trnnick / tstools Goto Github PK

They are not normalised now which may be a root of several problems of restricted models.

Hello, thanks for the wonderful package. In the new version when we have straigh tline time series, theta method trows an error, this possible could how the theta method tests seasonality? Below is a reproducible example.

tsa <- ts(rep(10,20),frequency = 4)
theta(tsa)
Error in if (season.exist == TRUE) { :
missing value where TRUE/FALSE needed

When running seasonal models there is this:

if (Stype != "N" & (obs/datafreq) < 2) {
message("Not enough observations for the seasonal model. Only non-seasonal models are available.")
Stype <- "N"
}

If initial.season and initial are provided is this necessary? I cannot see why

It needs to be faster so it can be used for intervals construction in ets2.

Please see below, when I run the following code I get an error. I'm using ubuntu xenial R version 3.4.4 (2018-03-15).

Please see below a reproducible example.

library("Mcomp")
m3.monthly <- subset(M3,"monthly")
m3.history = lapply(m3.monthly, [[, 'x')

forecast(theta(m3.history$N1402),h=12)
Error in if (diff(range(colSums(ynt))) <= .Machine$double.eps^0.5) { :
missing value where TRUE/FALSE needed

I have the following issue when i try to install the library

devtools::install_github("trnnick/TStools")
Downloading GitHub repo trnnick/TStools@master
checking for file 'C:\Users\g.nebot.troyano\AppData\Local\Temp\Rtmps7vNDA\remotes5bc838c81319\trnnick-TStools-4912b2e/DESCRIPTION' ...

• preparing 'TStools':
  checking DESCRIPTION meta-information ...
• checking for LF line-endings in source and make files and shell scripts
• checking for empty or unneeded directories
• looking to see if a 'data/datalist' file should be added
• building 'TStools_3.0.0.tar.gz'

Installing package into ‘C:/Users/g.nebot.troyano/Documents/R/win-library/3.6’
(as ‘lib’ is unspecified)

• installing source package 'TStools' ...
  ** using staged installation
  ** R
  ** data
  *** moving datasets to lazyload DB
  ** byte-compile and prepare package for lazy loading
  Error: object 'hm' is not exported by 'namespace:smooth'
  Ejecución interrumpida
  ERROR: lazy loading failed for package 'TStools'
• removing 'C:/Users/g.nebot.troyano/Documents/R/win-library/3.6/TStools'
  Error: Failed to install 'TStools' from GitHub:
  (convertido del aviso) installation of package ‘C:/Users/GNEBOT~1.TRO/AppData/Local/Temp/Rtmps7vNDA/file5bc87a347db4/TStools_3.0.0.tar.gz’ had non-zero exit status

How can i fix it?
Thanks in advance!

See Kolassa, 2011 (Combining exponential smoothing forecasts using Akaike weights) for the details.

The literature specifies that
|Rank1 - Rank2| > CD
for Rank1 to be significantly better than Rank2
However the whiskers in the plots extend both sides making the equation
|Rank1 - Rank2| > 2CD
Can we include another plot option with 1 sided whisker?

Hi,

Looks like there might be a bug in create.inputs(). In the Exogenous section this: 'if (length(xreg.lags[[i]]>0)){' should be 'if (length(xreg.lags[[i]])>0){' for the code to function correctly.

Thanks for a great package,
Gavin.

Can you add the initial' andinitial.season' in the output list of the es function?

For now the options are: nloptr::cobyla and alabama::auglag. The former one gives suboptimal estimates in cases with small ammount of parameters to estimate. The later (with "BFGS", "Nelder-Mead" and "nlminb") has problems with estimating seasonal models.

Hi,
I have tried several times to install the TStools package using the below command:
devtools::install_github("trnnick/TStools")

But have encountered below error:

• installing source package 'TStools' ...
  ** libs

*** arch - i386
c:/Rtools/mingw_32/bin/g++ -I"C:/R/R-331.1/include" -DNDEBUG -I"C:/R/R-3.3.1/library/Rcpp/include" -I"C:/R/R-3.3.1/library/RcppArmadillo/include" -I"d:/Compiler/gcc-4.9.3/local330/include" -O2 -Wall -mtune=core2 -c RcppExports.cpp -o RcppExports.o
c:/Rtools/mingw_32/bin/g++: not found
make: *** [RcppExports.o] Error 127
Warning: running command 'make -f "Makevars.win" -f "C:/R/R-331.1/etc/i386/Makeconf" -f "C:/R/R-33~1.1/share/make/winshlib.mk" SHLIB_LDFLAGS='$(SHLIB_CXXLDFLAGS)' SHLIB_LD='$(SHLIB_CXXLD)' SHLIB="TStools.dll" OBJECTS="RcppExports.o ssgeneralfun.o"' had status 2
ERROR: compilation failed for package 'TStools'

• removing 'C:/R/R-3.3.1/library/TStools'

Is there any known issues here? Thanks!

There must be a way to further enhance performance. I will just put here some ideas that we could try:

• "apply" instead of "for", pre-allocation of vectors instead of dynamic addition ("low hanging fruit")
• RcppParallel. new package on CRAN http://www.r-bloggers.com/rcppparallel-getting-r-and-c-to-work-some-more-in-parallel/
• parallel - for distributing "ZZZ" across threads

I think we need to update the photo. The MSc students were not sure if it was I! :p

Hey, can you discover what is the error?

> if (!require("devtools")){install.packages("devtools")}
> 
> devtools::install_github("trnnick/TStools")
Downloading github repo trnnick/TStools@master
Installing TStools
'/usr/lib/R/bin/R' --no-site-file --no-environ --no-save --no-restore CMD  \
  INSTALL '/tmp/RtmpvXkzpf/devtools292f439b5469/trnnick-TStools-79f883b'  \
  --library='/home/bruno/R/x86_64-pc-linux-gnu-library/3.2' --install-tests 

* installing *source* package ‘TStools’ ...
** libs
g++ -I/usr/share/R/include -DNDEBUG   -I"/home/bruno/R/x86_64-pc-linux-gnu-library/3.2/Rcpp/include" -I"/home/bruno/R/x86_64-pc-linux-gnu-library/3.2/RcppArmadillo/include"  `Rscript -e "Rcpp:::CxxFlags()"` -fpic  -g -O2 -fstack-protector --param=ssp-buffer-size=4 -Wformat -Werror=format-security -D_FORTIFY_SOURCE=2 -g  -c RcppExports.cpp -o RcppExports.o
g++ -I/usr/share/R/include -DNDEBUG   -I"/home/bruno/R/x86_64-pc-linux-gnu-library/3.2/Rcpp/include" -I"/home/bruno/R/x86_64-pc-linux-gnu-library/3.2/RcppArmadillo/include"  `Rscript -e "Rcpp:::CxxFlags()"` -fpic  -g -O2 -fstack-protector --param=ssp-buffer-size=4 -Wformat -Werror=format-security -D_FORTIFY_SOURCE=2 -g  -c ets2fun.cpp -o ets2fun.o
g++ -shared -L/usr/lib/R/lib -Wl,-Bsymbolic-functions -Wl,-z,relro -o TStools.so RcppExports.o ets2fun.o -L/usr/lib/R/lib -lR
installing to /home/bruno/R/x86_64-pc-linux-gnu-library/3.2/TStools/libs
** R
** data
*** moving datasets to lazyload DB
** preparing package for lazy loading
** help
*** installing help indices
** building package indices
** testing if installed package can be loaded
Error in dyn.load(file, DLLpath = DLLpath, ...) : 
  unable to load shared object '/home/bruno/R/x86_64-pc-linux-gnu-library/3.2/TStools/libs/TStools.so':
  /home/bruno/R/x86_64-pc-linux-gnu-library/3.2/TStools/libs/TStools.so: undefined symbol: dgetrf_
Error: loading failed
Execution halted
ERROR: loading failed
* removing ‘/home/bruno/R/x86_64-pc-linux-gnu-library/3.2/TStools’
Error: Command failed (1)
> 

I'm using ubuntu 14.04 LTS and R version 3.2.2

Thanks!

A couple of suggestions on the model input.
Should we replace X with C (to better connect with combination) and use X to restrict Z to only additive options?

Have you had the time to compared this with ets? Speed/accuracy?

I think it might be helpful to add manual bounds for the smoothing parameters of es.
Currently there are bounds=c("usual","admissible")
One may still want to restrict the search space further. Perhaps add one-two new inputs for manual min max bounds. These could be -inf +inf (or whatever) by default, but could then be restricted by the user to the desirable values. If that is a vector then in the cost function these can be passed as additional restrictions on top of the usual/admissible bounds.
I am in particularly interested in restricting phi further, but I can think of scenarios that this might be helpful for the other parameters.

One alternative would be to pass these directly as the limits for the usual bounds, though this may not be as intuitive for the user. I am not in favour of this option.

Also these is a typo on the bounds input at the es.Rd

For now the model selection for trace is done not 100% properly. See the example code:
x <- sim.ets("AAN",obs=100)
test <- ets2(x$data,h=10,trace=TRUE)

Usually multiplicative error model is chosen.