Zero variance with pyemu.Cov.from_parameter_data despite valid parbnds? about pyemu HOT 4 CLOSED

zip it... duh

pcf_8015_1.zip

from pyemu.

Looks like this is result of applying scale and offset the bounds before taking the log transform - in this case, that causes np.abs(parubnd) == np.abs(parlbnd). I remember this happening before and there an option to help: you can pass scale_offset=False to Cov .from_parameter_data() not apply scale and offset. I don't have a good answer as to conceptually whether these should or shouldn't be applied - any thoughts?

If you are accessing Cov.from_parameter_data() indirectly through helper function, you might not have access to scale_offset - let me know if we need to weave it through...

from pyemu.

Confirmed that offset was the problem. Switched these pars to no transformation. Cov.from_parameter_data() works fine now.

Correct, geostatistical_prior_builder does not see scale_offset.

from pyemu.

I've added a more verbose error message, suggesting to try changing scale_offset.

Also, I'm planning to refactor most of the higher-level draw methods to support the newly created from_mixed_draws and I'll add kwargs to all of these higher level draws...

from pyemu.