Bug: Confusion matrix misinterprets integer class labels as continuous range

[guillaume-vignal]

Description

When using plot_confusion_matrix with integer class labels that are not contiguous (e.g., [1, 2, 10]), the function incorrectly interprets them as a continuous numeric range rather than as categorical labels.

As a result, the confusion matrix displays empty intermediate classes (e.g., 3–9), creating unwanted spacing in the visualization.

Integer labels should be treated as categorical values, not as numeric intervals.

---

📌 Minimal Reproducible Example

from shapash.plots.plot_evaluation_metrics import plot_confusion_matrix

# Example lists of integer class labels
y_true_labels = [1, 2, 1, 10, 2, 1, 10, 2]
y_pred_labels = [1, 1, 1, 10, 2, 2, 10, 1]

plot_confusion_matrix(
    y_true=y_true_labels,
    y_pred=y_pred_labels,
    width=500,
    height=400
)

---

❗ Current Behavior

• The confusion matrix assumes labels are continuous integers.
• Intermediate class values (e.g., 3, 4, 5, 6, 7, 8, 9) are implicitly considered.
• The resulting plot contains large empty gaps between actual labels (2 → 10).

---

✅ Expected Behavior

• Labels should be treated as categorical values.
• Only observed unique labels should appear in the confusion matrix.
• No artificial gaps between 2 and 10.