🪲 #125 fix quarto NB problem

autopeptideml/utils/plots.py

@@ -196,9 +196,8 @@ def plot_model_vs_rep(history: pd.DataFrame, ax: plt.Axes = None,
                       figsize: Tuple[int, int] = (5, 5)):
 
     if ax is None:
-        fig = start_figure(
-            figsize[0]*len(history.run.unique())//20, figsize[1]
-        )
+        width = max(figsize[0], figsize[0]*len(history.run.unique())//20)
+        fig = start_figure(width, figsize[1])
         ax = fig.subplots(1, 1)
 
     history['Representation'] = history['representation']

tests/test_plots.py

@@ -0,0 +1,78 @@
+"""
+Tests for autopeptideml.utils.plots — specifically guarding against the
+zero-width-figure crash in plot_model_vs_rep when the HPO history contains
+fewer than 20 unique runs (which happens routinely with one representation
+and aggressive early-stopping).
+"""
+import matplotlib
+matplotlib.use('Agg')  # non-interactive backend — must come before pyplot import
+import matplotlib.pyplot as plt
+import numpy as np
+import pandas as pd
+import pytest
+
+from autopeptideml.utils.plots import plot_model_vs_rep, plot_optimization_history
+
+
+def _make_history(n_runs: int, reps: list, n_folds: int = 3) -> pd.DataFrame:
+    """Build a minimal HPO history DataFrame with the same schema as the real pipeline."""
+    rng = np.random.default_rng(0)
+    rows = []
+    for run in range(1, n_runs + 1):
+        rep = reps[(run - 1) % len(reps)]
+        for fold in range(n_folds):
+            rows.append({
+                'run': run,
+                'fold': fold,
+                'name': 'svm',
+                'representation': rep,
+                'variables': str({'C': 1.0}),
+                'mcc': float(rng.uniform(0.3, 0.8)),
+                'acc': float(rng.uniform(0.5, 0.9)),
+            })
+    return pd.DataFrame(rows)
+
+
+# ---------------------------------------------------------------------------
+# Bug regression: plot_model_vs_rep must not crash when n_runs < 20
+# (previously caused LinAlgError: Singular matrix via a zero-width figure)
+# ---------------------------------------------------------------------------
+
+@pytest.mark.parametrize("n_runs", [1, 2, 5, 10, 19])
+def test_plot_model_vs_rep_few_runs_one_rep(n_runs):
+    """plot_model_vs_rep must succeed with fewer than 20 unique runs, one rep."""
+    history = _make_history(n_runs=n_runs, reps=['one-hot'])
+    try:
+        plot_model_vs_rep(history=history, metric='mcc')
+    finally:
+        plt.close('all')
+
+
+@pytest.mark.parametrize("n_runs", [1, 2, 5, 10, 19])
+def test_plot_model_vs_rep_few_runs_two_reps(n_runs):
+    """plot_model_vs_rep must succeed with fewer than 20 unique runs, two reps."""
+    history = _make_history(n_runs=n_runs, reps=['one-hot', 'ecfp-16'])
+    try:
+        plot_model_vs_rep(history=history, metric='mcc')
+    finally:
+        plt.close('all')
+
+
+def test_plot_model_vs_rep_mutates_history():
+    """plot_model_vs_rep adds Representation/Model/Run columns so the QMD can drop them."""
+    history = _make_history(n_runs=5, reps=['one-hot'])
+    assert 'Representation' not in history.columns
+    plot_model_vs_rep(history=history, metric='mcc')
+    plt.close('all')
+    assert 'Representation' in history.columns
+    assert 'Model' in history.columns
+    assert 'Run' in history.columns
+
+
+def test_plot_optimization_history_few_runs():
+    """plot_optimization_history must succeed with very few runs."""
+    history = _make_history(n_runs=2, reps=['one-hot'])
+    try:
+        plot_optimization_history(history=history, metric='mcc')
+    finally:
+        plt.close('all')