Closed berndbischl closed 4 years ago
and really learn from the design in Impute.R and ImputeMethods.R to see how to delegate to specific algorithms, and how "reimpute" works during prediction time.
So, I worked on this since friday:
The code currently resides in the fda_pull1_task_featExtract branch.
It is mostly done similar to impute.R and imputeMethods.R
Current API:
t = extractFDAFeatures(fuelSubset.task,
feat.methods = list("UVVIS" = extractFDAMean(),
"NIR" = extractFDAFourier(trafo.coeff = "amplitude")))
It should return a Task
, but I have not solved that yet.
It currently returns a list, where
extractFDAFeatures
can be called for a data.frame
as well, then target
, fd.features
and fd.grids
, usually contained in the TaskDesc
have to be supplied as well.
For predict phase we can pass a Task and a extractFDAFeatDesc
t2 = reExtractFDAFeatures(fuelSubset.task, t$desc)
Questions:
Discuss function and object names!
For extractFDAFeatures.Task
and reExtractFDAFeatures.Task
we want to return a Task.
This can either be done by changing the data of the old task and removing the FDATask class
or by constructing a new Task and passing on the neccessary TaskDesc
values.
What else needs to be changed, so resampling automatically works?
We might want some way of appending extracted Features to a Task, for example if i want
min, max, mean
or Fourier amplitude
and phase
I would have to somehow do it once for each Method and then changeData?
Do we need a convertFDATaskToNormalTask
?
We could do this by simply wrapping around extractFDAFeatures.Task
or do we want to rename the
function?
@smilesun @lbeggel
This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.
ok, one problem is that we would like to have that function "act" on a task.
but we also need a basic "workhorse". for this we need to act on data (and target, do we really need the target for anything?).
so the current API is good, but we need to document that one line in "data" is always one complete time series.