- Added the signature
`c("disaggR","missing")`

for Ops group generic. (PR #54) - Inner calls to aggregate are now redirected to a faster non-exported function. (PR #55)
- Estimation spans and outliers are now handled in preset models in rePort and reView (PR #57)
- NULL labels are now removed even outside the plot margins (PR #59)
- The
`in_scatter`

function now substracts the outliers contributions from the low-frequency serie before computing the in_scatter comparison. (PR #63) - The
`in_scatter`

has now an additional arguments :`type`

(as every in_ function). For now, the only use for the type argument is to allow changes scatterplots for levels models. (PR #63) - The smoothed.part, for differencied benchmarks, has been set to a new base (its aggregated value is 0 in 2000). That way,
`reUseBenchmark`

is fixed if used on high-frequency series that have a different start than the previous one. This has no impact on the benchmarked serie. (PR #65)

- New vignette : Introduction to disaggR.
- The error of
`in_disaggR`

for wrong`type`

arguments has been changed, because it didn’t include “contributions”. `cex.axis`

,`xlim`

,`ylim`

,`cex.lab`

and`cex.main`

parameters now overwrite their default if used inside`plot(...)`

dots.- By default, in the
`plot`

and`autoplot`

methods, the axis annotations are now automatically set to be smaller if needed.

- Switched ggplot2, rmarkdown and shiny from imports to suggests. RColorBrewer has replaced scales as an import. disaggR can now be installed with far less dependencies. Hence, the
`autoplot`

generic is not reexported anymore. ggplot2 has to be attached to allow the use of`autoplot`

without the`ggplot2::`

prefix, by example with`library(ggplot2)`

.

- added support of outliers.
- In
`twoStepsBenchmark`

, the set.coeff names used to be replaced by`"hfserie"`

if`NCOL(hfserie) == 1L`

and`length(set.coeff) == 1L`

. This behavior was contradictory with the documentation if`set.coeff=c(constant=1L)`

. As for now, set.coeff names will never be replaced. Then, it makes the controls stricter because`set.coeff=c(x.name.herited.from.anywhere=1)`

will lead to an error. - as for the
`in`

time-series plots, the y window now ignores the infinite values.

- fixed some issues with ts.eps-delayed tsps.

- reView : a shiny reviewing application for twoStepsBenchmarks.
- rePort : a rmarkdown html report for twoStepsBenchmark and reView outputs.
- start.domain and end.domain know crops the hfserie
*after*having calculated the coefficients. That way, it is possible to evaluate the coefficients on a full hfserie, cropping them for the application. `in_sample`

now generates a more general class`"tscomparison"`

, with a`"in_sample"`

func attribute. In previous versions, the S3 class was named “insample”.- the new functions
`in_scatter`

,`in_benchmark`

and`in_revisions`

also produce tscomparisons, with plot and autoplot methods. - the graphics are prettier thanks to the package scales.
- the plot and autoplot methods now have xlab, ylab, start, end, col, lty, show.legend, main and mar arguments. The autoplot methods have also a theme argument.
- removed the c++ code to improve readability.
- most of the stats methods for time-series now coerce twoStepsBenchmarks or rateSmooths into time-series.
`reUseBenchmark`

now induces a`set.smoothed.part`

element in model.list if`reeval.smoothed.part`

is`FALSE`

.`threeRuleSmooth`

makes it easier than bflSmooth to procede to a rate smooth.

- various optimizations including cache for bflSmooth, which is now much faster, and alternative internal methods for time-series.
- added a weights arg to bflSmooth, that reproduces the
*lissage en taux*methodology. - the praislm and twoStepsBenchmark summaries now print some disclaimer to tell if the regression includes a differenciation.