-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathChangelog
65 lines (54 loc) · 3.08 KB
/
Changelog
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
# 2022-06-12 v0.3.1 [AG]
# Documentation converted to Roxygen.
# Added workflows for checking.
# 2022-04-16 v0.3.0 [AG]
# Replaced all snow and doParallel functionality with future package. The cores
argument is now removed from options of all functions and instead the user
can just setup a plan via the future package which is much simpler and allows
much more customization. This may break backwards compatibility with versions
less than this which made use of the cores argument.
# Bumped version to 0.3.0 across all affected packages.
# Made solver nlminb and autodiff (TRUE) defaults across all functions.
# Export the tsmoments method for analytical forecast moments
# 2022-04-08 v0.2.5 [AG]
- The innovations are now returned in the tsdecompose (component decomposition)
method for both the estimated and predicted model (in transformed space).
- The tsdecompose method now takes additional argument simplify to return the
standard named components Trend, Seasonal, ARMA and Irregular
- The tsdecompose returns the components pre-lagged (including the seed state)
so that they can be summed with the error term (Irregular) to obtain the exact
predictive distribution (in transformed space) against each date index.
# 2022-03-30 v0.2.4 [AG]
- Added an auto_issm function for automatic model selection (similar to auto_ets
in the tsets package).
# 2022-03-28 v0.2.3 [AG]
- Added tstransform to specification allowing for logit transformation
# 2022-03-16 v0.2.2 [AG]
- Added tscalibate and enhancements to the predict function
# 2021-09-03 v0.2.0 [AG]
- Fixed tsprofile error
- Added -DARMA_64BIT_WORD=1 in Makevars to avoid errors in very large systems.
# 2021-09-03 v0.1.9 [AG]
- Added option in tsbacktest to re-estimate the model every n periods expanding
the default of re-estimating every 1 period.
- Added option in tsbacktest to apply a custom function to each path draw (e.g.
sum, max etc) since we may be interested in having the distribution of some
statistic over the entire forecast horizon.
# 2021-08-08 v0.1.8 [AG]
- Added option for rescaling the mean and variance of the simulated prediction
distribution by their exact moments (per the Hyndman paper). Will no also return
the analytic_mean in the prediction object.
# 2021-06-20 v0.1.7 [AG]
- Added handling of missing values using the prediction filter step and calculation of the
likelihood only for non missing values.
- Added autodiff option (calls tsissmad package)
- Prettified the returned summary to display abbreviated model description and use
kable (similar to how tsets displays the summary).
# 2021-04-19 v0.1.6 [AG]
- Exported tsmetric method on predict object and fixed an issue with calculating
MASE (by passing lowest frequency to object).
# 2020-09-16 v0.1.5 [AG]
- Added parameter transformation function for unconstrained optimizers (optim/Nelder-Mead)
- Fixed specification for cases without seasonality or regular seasonality (bug in the ss_matrices.R file)
- Added some checks to tsbacktest to not fail completely if an estimation had problems, instead returning
NA's for the predictions of that run and calculating metrics on only good data (non NA)