Phase losing name and properties after refinement from single crystal map

[Erlendos12]

Hi! (sorry for the long post)
I have a single crystal map (.h5) which contains two phases of austenite and ferrite.

>>>print(xmap)
Phase   Orientations       Name  Space group  Point group  Proper point group  Color
    0    747 (37.4%)    ferrite        Im-3m         m-3m                 432      r
    1   1253 (62.6%)  austenite        Fm-3m         m-3m                 432   aqua

I would like to refine the crystal map's orientations by using refine orientation on each crystal map phase with the appropriate master pattern. I load the master pattern, make a deep copy of the crystal map phase with mp_key (name of the phase) with a for loop:

for mp_key, mp_path in self.mp_paths.items():
    mp: LazyEBSDMasterPattern = kp.load(
        mp_path,
        energy=energy,
        projection="lambert",
        hemisphere="upper",
        lazy=lazy,
    )
    print(f"\nRefining Phase: {mp.phase.name}")
    xmap_phase = xmap[mp_key].deepcopy()
    print(xmap_phase)

#Outputs (as expected)
Refining Phase: ferrite
Phase  Orientations     Name  Space group  Point group  Proper point group  Color
    0  747 (100.0%)  ferrite        Im-3m         m-3m                 432      r
Refining Phase: austenite
Phase   Orientations       Name  Space group  Point group  Proper point group  Color
    1  1253 (100.0%)  austenite        Fm-3m         m-3m                 432   aqua

I then perform refine_orientations, but now the output of the second phase has changed and are missing everything except orientations.

ref_xmaps[mp_key] = s.refine_orientation(
    xmap=xmap_phase,
    detector=det,
    master_pattern=mp,
    energy=energy,
    signal_mask=signal_mask,
    trust_region=[1, 1, 1],
    method=method,
    method_kwargs=ref_kwargs,
    compute=True
)
print(f"After refinement:\n{ref_xmaps[mp_key]}")

# Outputs
After refinement:
Phase  Orientations     Name  Space group  Point group  Proper point group     Color
    0  747 (100.0%)  ferrite        Im-3m         m-3m                 432  tab:blue
After refinement:
Phase   Orientations  Name  Space group  Point group  Proper point group       Color
    1  1253 (100.0%)  None         None         None                None  tab:orange

Am I doing something wrong when doing refinement this way? I know that the name and properties are kept when the phases are loaded from two different crystal maps instead of just the one containging both phases. Note that the signal s is the pattern which the crystal map originated from, but is not being indexed (neither pattern matching nor hough indexing) in the same code where refinement takes place. Colors also change after refinement, but these are maybe decided by the master pattern rather than the original crystal map?

Full code snippet with output at bottom:

...
ref_xmaps = {}
print(xmap)
for mp_key, mp_path in self.mp_paths.items():
    mp: LazyEBSDMasterPattern = kp.load(
        mp_path,
        energy=energy,
        projection="lambert",
        hemisphere="upper",
        lazy=lazy,
    )
    print(f"\nRefining Phase: {mp.phase.name}")
    xmap_phase = xmap[mp_key].deepcopy()
    print(xmap_phase)
    ref_xmaps[mp_key] = s.refine_orientation(
        xmap=xmap_phase,
        detector=det,
        master_pattern=mp,
        energy=energy,
        signal_mask=signal_mask,
        trust_region=[1, 1, 1],
        method=method,
        method_kwargs=ref_kwargs,
        compute=True
    )
    print(f"After refinement:\n{ref_xmaps[mp_key]}")
...

# Output
Phase   Orientations       Name  Space group  Point group  Proper point group  Color
    0    747 (37.4%)    ferrite        Im-3m         m-3m                 432      r
    1   1253 (62.6%)  austenite        Fm-3m         m-3m                 432   aqua
Properties: cm, fit, nmatch, pq
Scan unit: um

Refining Phase: ferrite
Phase  Orientations     Name  Space group  Point group  Proper point group  Color
    0  747 (100.0%)  ferrite        Im-3m         m-3m                 432      r
Properties: cm, fit, nmatch, pq
Scan unit: um
Refinement information:
  Method: LN_NELDERMEAD (local) from NLopt
  Trust region (+/-): [1 1 1]
  Relative tolerance: 0.0001
Refining 747 orientation(s):

[########################################] | 100% Completed | 7.82 s
Refinement speed: 95.23195 patterns/s
After refinement:
Phase  Orientations     Name  Space group  Point group  Proper point group     Color
    0  747 (100.0%)  ferrite        Im-3m         m-3m                 432  tab:blue
Properties: scores, num_evals
Scan unit: um

Refining Phase: austenite
Phase   Orientations       Name  Space group  Point group  Proper point group  Color
    1  1253 (100.0%)  austenite        Fm-3m         m-3m                 432   aqua
Properties: cm, fit, nmatch, pq
Scan unit: um
Refinement information:
  Method: LN_NELDERMEAD (local) from NLopt
  Trust region (+/-): [1 1 1]
  Relative tolerance: 0.0001
Refining 1253 orientation(s):

[########################################] | 100% Completed | 13.45 s
Refinement speed: 92.94377 patterns/s
After refinement:
Phase   Orientations  Name  Space group  Point group  Proper point group       Color
    1  1253 (100.0%)  None         None         None                None  tab:orange
Properties: scores, num_evals
Scan unit: um

[hakonanes]

Hi @Erlendos12,

thank you for asking. The loss of properties is by design. There are some auxiliary properties, like the orientation similarity (OS) computed from dictionary indexing results, which aren't valid for refined maps. Therefore, I think it is safest to return only properties we know are valid for the refined map. If we want to add them to the refined map, we have to do so manually.

The "intended way" of refining only some points in a map is to pass a navigation_mask, with points to refine equal to False (points to "mask out" are True), to the refinement methods EBSD.refine_*(). This procedure is demonstrated in the Hybrid indexing tutorial. It seems to me like this workflow does what you want. Could you try it out?

That said, the loss of phase information is strange and seems like a bug. Can you check xmap.phases_in_data? I'm wondering if the Austenite phase got a phase ID of 0 when wrapping the refinement results in a new crystal map, while the phase_id array keeps the phase ID of 1...

[hakonanes]

Great! Thank you for doing the test.

Will let you know once orix 0.11.1 is available so that you can use this version instead of my branch.

[hakonanes]

We just released orix 0.11.1, so kikuchipy should set this as the minimal version in a patch release 0.8.2 to avoid people encountering this issue. Can you make this change, @Erlendos12? The changes required are:

• Update minimal orix version in setup.py
• Add a one-line note in the changelog under "Changed" stating this change

Since this is a patch, the changes should go directly to main, so remember to branch off of main and not develop when making the fix-branch.

[Erlendos12]

Yes, that I can do. Glad to be of help.

[hakonanes]

Great! Let me know if you're unsure of anything.

[hakonanes]

You can now depend on kikuchipy v0.8.2 in your GUI. Keep the good issue reports coming! (I hope to look at #622 during the weekend.)