In djbetancourt-gh/funGp: Gaussian Process Models for Scalar and Functional Inputs

TODO or not TODO

• (x) is used for tasks that are completed,
• (WIP) is used for tasks that are in progress,
• (?) Idea to be discussed.
• () FB 25/02/2022 test git push
• () FB 07/03/2022 test git push macbook

Checks

• (x) Check that the dimensions/the content of the data given as inputs for the predict and the simulate methods are consistent with the content of the object used to predict or simulate from.

summary methods

• (x) Add a summary method for the class "fgpm".

• (x) Add a summary method for the class "Xfgpm".

plot methods

• (x) Add a plot method for the class "fgpm".

• (x) Add a plot method for the class "Xfgpm".

• (WIP) For the plot methods that produce two plots on a same graphic, add an extra argument horiz to allow the user to chose the the way in which the plots are set. This will be useful for reports or articles such as the JSS.

• Add an horiz argument in the plotmethod for the S3 class "predict.fgpm"and in the plotPred method/function.

• (?) Add autoplot methods allowing to get *ggplot2 graphics.

Class "Xfgpm"

• (x) Store a copy of sIn, fIn and sOut as three new slots of the class. This will help at recreating any fgpm model visited from the data used in the fit. Maybe write a new function for that aim? How to specify the model to be re-created: by its number?

• (x) Improve the summary method. Should work correctly even when there are many inputs. Maybe add options to provide details on the scalar and the functional inputs separately in order to avoid having a table which is too wide.

• (x) When an object is created by using fgpm_factory, the progression bars whould be displayed only in an interactive session i.e., where interactive() is TRUE.

S3 classes "predict.fgpm" and "simulate.fgpm"

• (x) Set the S3 class ot the output of the predict and simulate methods to "predict.fgpm" and "simulate.fgpm" both inheriting from "list". This allows to write (S3) plot methods without breaking the inherited standard behaviours (in print, ...).

Simplifications

• (?) Should we still export the methods plotSims, plotPred, ... ? As a compromise we could continue to export the functions, but no longer turn them into S4 generics and methods because e.g., plotEvol is unlikely to be mnemonic for the user or to be used in some other package.

• (?) Rename the formal arguments of the predict and simulate methods for the class "fgpm" to use the same formal names? Could be for instance sIn.new and fIn.new or even simply sInand fIn, since it is clear that this must be "new" values.

• (?) Rename fgpm_factory into Xfgpm to conform to the R rule: the creator of a class has the same name as the class itself.

Tests

• (?) Add tests to make sure that the checks on the size of objects actually do what we expect.

New functions

• Add a function model_def (or other name) that takes Xfgpm object and returns a parsed code for one or several model. Bys using eval on this It could also return a list of model definitions. Then eval(model_def[[4]]) would rebuild the 4-th model. Using fIn and sIn in the current env?

Package documentation

• Read again carefully the doc for the new functions and methods. Find the "XXXY" code that shows something that still requires work.

• In the examples the output provided as comments should be removed, because this is an unusual practice in R. This behaviour is for vignettes. Some examples can be made simpler.

• In the examples the plot methods should be used in place of plotLOO, plotPreds, ... Similarly summary can be useful.

modelDef

• Find out why the class "modelCall" (with no method other than show exists. Maybe then rename modelDef into modelCall?

• Make the Examples section faster to run (using dontrun?) and improve the example which leads to a much smaller Q2 when evaluated on the new data.

Misc

• No longer export formatShort.

NEWS

Try

library(funGp)
example(plot.predict.fgpm)
example(plt.simulate.fgpm)

djbetancourt-gh/funGp documentation built on May 13, 2024, 1:17 p.m.