In GabelHub/FAMoS: A Flexible Algorithm for Model Selection

knitr::opts_chunk$set(echo = FALSE)

FAMoS 0.3.0

Major changes

• Users can now enable FAMoS reattempts: If all search methods are exhausted, FAMoS will look for a distant model, jump there and continue the search. The algorithm terminates if the best model is re-encountered or if no new models are available.

• The option 'future.off' is now renamed to 'use.futures' to allow for a more intuitive use.

Minor changes

• FAMoS introduces the option 'interactive.session', a logical which allows to turn off interactive requests, e.g. when running non-locally.

• The function 'get.most.distant' now removes duplicates before looking for complementary models.

• The plotting function 'famos.performance' has an additional option to specify SCVs for reattempts

Bug fixes

• FAMoS now correctly assesses the number of managed future jobs in each iteration when 'refit = FALSE'.

• FAMoS now returns the best model if the algorithm terminates due to insufficient models for testing.

FAMoS 0.2.0

Major changes

• FAMoS can now handle cost functions that do not rely on AIC(c) or BIC. Any selection criterion can be used and FAMoS will attempt to minimise it. Therefore, the option nr.of.data was removed. The selection criterion value now needs to be returned by the cost function fit.fn.

• FAMoS can now use other fitting routines than optim. The custom fitting routines can be included in the cost function fit.fn. The use of a custom fitting routine requires to set use.optim = FALSE. In this case, the cost function fit.fn needs to return a list containing the selection criterion value as well as a named parameter vector describing the optimised fit.

Minor changes

• Graphic settings get returned to their previous values after calling FAMoS plotting functions.

• Log files are no longer saved using sink.

• The default number for optim.runs was set to 1.

• Futures are no longer used as default option.

• The function sc.order now plots logarithmically, if the ratio between minimal and maximal values is larger than 300.

• FAMoS now checks during each run, if the used cost function has changed. In this case, an interactive warning will be issued to the user.

• Included an option to silence the verbose output of FAMoS (verbose = FALSE).

• FAMoS now returns the total number of tested models.

• If a custom fitting routine is used, no previous applicability checks are performed.

Bug fixes

• Fixed the problem of multiple best models. If multiple best models are found, only the first one is used.

• The function famos.performance now displays the grid properly.

GabelHub/FAMoS documentation built on June 12, 2025, 2:16 p.m.