Closed mogrodnik closed 2 years ago
I was expecting that user gold CI will be failed after these changes. One can see from build log that highest differences appear in these datafields, which contain the spectral cutoffs or are close to them, as the manual (forced) transfer was performed mainly in case of failure in detecting cutoff or predicting bin deactivation. I will now double-check to make sure that these news results are correct.
It looks like we need to upgrade gdf_distance and add an autodetection of vectors od data. Things such as cree%02d
or mag_field_[xyz]
should not be compared component by component like scalars because it may generate huge distance when the difference occurs on irrelevant component.
Fixes problem of broken spectrum, caused by forced manual transfers of 'n' and 'e', disrupting its distribution in the bins across the grid. The problem manifests itself by gradual formation of sharp breaks in the spectrum, which form in poorly populated parts of the grid. These sharp breaks may persist throughout the whole simulation. Results obtained using CRESP module prior to these changes were affected.