Open lennijusten opened 3 years ago
Can I just check quickly - metpy v1.0.1 doesn't use Dask yet for wet bulb temperature calculations does it? Because none of my workers are doing anything when I use it (passing in chunked xarray DataArrays).
@aurelgriesser Due to the current implementation of moist_lapse
, wet_bulb_temperature
is iterating point-by-point through the arrays using nditer
; there's probably any number of reasons in that chain of calls where the Dask arrays end up converted into plain numpy arrays. We definitely want to improve this is the somewhat near future.
Can I just check quickly - metpy v1.0.1 doesn't use Dask yet for wet bulb temperature calculations does it? Because none of my workers are doing anything when I use it (passing in chunked xarray DataArrays).
@aurelgriesser To speed up the wetbulb calculations I ended up calculating wetbulb over a range of reasonable temperature, pressure, and dewpoint. The result is a 3D volume of calculated wetbulb temperatures that can be queried for any combination of temp, pressure, and dewpoint.
Here is a link to that project if you are interested: https://github.com/lennijusten/Wetbulb-Temp/tree/main/Reference%20Grid
Hello, I'm using Dask to chunk xarrays and pass them into the
metpy.calc.wet_bulb_temperature
function and am getting a cryptic error message that I believe is due to something breaking internally in the wetbulb function. I understand from Unidata/MetPy#1479 that Dask integration is an ongoing project, so my intention is just to document this issue.The dask cluster works fine when calculating dewpoint and for the first set of
metpy.calc.wet_bulb_temperature
processes like opening the datasets, boradcasting, etc, but then it freezes progress at the process shown in the screenshot and returns the following error:The process actually keeps running after this, but the dask interface shows that none of the workers are being used.
Relevant packages in my Conda environment: