` parameter the formula in the resulting model is the entire function call. Borrowing from the glm code we can get...

TimTaylor

Improve cross validation to handle time structure

3

See here: https://stats.stackexchange.com/a/14109

dirkschumacher

When using more than one metric the column sorting in the leaderboard is not stable. Thus the best model is not selected by the first metric specified by the user...

dirkschumacher

Everything works great, just 2 minor suggested improvements to make users' lives easier: - You have custom error catching if `max_k` is too large in `asmodee()` , would be nice...

qleclerc

Would be nice to include that for testing, but compiling the model takes too much time for regular testing. Maybe we can add a larger integration test similiar to the...

dirkschumacher