How do I prevent ConfusionMatrixDisplay.from_predictions from showing the plot?

Question

ConfusionMatrixDisplay docs say:

It is recommend to use from_estimator or from_predictions to create a ConfusionMatrixDisplay.

However, I want to plot the matrix manually on some axes I configure, and when I use from_predictions, I can't prevent it from plotting the matrix. The result is that I get two plots shown: one from the from_predictions call, and one from cmp.plot().

For example, this results in two plots:

import numpy as np
from sklearn.metrics import ConfusionMatrixDisplay, confusion_matrix
import matplotlib.pyplot as plt

cmp = ConfusionMatrixDisplay.from_predictions(np.arange(25), np.arange(25))
fig, ax = plt.subplots(figsize=(10,10))
cmp.plot(ax=ax)

But this results in one:

import numpy as np
from sklearn.metrics import ConfusionMatrixDisplay, confusion_matrix
import matplotlib.pyplot as plt

cm = confusion_matrix(np.arange(25), np.arange(25))
cmp = ConfusionMatrixDisplay(cm, display_labels=np.arange(25))
fig, ax = plt.subplots(figsize=(10,10))
cmp.plot(ax=ax)

I would like to use from_predictions, but to have it not plot, like using the constructor.

Notebook: https://drive.google.com/file/d/1lUPpp0dYfuIk1z8o7QfdUlQTt9eJ1dy4/view?usp=sharing

Answer 1

The from_predictions has an extra keyword argument that you can use to plot the figure on an existing Axes object (see also the documentation, which you can use to replace the cmp.plot(ax=ax) call. The following should therefore do what you're looking for:

fig, ax = plt.subplots(figsize=(10, 10))
cmp = ConfusionMatrixDisplay.from_predictions(np.arange(25), np.arange(25), ax=ax)