Closed brendan-m-murphy closed 1 year ago
I pushed a fix for this. Quick fixes include: using encoding = None
or setting zlib=False
in the comp
dictionary. The fix I added only removes compression for data variables with data type "Object" or "Unicode" (in practice, this just means the "sitecodes" data variable); this gives files that are about 3 times smaller than the quick fixes.
Sometimes the "Post-processing output" in
inversions_pymc.inferpymc_postprocessouts
fails due to a "Filter error":Steps to recreate
The following
.ini
file was used to run a small inversion on the login and compute notes of Blue Pebble. The obs/ancillary data is from the object store in/group/chemistry/acrg/ES_store
.Attempted fixes
Changing the
encoding
option toNone
into_netcdf
(around line 600 in inversions_pymc.py) seem to fix the problem. This might not be a satisfactory fix since it turns off compression.This issue https://github.com/Unidata/netcdf4-python/issues/1175 suggests that the problem might be caused by the
zlib=True
option being used on "variable length data", but I'm not sure exactly where this occurs ininferpymc_postprocessouts
.