StochasticAnalytics / emClarity

GNU Lesser General Public License v3.0
41 stars 6 forks source link

simplify classification for release #37

Closed bHimes closed 6 years ago

bHimes commented 6 years ago

Classification is too complex for general use.

1) Remove options for class average alignment from public view, this removes a lot of parameters and commands. It does prevent a multi-ref alignment, but that was not discussed in the paper and likely needs work to be robust anyhow.

2) Make the default the "-1" flag behavior, which looks at even and odd volumes together - since there is no alignment at that point, this is okay and cuts the number of steps in half.

3) When a user makes an imod model to indicate classes for removal, convert that internally instead of requiring user to do model2point.

4) Add BH_skipClassAlignment to emClarity and combine ClassAlignment/RawAlignment steps into one call. Simplify args to just param and cycle number since we have done (2)

5) This means the entire "Class reference and alignment" section of the parameter file can be removed. The exception being Cls_className, and Cls_classes_odd (can drop the odd) specifying symmetry and class idx for averaging.

6) Fsc_refVector_odd/eve would no longer be needed since these are only relevant for a multi-reference alignment.

7) Default Fsc_bfactor to 0 and allow as option in avg FinalAlignment.

8) Move fscGoldSplitOnTomos to advanced options, then the whole "Fsc parameters" section is gone.

9) Add clarification to the Classification parameters section, default flgWMDs to 3 and remove it and flgWedgeComparision from the parameter file.

10) This isn't about classification, but in the advanced options, remove duplicateRadius and duplicate sampling, and set instead to use some fraction of the particle radius, and a binning that gives a pixel size ~ 10 Ang.

bHimes commented 6 years ago

Finished with - 2,3,4

bHimes commented 6 years ago

done