| NEWS | R Documentation |
News for Package FlexParamCurve
Disagreement between package versions in the help files is resolved
Models that fail to fit are now assessed in the code using inherits rather than class "==" to resolve several issues with failed fitting of very complex models and fix errors with pn.modselect.step
Updated to work on new R code requirements
Fixed 4 bugs from 1.5.3 (FlexParamCurve 1.5.4 was not generally circulated):
1) verbose from modpar is now transitioning to pn.mod.compare and pn.modselect.step
2) modpar, pn.mod.compare and pn.modselect.step will now throw an error if any data values (x, y or grp) are NA
3) in pn.mod.compare and pn.modselect.step, existing = TRUE now works and will not fit any existing models within the working environment. These are transfered to the package environment, used, and transfered back
4) optional constraints on either end of curve (first.y, x.at.first.y, last.y, x.at.last.y) are now functioning
(note FlexParamCurve 1.5.2 was not generally circulated thus changes detailed are additions since 1.5.1)
Change in default approach to fitting monotonic curves:
Monotonic curves are exemplified by the logistic and Richards curve, and do not incorporate a recessional part of the curve. In previous versions, FlexParamCurve fixed recessional parameters to the values in the specified pn.options object (see
?modpar). However, if data used in modpar supported recessional parameters non-intuitive recessional plots were generated using monotonic curve models (modno 12
or 32). The new approach implemented in v1.5.3 is to by default ignore recessional parameters whenever modno 12 or 32 are optimized or plotted. To access the original approach (for those who wish to optimize pre-recession parameters only but retain fixed recession trajectories) set force.nonmonotonic = TRUE in modpar
Update of model selection routines (pn.mod.compare and pn.modselect.step)
The new default for pn.mod.compare is to proceed using variable models (those with a shape parameter, M, for the pre-recessional part of the curve) for subsequent model selection if neither of the initial models successfully converge. User preferences can be selected using either forcemod = 3 or forcemod = 4; see ?pn.mod.compare. In both selection routines the code has been adjusted to make forcemod work appropriately.
Implementation of required package dependencies: The new dependency information required by CRAN has been incorporated.
IMPORTANT: Change in the use of the Global Environment: In previous versions, FlexParamCurve has automatically written the pn.options file (with fitting estimates and user custom fitting instructions) to the Global Environment. Additionally, the model selection routines outputed all the fitted nlsList models to the Global Environment. To comform to R good practice for packages, the user can now specify that the Global Environment is not written to. To make this version back-compatible with previous FlexParamCurve versions, the default is to write to the Global Environment but this can be switched off setting the parameter Envir to a value for an alternative environment to write to. Internal parameters are written to the namespace environment, FlexParamCurve:::FPCEnv. This applies to the following function calls: modpar, SSposnegRichards, pn.mod.compare, pn.modselect.step, posnegRichards.eqn and change.pnparameters.
Addition of new function: get.mod.This is mostly used internally but can be used by user to copy files from one environment to another. By default this will copy files output from model selection routines (pn.mod.compare and pn.modselect.step) from the packge namespace to the global environment.
Correction of faulty examples. FlexParamCurve v1.4.3 was archived for reason of failing examples. These have been revised accordingly and pass appropriate tests.
Addition of a NEWS file: This file contains news for all recent versions of FlexParamCurve that were previously found at https://pennstate.academia.edu/SteveOswald/Posts Please check this website for instant updates
Use of Lazy Loading for data files:The package now uses lazyloading for all data files used in examples. Thus all data files can be used directly without the need to call data beforehand.
Bug fix for SSposnegRichards: Correction of a problem for selection routines pn.mod.compare and pn.modselect.step that was caused by failing to specify taper.ends values explicitly when initially running modpar. These two selection routines will now work alone whether specifying taper.ends or not and regardless of previous runs of modpar.
Error message update in pn.modselect.step: Addition of informative error message for this selection routine if no nlsList models are successfully fitted - induces safe termination of the routine with appropriate error message.
FlexParamCurve Documentation available: Paper, Online Tutorial and Introduction to the R package: Methods in Ecology and Evolution recently published a showcase paper on FlexParamCurve and is hosting two videos on the package, as online extras. These include an ~8-min introduction as to the purpose and utility of the package and a 12-min tutorial in R running you through some of the main functions in FlexParamCurve, including curve selection, model plotting and using FlexParamCurve models within mixed effects models. The paper is available at: http://onlinelibrary.wiley.com/doi/10.1111/j.2041-210X.2012.00231.x/abstract. Videos can be found at: http://www.methodsinecologyandevolution.org/view/0/podcasts.html
Major code updates have been made to the package in this version:
selfStart ( SSposnegRichards) and model selection routines run ~ 10 x faster and are more reliable!
more user-options to modpar and model selection routines to allow tapering at either end of the curve ( taper.ends)
custom specification of parameter bounds ( width.bounds)
passing arguments to nls and nlsList within modpar
automated bounds expansion for data exhibiting very different curves between different levels
modpar and SSposnegRichards require users to specify the name of a list object to write model parameters and fitting options to, so that multiple models can be fitted alternatively within one session
Various inconsistent names of functions have been updated
This new version is showcased in a forthcoming Methods in Ecology and Evolution Application paper
Thanks to Timothy Paine for his helpful suggestions that refined this version of FlexParamCurve
Significant improvement in the accuracy of optimizations so this version is to be prefered over v1.1. As in version 1.1, it can be used for a variety of complex parameteric growth curves or any relationships that exhibit multiple S-shaped curves between predictors and response variables
Added flexibility to model curves that follow a very general double-Richards type (not just positive-negative trajectory), e.g. it can now perform double-positive (or double-negative formulations), e.g. Meyer, P.(1994) Bi-logistic growth. Technological Forecasting and Social Change. 47: 89-102)
Ability to model two component models (where separate curves do not interact) as well as the traditional double curves. e.g. of two component models: Ross, J.L. et al. (1994) Age,growth, mortality, and reproductive biology of red drums in North Carolina waters. Transactions of the American Fisheries Society 124: 37-54. OR Murphy, S. et al. (2009) Importance of biological parameters in assessing the status of Delphinus delphis. Marine Ecology Progress Series 388: 273-291
It can estimate double-positive richards and double negative richards and should be able to accomodate a wide variety of data [if you have a dataset that doesn't fit then let me know and I'll work it in]
It is slightly slower for more complex datasets: speeding up optimization algorithms is in the pipeline
Option verbose added for use in modpar and SSposnegRichards to allow trouble shooting
I am happy to announce that the FlexParamCurve R package is now available on CRAN (http://cran.r-project.org/web/packages/FlexParamCurve/index.html)
This package provides a flexible selfStart function and model selection tools for fitting nls, nlsList and nlme models for data that follow positive and negative trajectories. The package was designed with growth analysis (see Huin & Prince 2000. Chick Growth in Albatrosses: Curve Fitting with a Twist. J. Avian Biol. 31(3): 418-425) in mind but provides support for any curve-fitting requiring combined positive-negative trajectories
It can be downloaded directly from R in the usual way: install.packages("FlexParamCurve")