clarify element-errors-ptable-heatmap.svelte normalized checkbox purpose

with more detailed label
also remove abstract from preprint/references.yaml
combine 2023-03-(04|06)-chgnet-0.2.0 preds with 500 and 2000 relax steps into one file (2023-03-06-chgnet-0.2.0-wbm-IS2RE.csv.gz)

Author: janosh

data/wbm/eda.py

@@ -4,7 +4,6 @@
 import numpy as np
 import pandas as pd
 import plotly.express as px
-import plotly.graph_objects as go
 from pymatgen.core import Composition
 from pymatviz import (
     count_elements,
@@ -107,7 +106,7 @@
 fig = px.bar(
     x=bins[bins < 0], y=left_counts, opacity=0.7, labels={"x": x_label, "y": "Counts"}
 )
-fig.add_trace(go.Bar(x=bins[bins >= 0], y=right_counts, opacity=0.7))
+fig.add_bar(x=bins[bins >= 0], y=right_counts, opacity=0.7)
 
 if col.startswith("e_above_hull"):
     n_stable = sum(df_wbm[col] <= STABILITY_THRESHOLD)

matbench_discovery/plots.py

@@ -625,15 +625,14 @@ def rolling_mae_vs_hull_dist(
                     color=plotly_colors[idx], dash=plotly_line_styles[idx], width=3
                 )
             # marker_spacing = 2
-            # trace = go.Scatter(
+            # fig.add_scatter(
             #     x=trace.x[::marker_spacing],
             #     y=trace.y[::marker_spacing],
             #     mode="markers",
             #     marker=dict(symbol=marker, color=trace.line.color, size=8),
             #     showlegend=False,
             #     legendgroup=getattr(trace, "legendgroup", None),
             # )
-            # fig.add_trace(trace)
     return fig, df_rolling_err, df_err_std
 
 

models/chgnet/2023-03-06-chgnet-0.2.0-wbm-IS2RE.csv.gz

@@ -89,19 +89,17 @@ def plot_energy_and_forces(
     """Plot energy and forces as a function of relaxation step."""
     fig = go.Figure()
     # energy trace = primary y-axis
-    fig.add_trace(go.Scatter(x=df.index, y=df[e_col], mode="lines", name="Energy"))
+    fig.add_scatter(x=df.index, y=df[e_col], mode="lines", name="Energy")
     # get energy line color
     line_color = fig.data[0].line.color
 
     # forces trace = secondary y-axis
-    fig.add_trace(
-        go.Scatter(
-            x=df.index,
-            y=df[force_col],
-            mode="lines",
-            name="Forces",
-            yaxis="y2",
-        ),
+    fig.add_scatter(
+        x=df.index,
+        y=df[force_col],
+        mode="lines",
+        name="Forces",
+        yaxis="y2",
     )
 
     fig.update_layout(

models/chgnet/ctk_trajectory_viewer.py

@@ -84,14 +84,13 @@
     large_errors = df_each_err[model].abs().nlargest(n_structs)
     small_errors = df_each_err[model].abs().nsmallest(n_structs)
     for label, errors in zip(("min", "max"), (large_errors, small_errors)):
-        scatter = go.Histogram(
+        fig.add_histogram(
             x=df_wbm.loc[errors.index][fp_diff_col].values,
             name=f"{model} err<sub>{label}</sub>",
             visible="legendonly" if idx else True,
             legendgroup=model,
             hovertemplate=("SSFP diff: %{x:.2f}<br>Count: %{y}"),
         )
-        fig.add_trace(scatter)
 
 title = (
     f"Norm-diff between initial/final SiteStatsFingerprint<br>"
@@ -121,7 +120,7 @@
 for idx, model in enumerate(df_metrics):
     errors = df_each_err[model].abs().nlargest(n_structs)
     model_mae = errors.mean().round(3)
-    scatter = go.Scatter(
+    fig.add_scatter(
         x=df_wbm.loc[errors.index][fp_diff_col].values,
         y=errors.values,
         mode="markers",
@@ -136,7 +135,6 @@
         customdata=df_wbm.loc[errors.index][["material_id", "formula"]].values,
         legendrank=model_mae,
     )
-    fig.add_trace(scatter)
 
 title = (
     f"Norm-diff between initial/final SiteStatsFingerprint<br>"
@@ -246,7 +244,7 @@
     model_mae = df_each_err[model].loc[df_largest_fp_diff.index].abs().mean()
 
     visible = "legendonly" if idx else True
-    scatter = go.Scatter(
+    fig.add_scatter(
         x=df_largest_fp_diff.values,
         y=df_each_err[model].loc[df_largest_fp_diff.index].abs(),
         mode="markers",
@@ -265,7 +263,6 @@
         marker=dict(color=color),
         legendrank=model_mae,
     )
-    fig.add_trace(scatter)
     # add dashed mean line for each model that toggles with the scatter plot
     # fig.add_hline(
     #     y=model_mae,
@@ -357,14 +354,13 @@
     fig = go.Figure()
     for model in df_metrics:
         errors = getattr(df_each_err[model].abs(), which)(n_structs)
-        violin = go.Violin(
+        fig.add_violin(
             x=df_wbm.loc[errors.index][fp_diff_col].values,
             name=f"{model} err<sub>{label}</sub>",
             legendgroup=model,
             hovertemplate=("SSFP diff: %{x:.2f}<br>Count: %{y}"),
             spanmode="hard",
         )
-        fig.add_trace(violin)
     fig.layout.update(showlegend=False)
     fig.layout.xaxis.title = "SSFP norm-diff before/after relaxation"
     fig.show()

scripts/analyze_model_failure_cases.py

@@ -62,8 +62,7 @@
 for idx, model in enumerate(models):
     ys = [df_each_err[model].quantile(quant) for quant in (0.05, 0.25, 0.5, 0.75, 0.95)]
 
-    box_plot = go.Box(y=ys, name=model, width=0.7)
-    fig.add_trace(box_plot)
+    fig.add_box(y=ys, name=model, width=0.7)
 
     # Add an annotation for the interquartile range
     IQR = ys[3] - ys[1]

scripts/model_figs/make_hull_dist_box_plot.py

@@ -5,7 +5,6 @@
 from typing import Final
 
 import numpy as np
-import plotly.graph_objects as go
 from pymatviz.io import save_fig
 
 from matbench_discovery import PDF_FIGS, SITE_FIGS
@@ -65,11 +64,10 @@
     counts, bins = np.histogram(
         df_preds[each_true_col] + noise, bins=100, range=fig.layout.xaxis.range
     )
-    marginal_trace = go.Scatter(
+    fig.add_scatter(
         x=bins, y=counts, name="Density", fill="tozeroy", showlegend=False, yaxis="y2"
     )
-    marginal_trace.marker.color = "rgba(0, 150, 200, 1)"
-    fig.add_trace(marginal_trace)
+    fig.data[-1].marker.color = "rgba(0, 150, 200, 1)"
 
     # update layout to include marginal plot
     fig.layout.update(

scripts/model_figs/rolling_mae_vs_hull_dist_models.py

@@ -46,6 +46,7 @@
       color_scale={color_scale[0]}
       {log}
       bind:active_element={active_wbm_elem}
+      show_photo={false}
     >
       <TableInset slot="inset">
         <label for="log">Log color scale<Toggle id="log" bind:checked={log} /></label>
@@ -81,6 +82,7 @@
       color_scale={color_scale[0]}
       {log}
       bind:active_element={active_mp_elem}
+      show_photo={false}
     >
       <TableInset slot="inset">
         <label for="log">Log color scale<Toggle id="log" bind:checked={log} /></label>

site/src/routes/data/+page.svelte

@@ -71,6 +71,7 @@ structure was generated in).
   {log}
   color_scale={color_scale[0]}
   bind:active_element
+  show_photo={false}
 >
   <TableInset slot="inset">
     <PtableInset element={active_element} elem_counts={active_counts} />

site/src/routes/data/tmi/+page.svelte

@@ -69,9 +69,14 @@
     {cbar_max}
   </label>
   <label>
-    Divide errors by test set energies std. dev. over structures containing each element
+    Divide each element value by its std. dev. of target energies over all test structures
+    containing a given element
     <input type="checkbox" bind:checked={normalized} />
   </label>
+  <small>
+    This is meant to correct for the fact that some elements are inherently more difficult
+    to predict since some have a wider distribution of convex hull distances.
+  </small>
 </form>
 
 <PeriodicTable
@@ -82,6 +87,7 @@
   tile_props={{
     precision: `0.2`,
   }}
+  show_photo={false}
 >
   <TableInset slot="inset" style="align-content: center;">
     <PtableInset
@@ -113,4 +119,9 @@
     place-content: center;
     gap: 1ex;
   }
+  form label + small {
+    max-width: 60em;
+    margin: 0 auto;
+    text-align: center;
+  }
 </style>