Re: [ferret_users] Selecting groups of data

["Ansley C. Manke" <ansley.b.manke@xxxxxxxx>]

Hi Alexander,
 
No, this kind of striding isn't available in Ferret/PyFerret.
 
Do you want to compute a seasonal climatology? Or an average to an axis 
with 4 points per year? If so then a regridding operation can do it for 
you without your having to specify strides yourself
 
For climatologies, here are some details -- see the note at the bottom 
about the "clim_axes" datasets which define a variety of seasonal 
calendar axes for both the Gregorian calendar and other calendar definitions
 
../../faq/how-do-i-calculate-climatologies-and-climatological-anomalies
 
To regrid to a seasonal calendar, you would define your calendar axis 
and use a regridding operation.
 
If the grid is still too large to work with for one of the above 
operations, or if you're doing some other kind of analysis you can work 
on the grid in pieces, appending to a netCDF file in some dimension 
other than time. You need to first set up the file with the full 
range, and then append into it, filling the grid piece by piece.
 
../../documentation/users-guide/converting-to-netcdf/SIMPLE-CONVERSIONS-USING-FERRET#_VPINDEXENTRY_1071
 
-Ansley
 
On 1/3/2020 7:23 AM, Alexander Audet wrote:
> Ferret User Community
>                    I am trying to work with a very 
> large grided dataset and trying to get as high a resolution 
> aspossible. I am trying to correlate 1 season of data across a number 
> of years to a time series. Thus, it appears that I need to average 
> 498355200 points of data into four seasons then perform a stride so 
> that only the winter season is selected (right now defined as JFM, 
> later I will need to compare to DJF).
>  
>                     The problem is that there are 
> too many data points for my ferret to averagethe data into 4 seasons 
> as it gives me a memory error when I try to work with the resulting 
> variable. If I first trim down the data (stride over several degrees 
> Lat and Long) into a variable, the correlation does work. I think that 
> if I could first isolate JFM monthly data from the rest of the year, 
> then average those, I will be able to perform@AVE over a larger part 
> of the dataset of interest to me and create a higher 
> resolutioncorrelation.
>  
>                     Is there a way to select 
> chunks of data? Say if I had 24 months over 2 years, so that L=1:24, 
> somehow select 1:3, then 13:15? I could imagine using a command like 
> 1:24:3:12, (select 3 points at every 12th value). My browsing through 
> the documentation makes it seem impossible, but it seems like such a 
> simple request, I have a hard time believing that.
>  
> Alexander Audet