Gb/fix gpool

[grantbuster]

No description provided.

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 88.08%. Comparing base (1ce0ef1) to head (b634576).

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #50      +/-   ##
==========================================
+ Coverage   88.06%   88.08%   +0.01%     
==========================================
  Files          27       27              
  Lines        3185     3188       +3     
==========================================
+ Hits         2805     2808       +3     
  Misses        380      380              

Flag	Coverage Δ
unittests	88.08% <100.00%> (+0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[grantbuster]

@bnb32 option to review. The goal here is to implement a pooling layer with a gaussian kernel instead of a uniform average. i flip flopped between a tensorflow conv2d layer (implemented in the previous PR) with fixed kernel versus applying a kernel directly with the tf convolution method. The conv2d layer won't actually work because this needs to operate on each feature independently and a conv2d will apply a uniform kernel to all features and then sum across all features yielding a constant output.

[bnb32]

Cool. Were you thinking this might help with sharp edged artifacts?

[grantbuster]

Cool. Were you thinking this might help with sharp edged artifacts?

huh maybe but no i implemented this for the UHI work where we're training simple networks on point observations but want to include some input wide-field spatial information.

[bnb32]

Cool. Were you thinking this might help with sharp edged artifacts?

huh maybe but no i implemented this for the UHI work where we're training simple networks on point observations but want to include some input wide-field spatial information.

ah yeah makes sense.