Timedependens by myksa · Pull Request #11 · mpes-kit/fuller

myksa · 2025-11-03T16:02:09Z

examples 01,02,03 and 05 no real changes only used a different venv and python version
example 04: 1. used 10 copyes of the old data to have a time dependent example.
2. added time constant
3. added data preprocessing steps
4. added time constant for plotting
5. changed the epochs from 150 to 100
For the mrfRec file:

Line 25 to 70: added time data and changed the Intensity data to 4d input
Line 86 to 93: changed Initialize band structure form 2D (kx,ky) to 3d (kx,ky,t)
Line 121 and 131: added time data to fromFile
Line 223: in initializeBand the theoretical data is given, the data is given as a 2D array, for the reconstruction the data is expendet into the time domain
Line 238 to 240: smooting of the intensity is done for all times seperatly
Line 273 to 281: normalization is done for all times seperatly
Line 314 to 338: symmetrize for each time seperatly
Line 364 to 395: for the seq reconstruction i removed the curv. functions and added the time domain. The seq is still work in procress.
Line 405 to 430: the squarediff is calculated for the 3D MRF not the 2D case. LogP is in this case a cube which get initilised with a new function called _initCube(2) which can be found at the end of the code. LogP is then calculated for the time dependent case.
Line 437 to 445: the total LogP is calculated for the cube (3D case) instead of the square (2D) case.
Line 450 and 457: the black and white nodes are calculated for the 3D case.
Line 492 to 513: added the time data and changed the calculations for the time dependent case. I gets transposed to have the needed format of (kx,ky,t,E)
Line 517 to 537: added the time data and changed to calculation of the nodes to the 3D case.
Line 539 to 545: added time data for the results.
Line 764 to 783: calculate logP for timedepentend data
Line 791 to 985: changed the plot functions to time dependent data
1069 to 1082: _initCube(2) funtion for the 3D MRF model

rettigl

Generally, most changes make sense to me, but carefully check the order of your indices. I think the bug could be there.

rettigl · 2025-11-05T12:12:19Z

fuller/mrfRec.py

+        Delay: 1D array | None
+            Delay along time axis as numpy array.
+        I: 4D array | None
+            Measured intensity wrt momentum, time and energy, generated if None.


Do you consider the data as I(kx, ky, t, E) or I(kx, ky, E, t)?

right now i only consider I(kx, ky, E, t), i will change the Description. The format I(kx, ky, E, t) comes from the nexus data set.

rettigl · 2025-11-05T12:14:33Z

fuller/mrfRec.py

        ind1d = np.argmin(np.abs(EE - EE0), 1)
        self.indEb = ind1d.reshape(self.E0.shape)
+        self.indEb = np.repeat(self.indEb[ :, :, np.newaxis], self.lengthdelay, axis=2)
+        # depends on the shape of E0, i assume E0 is the same for all times


Describe what you assume in the docstring, and possibly test for the shape

rettigl · 2025-11-05T12:15:03Z

fuller/mrfRec.py


-        self.I = ndimage.gaussian_filter(self.I, sigma=sigma)
+        for t in range(len(self.delay)):
+            self.I[:,:,:,t] = ndimage.gaussian_filter(self.I[:,:,:,t], sigma=sigma)


So this also assumes I(kx, ky, E, t)

rettigl · 2025-11-05T12:16:05Z

fuller/mrfRec.py

+            for t in range(self.lengthdelay):
+                self.I[:,:,:,t] = mclahe(
+                    self.I[:,:,:,t] ,
+                    kernel_size=kernel_size,
+                    n_bins=n_bins,
+                    clip_limit=clip_limit,
+                    use_gpu=use_gpu,
+                )


we should discuss if this makes sense. Maybe generate one set of parameters, and apply everywhere. Not sure if mclahe can do this rn.

I think i already tried to do this for the 4d intensity. If i remenber correctly i got an error, but i can try again.

rettigl · 2025-11-05T12:38:42Z

fuller/mrfRec.py

-                        logP -= (ECurv[self.indEb[indx, indy - 2]] - 2 * ECurv[self.indEb[indx, indy - 1]] + ECurv) ** 2
-                    if indy < (self.lengthKy - 1):
-                        logP -= (ECurv[self.indEb[indx, indy - 1]] - 2 * ECurv + ECurv[self.indEb[indx, indy + 1]]) ** 2
+                logP -= (ENN - ENN[self.indEb[indx, indy - 1, indt]]) ** 2


The way this is defined here, it will take the same eta along all directions. This should be extended after testing to provide a len(3) eta array for the three dimensions.

I did the extention and i am testing the results right now. If everything looks correct i will update this with the testing of len(3).

rettigl · 2025-11-05T15:41:23Z

fuller/mrfRec.py

+            logP -= np.sum(
+                (Eb[ 0 : (self.lengthKx - 1), :,:] - Eb[1 : self.lengthKx, :,:]) ** 2
+            ) / (2 * self.eta**2)


different etas

I will update this when i find out were this bug with the "ripples" comes from.

rettigl · 2025-11-05T15:43:23Z

fuller/mrfRec.py


        **Parameters**\n
-        kx, ky: 1D array, 1D array | None, None
+        kx, ky, time: 1D array, 1D array | None, None


They are indices and not 1d arrays, no?

Yes, i will change this.

rettigl · 2025-11-05T15:44:25Z

fuller/mrfRec.py

@@ -846,14 +934,19 @@ def plotBands(self, surfPlot=False, cmapName="viridis", figsize=[9, 9], equal_ax
        equal_axes: bool | False


Add parameter

rettigl · 2025-11-05T15:45:12Z

fuller/mrfRec.py

+        if time is None:
+            t = 0
+        if time is not None:
+            t = time


I suppose this should be a time rather than an index like for the function above.

You can use one variable

rettigl · 2025-11-05T15:49:38Z

fuller/mrfRec.py

-            ax.plot_surface(x, y, np.transpose(self.getEb()))
+            ax.plot_surface(x, y, np.transpose(Eb[:,:,t]))


does it matter if you use getEb or just Eb?

I did matter in the beginning while i updatet the code for time dependens. I can check again if this is still needed.

myksa added 4 commits October 31, 2025 17:23

Update for time dependend data

8683f93

Cleared example outputs

6eafa6d

Fixing small formation errors

a1f029b

removed mpes

ed31c14

rettigl reviewed Nov 5, 2025

View reviewed changes

		@@ -846,14 +934,19 @@ def plotBands(self, surfPlot=False, cmapName="viridis", figsize=[9, 9], equal_ax
		equal_axes: bool \| False

		ax.plot_surface(x, y, np.transpose(self.getEb()))
		ax.plot_surface(x, y, np.transpose(Eb[:,:,t]))

Conversation

myksa commented Nov 3, 2025

Uh oh!

rettigl left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants