Phonons analysis app #223

joehart2001 · 2025-12-17T15:46:10Z

Pre-review checklist for PR author

PR author must check the checkboxes below when creating the PR.

I've confirmed the contribution guidelines.

Summary

phonon analysis and app, without the calc for the moment.

Linked issue

Resolves #222

Progress

Calculations
Analysis
Application
Documentation

Testing

mattersim and mace-omat-0

New decorators/callbacks

decorators:

cell_to_scatter: table cell -> scatter plot

callbacks:

for model-specific assets e.g. phonon dispersions. for weas, structures are model independent, meaning we need these new callbacks to propagate model specific info.
- scatter_and_assets_from_table
- model_asset_from_scatter

i think theres potential to combine these with other callbacks so lets dicuss? e.g. make table -> scatter and scatter to structure more general (e.g. scatter -> any asset). but maybe we want to keep it separate so its more simple for users

ml_peg/analysis/bulk_crystal/phonons/metrics.yml

ml_peg/app/bulk_crystal/phonons/app_phonons.py

ml_peg/analysis/bulk_crystal/phonons/analyse_phonons.py

ElliottKasoar · 2025-12-18T16:23:37Z

ml_peg/analysis/bulk_crystal/phonons/analyse_phonons.py

+    cumulative_dist, i = 0.0, 0
+    connections = [True] + connections
+
+    for seg_dist, connected in zip(distances, connections, strict=False):


With strict=False this will stop when either of the two lists ends, even if the other is longer. Is this what you want?

ml_peg/analysis/bulk_crystal/phonons/analyse_phonons.py

ml_peg/app/utils/build_callbacks.py

ElliottKasoar · 2025-12-18T17:22:09Z

ml_peg/app/utils/build_callbacks.py

+        return Div("Click on a metric to view the structure.")

+
+def scatter_and_assets_from_table(


Why can't this by handed by something like plot_from_table_cell?

it also keeps the model data so that we can access model-specific assets from the scatter plot. (with weas, we dont take model-specific assets as the structures are all the same (right?))

ElliottKasoar · 2025-12-18T17:28:49Z

ml_peg/app/utils/build_callbacks.py

+        return content, meta, active_cell
+
+
+def model_asset_from_scatter(


Where does the model-specific part come in?

As above, I also wonder if asset it the most useful description? Is this actually loading from /assets?

The model context is carried via scatter_metadata: when the user clicks a table cell, scatter_and_assets_from_table stores the active model, metric, and the list of points.

asset is just a general way of saying "secondary view component" e.g. scatter -> asset e.g. scatter -> structure or png or other. i was trying to make a general. (wouldnt use for strucutures here as model independent, but maybe if we had specific relaxed structures which were model dependent then we could use this for that too)

ml_peg/app/bulk_crystal/phonons/app_phonons.py

ml_peg/app/bulk_crystal/phonons/interactive_helpers.py

ml_peg/analysis/utils/decorators.py

ElliottKasoar · 2025-12-18T18:16:31Z

ml_peg/analysis/utils/decorators.py

    return plot_parity_decorator


+def cell_to_scatter(


Would it make sense to make this an option for the exist scatter decorator, rather than a separate one? I think basically all the logic is the same, it's just whether we combine the traces, right? We may even want to do individual + combined increasingly e.g. when scaling starts to get messy

ml_peg/app/bulk_crystal/phonons/interactive_helpers.py

ElliottKasoar · 2025-12-18T19:41:27Z

ml_peg/app/utils/plot_helpers.py

Why do we build these as part of the app, rather than saving them to json and loading them as we normally do?

save on analysis time

ElliottKasoar · 2025-12-18T19:46:51Z

ml_peg/app/utils/plot_helpers.py

+    Parameters
+    ----------
+    points
+        Sequence of metadata dictionaries containing reference/prediction values.


This is a bit confusing because it makes it sound like data rather than metadata, but I think it's because we save various things (including but not limited to the x/y values) of each point in these dicts? Probably worth a making the connection to "points" in this docstring too, since out of context it's quite confusing

… on the fly)

…, generalise non-specific helpers to plot_helpers.py and clean up

Co-authored-by: Elliott Kasoar <[email protected]>

joehart2001 · 2026-01-12T11:15:36Z

Issue to fix: when rows are reordered, the interactive cells show the original model in that position

Co-authored-by: Elliott Kasoar <[email protected]>

joehart2001 · 2026-01-14T21:08:37Z

So i've gotten to the bottom of the band mismatches.

In the calculation, when the primitive cell isnt present in the ref data, i use primitive cell "auto", which works in most cases, but sometimes the auto algorithm predicts a different number of atoms (so multiple cells).
This means there are more atoms and as no. branches = 3N (N = atoms) then we get a mismatch in the number of branches for the BZ MAE (but not the k-points as this is based off the ref always).
for mp-0a it was 38 samples which got this error so weren't included in the BZ MAE calculation, but i dont think this effects any other metrics.

ElliottKasoar · 2026-01-15T11:13:29Z

.gitignore

 __pycache__/
-ml_peg/app/data/
+ml_peg/app/data/*
+!ml_peg/app/data/onboarding/


I know it's a tiny change, but can we do this separately e.g. its own PR? It's unrelated to phonons

joehart2001 · 2026-01-15T15:44:44Z

So i've gotten to the bottom of the band mismatches.

In the calculation, when the primitive cell isnt present in the ref data, i use primitive cell "auto", which works in most cases, but sometimes the auto algorithm predicts a different number of atoms (so multiple cells).

This means there are more atoms and as no. branches = 3N (N = atoms) then we get a mismatch in the number of branches for the BZ MAE (but not the k-points as this is based off the ref always).

for mp-0a it was 38 samples which got this error so weren't included in the BZ MAE calculation, but i dont think this effects any other metrics.

So a fix that works for some test samples is instead of auto, just calculating the primitive matrix:

                unitcell = phonons_pred.unitcell
                primitive_cell = phonons_pred.primitive
                primitive_matrix = np.linalg.inv(np.array(unitcell.get_cell())) @ np.array(primitive_cell.get_cell())

We probably need to rerun some phonons, but using these larger cells we used before doenst effect the max or min freqs etc, mainly the BZ MAE

joehart2001 added the new benchmark Proposals and suggestions for new benchmarks label Dec 17, 2025

joehart2001 requested a review from ElliottKasoar December 17, 2025 15:46

ElliottKasoar reviewed Dec 18, 2025

View reviewed changes

ml_peg/analysis/bulk_crystal/phonons/analyse_phonons.py Outdated Show resolved Hide resolved

ElliottKasoar reviewed Dec 18, 2025

View reviewed changes

joehart2001 and others added 17 commits January 5, 2026 15:46

add analysis and update gitignore with npz

945e73f

phonon app and callbacks for scatter plot and png

fe0d4a0

fix phonon dispersion plotting and BZ MAE calculation

8254c6a

edit CM, metrics, convert to kelvin

ab19881

switch CM axes

b3dfa75

cell to scatter decorator

dd6e778

fix metric units, load saved scatters instead of on the fly (only png…

4fc70a8

… on the fly)

generalise phonon plotting decorators + streamline

95b6af5

clean up phonon-specfici helpers from app into interactive_helpers.py…

e8de636

…, generalise non-specific helpers to plot_helpers.py and clean up

rename functions and add clear up doc strings

9dc0aba

clean up phonon analysis

a0e8a5d

update metrics.yml

dfab50c

Update ml_peg/analysis/bulk_crystal/phonons/metrics.yml

c9a835f

Co-authored-by: Elliott Kasoar <[email protected]>

Update ml_peg/app/bulk_crystal/phonons/app_phonons.py

edf0e0c

Co-authored-by: Elliott Kasoar <[email protected]>

update metrics.yml

f87b557

use ref path, reuse CM logic, sort strings correctly, fix metrics

e488f65

Apply pre-commit

7e2117a

ElliottKasoar force-pushed the phonons-analysis-app branch from 831bf6e to 7e2117a Compare January 5, 2026 16:55

joehart2001 and others added 5 commits January 13, 2026 15:05

Add tutorial (#221)

d221e1f

Co-authored-by: Elliott Kasoar <[email protected]>

Add PET-MAD model (#244)

882c9cc

change store id name and add comments

240d925

move data load into register callbacks

92cd50a

clear up assets and meta meaning, change to metadata, address comments

c8884dd

joehart2001 added 2 commits January 14, 2026 17:43

improve lookup function

7870ef1

clarify points docstring

61ba825

ElliottKasoar reviewed Jan 15, 2026

View reviewed changes

add comments and print statements

837e734

		return Div("Click on a metric to view the structure.")


		def scatter_and_assets_from_table(

		return content, meta, active_cell


		def model_asset_from_scatter(

Phonons analysis app #223

Are you sure you want to change the base?

Phonons analysis app #223

Uh oh!

Conversation

joehart2001 commented Dec 17, 2025

Pre-review checklist for PR author

Summary

Linked issue

Progress

Testing

New decorators/callbacks

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

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!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

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!

joehart2001 commented Jan 12, 2026

Uh oh!

joehart2001 commented Jan 14, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

joehart2001 commented Jan 15, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants