hexagonal convolution

Open geoHeil opened this issue 6 years ago • 2 comments

Hi,

https://www.youtube.com/watch?v=z3PaGIQTFSE greatly explains efficient spatial smoothing using H3 hexagons and convolutions.

Is experimentalH3ToLocalIj the convolution suitable coordinate system outlined in the video? _Uber_Open_Summit_2018__Hexagon_Convolution_for_Data_Smoothing___Forecasting_-_YouTube

Can I find further information regarding this topic somewhere? This sounds interesting.

May 30 '19 05:05 geoHeil

Yes, experimentalH3ToLocalIj is the API used for the conversion.

What other further information would you like to get on this topic?

Jun 07 '19 22:06 chonguber

Thanks @chonguber . In the video, you mention this runs on GPUs/pytorch. Do you also run this in a distributed computing environment spark)? When I do not require deep learning, for now, and just want the convolution operation for smoothing, how is an efficient way to implement this in spark, i.e. without causing lots of shuffles?

Jun 08 '19 03:06 geoHeil