- diffusion is now released under LGPLv2.1. The previous version under GPL(>=2) can be downloaded from here: https://github.com/mamut86/diffusion/releases/tag/v0.3.1
- new automated removal of leading and trailing NAs. Error abort if NA within series
- new search optimisation argument (multisol)
- new option to partially fix/optimise parameters w in both diffusion and seqdiffusion
- new bootloss argument to control fo r different loss function of the bootstrapping
- added documentation for tsMetal.RData
- added internal sum squared error function for the optimiser
- added passing optim setting int trough Gompertz and GSGompertz initialisation
- added error handling for optimx function
- renamed argument l to loss (for l-norm selection)
- renamed argument x to y (the response variable)
- renamed argument optim to method (the optimiser selection)
- introduced argument initpar to allow manual initialisation, preset and automatic
- introduced argument mscal to scale market potential for better optimisation results
- introduced argument bootloss to choose from different loss function for bootstraps
- changed non-critical warnings to type message, to allow easy suppression
- is.diffusion() and is.bass() functions
- renamed option variable l to loss (for l-norm selection)
- renamed option variable x to y (the response variable)
- updated documentation for diffusion() - Thx Ivan!
- removed tsBroadband dataset due to data privacy concerns
- new search optimisation option
- issues with log errors - reverted error measure to v.0.2.7 for the time being
- fixed an initalisation issue on the weibull model
- fixed titles for graphs
- fixed typo in documentation for predict function
- fixed fcst table to contain same number as horizons
- fixed number of parameters that go into G/GS (only relevant if w.idx is active)
- fixed error when plot.seqdiffusion() was used with cleanlead = F
- fixed error in gompertz when cleanlead = F
- fixed the tsmetal issue - the object now has the correct name
- fixed error when plotting ts-object with cumulative = F
- fixed an error when eliminate was TRUE with 0 pval reps to stop.
- fixed the sequential diffusion pval elimination process
- fixed error when linearization fails and reverting now to preset
- fixed error when bootstrapping produced NA now ignoring NA and warning.
- included error check for pvalreps in seqdiffusion
- included error check for y to be vector or ts-object / matrix or mts for seqdiffusion
- included a revert method in case optimization fails and warning messages
- added Weibull curve
- Changed the optimisation errors to log
- update description file
- fixed dataset urls
- updated naming for datasets
- merged examples into main code
- renamed sgompertz function to gsgompertz to reflect the correct name Gamma/Shifted Gompertz
- fixed some missing functions in norton-bass function
- Fixed roxygen2 documentation
- Fixed datasets
- Added difcurve function to produce curves for given parameters
- Reintroduced predict, print and plot functions for diffusion objects
- Created new diffusion class for outputs of diffusion()
- Hidden (until fixed) seqdiffusion and norton-bass
- Added method predict() for diffusion models
- Added maxiter and tolerance option control
- New option np in predict() which allows to generate new product forecasts with 0 data
- Fixed print issue for Gompertz
- Updated some of the documentation so it contains information about the new optim = parameter
- introducing optim paramter choice to allow for Nelder-Mead or Hooke-Jeeves method
- Fixed various RoxyGen2 issues and
- Fixed plotting issue
- Added print and plot to Namespace
- Updated some of the documentation so it includes details about parameter order of S-Gompertz
- Replaced if statements with switches where necessary - closing Issue #1
- Updated the optimisation process. This should give better results but still not perfect
- Added better commented for diffusionPlot() function - closing Issue #2
- Open tasks in News depreciated and now tracking them with Issues on github
- All the parameter optimisation are now using Subplex algorithm from the nloptr package
- In diffusion() When providing w, pval length was not using the right parameter
- Several small variable name errors due to the integration
- Introducing shifted-gompertz model to the diffusion() function
- Some typo in the print out function fixed
- carstock.RData file should now be available
- Depreciating function Bass() and Gompertz()
- Introducing function diffusion() and seqdiffusion (kindly donated by N. Kourentzes)
- Added dataset safari
- Added dataset windows
- Added cost function for Norton-Bass model
- Included curve fitting into Nortonbass() function
- Added curve fitting ability and insample error estimation for the Gompertz curve
- Added data "broadband"
- Added curve fitting and insample error estimation for Norton-Bass model