fix ModelCard not handling multiple training sets correctly

plus other minor fixes

Author: janosh

scripts/model_figs/metrics_tables.py

@@ -95,11 +95,14 @@
         )
 
         if n_materials_total != n_structs_total:
-            title = "Number of materials (and structures) in training set"
+            title = (
+                "Number of materials in training set
In parenthesis=number of "
+                "structures, usually from multiple DFT relaxation frames per material"
+            )
             train_size_str = (
                 f"<span {title=} data-sort-value={n_materials_total}>"
                 f"{si_fmt(n_materials_total, fmt='.0f')}</span>"
-                f"<small {title=}> ({si_fmt(n_structs_total, fmt='.1f')})</small>"
+                f" <small {title=}> ({si_fmt(n_structs_total, fmt='.1f')})</small>"
             )
 
         dataset_links = []
@@ -112,7 +115,7 @@
                 dataset_links += [key]
         dataset_str = "+".join(dataset_links)
 
-        tooltip = "\\n".join(tooltip_lines)
+        tooltip = "&#013;".join(tooltip_lines)
         train_size_str = (
             f"<span title='{tooltip}'>{train_size_str} ({dataset_str})</span>"
         )

scripts/model_figs/rolling_mae_vs_hull_dist_models.py

@@ -44,6 +44,7 @@
     df_rolling_err=df_err,
     df_err_std=df_std,
     show_dummy_mae=False,
+    height=600,
 )
 
 if backend == MATPLOTLIB:
@@ -55,7 +56,7 @@
     for line in fig.lines:
         line._linewidth *= 2  # noqa: SLF001
 else:
-    show_n_best_models = len(models)
+    show_n_best_models = 8
     for trace in fig.data:
         model = trace.name.split(" MAE=")[0]
         if model in df_metrics.T.sort_values("MAE").index[show_n_best_models:]:

site/src/figs/each-parity-models-7x2.svelte

@@ -1,6 +1,7 @@
 <script lang="ts">
   import type { ModelData, ModelStatLabel } from '$lib'
   import { AuthorBrief } from '$lib'
+  import TRAINING_SETS from '$root/data/training-sets.yml'
   import { repository } from '$site/package.json'
   import Icon from '@iconify/svelte'
   import { pretty_num } from 'elementari'
@@ -91,17 +92,27 @@
     {/if}
   </span>
   {#if training_set}
-    {@const { n_structures, url, title, n_materials } = training_set}
-    {@const pretty_n_mat =
-      typeof n_materials == `number` ? pretty_num(n_materials) : n_materials}
-    {@const n_mat_str = n_materials ? ` from ${pretty_n_mat} materials` : ``}
+    {@const training_sets = Array.isArray(training_set) ? training_set : [training_set]}
     <span style="grid-column: span 2;">
       <Icon icon="mdi:database" inline />
       Training set:
-      <a href={url}>{title}</a>
-      <small>
-        ({pretty_num(n_structures)} structures{n_mat_str})
-      </small>
+      {#each training_sets as training_set_or_key, idx}
+        {#if idx > 0}
+          &nbsp;+&nbsp;
+        {/if}
+        {@const training_set =
+          typeof training_set_or_key == `string`
+            ? TRAINING_SETS[training_set_or_key]
+            : training_set_or_key}
+        {@const { n_structures, url, title, n_materials } = training_set}
+        {@const pretty_n_mat =
+          typeof n_materials == `number` ? pretty_num(n_materials) : n_materials}
+        {@const n_mat_str = n_materials ? ` from ${pretty_n_mat} materials` : ``}
+        <a href={url}>{title}</a>
+        <small>
+          ({pretty_num(n_structures)} structures{n_mat_str})
+        </small>
+      {/each}
     </span>
   {/if}
 </p>
@@ -127,7 +138,10 @@
         </ul>
       {/if}
     </section>
-    <section transition:slide={{ duration: 200 }}>
+    <section
+      transition:slide={{ duration: 200 }}
+      style="overflow: hidden; white-space: nowrap; text-overflow: ellipsis;"
+    >
       <h3>Package versions</h3>
       <ul>
         {#each Object.entries(data.requirements ?? {}) as [name, version]}

site/src/figs/each-parity-models-8x2.svelte

@@ -5,9 +5,9 @@
   import BoxHullDistErrors from '$figs/box-hull-dist-errors.svelte'
   import CumulativeMae from '$figs/cumulative-mae.svelte'
   import CumulativePrecisionRecall from '$figs/cumulative-precision-recall.svelte'
-  import EachParityModels from '$figs/each-parity-models-7x2.svelte'
+  import EachParityModels from '$figs/each-parity-models-8x2.svelte'
   import ElementPrevalenceVsError from '$figs/element-prevalence-vs-error.svelte'
-  import HistClfPredHullDistModels from '$figs/hist-clf-pred-hull-dist-models-7x2.svelte'
+  import HistClfPredHullDistModels from '$figs/hist-clf-pred-hull-dist-models-8x2.svelte'
   import HullDistParityWrenformerFailures from '$figs/hull-dist-parity-wrenformer-failures.svelte'
   import LargestErrorScatterSelect from './largest-error-scatter-select.svelte'
   import MetricsTableUniqProtos from '$figs/metrics-table-uniq-protos.svelte'