ee_filters

[zackarno]

I like the new ee_filter functions. It would be really cool to be able to plug these into all sorts of code and not need to think about mapping with ee_utils_pyfunc & ee$filter$calendarRange anymore. One potential issue with doing this though is that it seems the final step of these functions is to composite the image with the chosen stat/reducer. Therefore, I am wondering if it might make sense to:

1. Separate the actual filter functionality into it’s own function(s). Something like ee_filter_calendar or even ee_filter_calendar_range. I feel like they could be all merged into one filter with a “by” argument (“month”,”year”,”year_month”)
2. Create a second function which does what the current ee_filter functions do which is filter + composite/reduce. This function would wrap the new filter function mentioned above. I am thinking something like ee_temporal_composite or ee_calendar_composite. I think this naming/verbage would be consistent with GEE documentation on compositing

Below is a rough skeleton of how I am currently imagining the functions could fit together just to give you the basic idea. There is a bunch missing and I don't think it simplifies the function code a great deal, but i think it could improve the syntax and also make the filter more portable. Let me know what you think. If you like the idea, I would be happy to give it a go.

#' @param imageCol imageCollection
#' @param by time unit ("month", "year" , "year_month")
#' @param months \code{numeric} vector, e.g. c(1,12).
#' @param years \code{numeric} vector min/max years e.g. c(2013, 2016).
#' 
ee_filter_calendar_range(imageCol, by, month, years){
  
  if(by=="months"){
    
  }
  if(by="years"){
    
  }
  if(by="year_month")
}
  
}


#' @param imageCol imageCollection
#' @param by time unit ("month", "year" , "year_month")
#' @param months \code{numeric} vector, e.g. c(1,12).
#' @param years \code{numeric} vector min/max years e.g. c(2013, 2016).
#' @param stat \code{character} statistic
 

ee_calendar_composite<- function(imageCol,by,stat,months, years){
  

  ee_reducer <- stat_to_reducer(stat)
  # filter
  imageCol_temp_filtered <- ee_calendar_filter(imageCol,by=by, months= months, years=years)
  # composite
  imageCol_temp_composite <-ee_reducer(imageCol_temp_filtered)
  
 # have to map again to set the properties - kind of annoying. I guess we could write a function to set properties based on "by", "month", 
 # and year
 
  if(by=="months"){
    months = ee$List$sequence(months[1], months[2])
    ee$ImageCollection$fromImages(
      months$map(rgee::ee_utils_pyfunc(function (m) {
        indexString = ee$Number(m)$format('%03d')
        ic_temp_filtered <-imageCol_temp_composite $
          set('system:index', indexString)$
          set('year',0000)$
          set('month',m)$
          set('date',ee$Date$fromYMD(1,m,1))$
          set('system:time_start',ee$Date$millis(ee$Date$fromYMD(1,m,1)))
      }
      )))
  }  
  if(by=="years"){
    
    # write some code
  }
  if(by=="year_month"){
    
    # some code
  }
  
    
}

[joshualerickson]

I'm looking at it right now. I like it. I have some questions below.

ee_filter_calendar_range()
So the first function ee_filter_calendar_range() would this have functionality like below? Where you are essentially abstracting away ee$Filter$calendarRange() and imageCollection$filterDate()? I wonder if we'd even need the by = call? We could just catch whats missing or what's not missing and apply filter from there.

imageCol |> ee_filter_calendar_range(months = c(8,12))
imageCol |> ee_filter_calendar_range(months = c(8,12), years = c(1990, 1999))

ee_calendar_composite()

This is essentially the ee_filter_* but with a by = argument, right? I think we could just have a top-level function ee_calendar_composite() and then whatever the by argument is that will call the ee_filter_* functions. I like this idea of combining, 👍.

I'm down if you want to tackle this!

[zackarno]

Nice, I will try combining and make a version where the condition is the "by" argument and another where it is based on which inputs are filled (month vs year vs month and year). We can then test them out and see how they feel