How to plot a confusion matrix with scikit-learn

A confusion matrix turns held-out classification predictions into a class-by-class count table. In scikit-learn, plotting the matrix makes label mix-ups visible so a classifier review does not depend on accuracy alone.

The ConfusionMatrixDisplay helper can draw the matrix from a fitted estimator or from already computed predictions. Using from_predictions() fits evaluation scripts where y_test and y_pred already exist, and it avoids asking the estimator to predict again.

Keep the label order consistent between the printed matrix, display_labels, and any classification report. Rows represent true labels and columns represent predicted labels, so diagonal cells are correct classifications and off-diagonal cells show the mistakes to review.

1. Create confusion_matrix_plot_demo.py with held-out labels, predictions, and a noninteractive Matplotlib backend.

   confusion_matrix_plot_demo.py

   from pathlib import Path
    
   import matplotlib
    
   matplotlib.use("Agg")
    
   import matplotlib.pyplot as plt
   import sklearn
   from sklearn.datasets import load_wine
   from sklearn.metrics import ConfusionMatrixDisplay, confusion_matrix
   from sklearn.model_selection import train_test_split
   from sklearn.tree import DecisionTreeClassifier
    
    
   wine = load_wine()
   X_train, X_test, y_train, y_test = train_test_split(
       wine.data,
       wine.target,
       test_size=0.30,
       stratify=wine.target,
       random_state=7,
   )
    
   model = DecisionTreeClassifier(max_depth=3, random_state=7)
   model.fit(X_train, y_train)
    
   y_pred = model.predict(X_test)
   matrix = confusion_matrix(y_test, y_pred)
   output_path = Path("confusion-matrix-display.png")
    
   fig, ax = plt.subplots(figsize=(6, 5))
   ConfusionMatrixDisplay.from_predictions(
       y_test,
       y_pred,
       display_labels=wine.target_names,
       values_format="d",
       cmap="Blues",
       colorbar=False,
       ax=ax,
   )
   ax.set_title("Wine classifier confusion matrix")
   fig.tight_layout()
   fig.savefig(output_path, dpi=160, bbox_inches="tight")
    
   print(f"scikit-learn {sklearn.__version__}")
   print("labels: " + ", ".join(wine.target_names))
   print("confusion matrix:")
   print(matrix)
   print(f"saved plot: {output_path}")

   matplotlib.use(“Agg”) saves the plot from a shell, container, or job runner that does not have an interactive display.

2. Run the plotting script.

   $ python3 confusion_matrix_plot_demo.py
   scikit-learn 1.9.0
   labels: class_0, class_1, class_2
   confusion matrix:
   [[15  3  0]
    [ 0 20  1]
    [ 0  0 15]]
   saved plot: confusion-matrix-display.png

3. Check the off-diagonal counts before accepting the plot.

   The class_0 row shows 3 samples predicted as class_1, and the class_1 row shows 1 sample predicted as class_2. The diagonal cells show 15, 20, and 15 correct predictions.

4. Open the saved PNG and confirm that the labels and counts match the printed matrix.

   Fix display_labels or pass an explicit labels order before sharing the plot if the class names do not match the matrix order.