Interactive 3-D plots in Jupiter notebook

Question

I have the following code which does produce a 3-D plot in jupyter notebook, but I am unable to get it to be interactive so I can rotate it with the mouse.

import matplotlib.pyplot as plt
from mpl_toolkits.mplot3d import Axes3D
import numpy as np
import pandas as pd

# Generate random dummy data
np.random.seed(42)
X = pd.DataFrame(np.random.randn(100, 3), columns=['feature1', 'feature2', 'feature3'])
VT = np.random.randn(3, 3)
U = np.random.choice([0, 1], size=100)


fig2 = plt.figure()
ax = fig2.add_subplot(111, projection='3d')

for j in range(X.shape[0]):
    x = VT[0, :] @ X.iloc[j, :].T
    y = VT[1, :] @ X.iloc[j, :].T
    z = VT[2, :] @ X.iloc[j, :].T

    if U[j] == 1:
        ax.scatter(x, y, z, marker='x', color='r', s=50)
    else:
        ax.scatter(x, y, z, marker='o', color='b', s=50)

ax.view_init(25, 20)
plt.show()

I have looked into the issue on several sites (Stack Overflow and GitHub), but it is not quite clear what the remedy may be. using %matplotlib qt was suggested, but it throws the exception:

ImportError: Failed to import any of the following Qt binding modules: PyQt6, PySide6, PyQt5, PySide2

Trying to install PyQt5 also failed in my Kaggle environment

Answer 1

If you install ipympl, this variation of your code will work:

%matplotlib ipympl
import matplotlib.pyplot as plt
from mpl_toolkits.mplot3d import Axes3D
import numpy as np
import pandas as pd

# Generate random dummy data
np.random.seed(42)
X_t = pd.DataFrame(np.random.randn(100, 3), columns=['feature1', 'feature2', 'feature3'])
VT = np.random.randn(3, 3)
U = np.random.choice([0, 1], size=100)


fig2 = plt.figure()
ax = fig2.add_subplot(111, projection='3d')

for j in range(X_t.shape[0]):
    x = VT[0, :] @ X_t.iloc[j, :].T
    y = VT[1, :] @ X_t.iloc[j, :].T
    z = VT[2, :] @ X_t.iloc[j, :].T

    if U[j] == 1:
        ax.scatter(x, y, z, marker='x', color='r', s=50)
    else:
        ax.scatter(x, y, z, marker='o', color='b', s=50)

ax.view_init(25, 20)
plt.show()

The ipympl Github repo under the matplotlib organization provides additional concise background on usage.

To demonstrate it without touching your system:
Click here to launch directly into a Jupyter session served via MyBinder. Open a new notebook and paste that code into there and run it. You should be able to left click on the graph and the drag to where you want. Release the left-click when it is where you want. Hit the 'Home' button on the tool widget in the upper left side of the output area to reset the view to the initial point.

That session already had the equivalent of pip install ipympl run it. That is what you are seeking for modern interactive plotting with Jupyter, see here. Note that last link also details that in addition to %matplotlib ipympl, you can use %matplotlib widget, too. I explain more about the inline default of modern Jupyter and setting the backend option towards the top of your notebook in my comments here for a pertinent reply to a related question about enabling the interactive backend.

Related answer about plots is also here.