Inconsistent behavior under presence of variable with zero-only coefficients

[hdolfen]

When plotting two-dimensional polynomials, I noticed that when a variable has only 0 coefficients it cannot be evaluated. This can result in an error or unexpected behavior.

To Reproduce
Running the example results in the same evaluation of the second and third polynomial in the array. These are polynomials of degree 1, and they should be oriented differently.

import chaospy
import numpy as np
import matplotlib.pyplot as plt

X = 1  # m
Y = 1  # degrees

# Generate distribution
distribution = chaospy.J(chaospy.Uniform(0, X), chaospy.Uniform(0, Y))

x, y = np.linspace((0, 0), (X, Y), 100).T
x_grid, y_grid = np.meshgrid(x, y)

polynomial = chaospy.generate_expansion(2, distribution)

q0, q1 = chaospy.variable(2)

# polynomial = polynomial + (q0 + q1) * 1e-100

print(polynomial)

for i in range(len(polynomial)):
    z = polynomial[i](x_grid, y_grid)
    fig = plt.figure()
    ax = fig.add_subplot(projection='3d')
    ax.plot_surface(x_grid, y_grid, z)
plt.show()

Expected behavior
The issue can be solved by adding the following.

polynomial = polynomial + (q0 + q1) * 1e-100

This feels like a hack though.

Desktop (please complete the following information):

• OS: Linux CentOS
• Python version: 3.9.12
• Chaospy version 4.3.13

[jonathf]

Iterating over a polynomial array of 2 dimensions reduces to 1 dimensions when there the other variable isn't present. In other words: [1, x, y, xy] = [f1(), f2(x), f3(y), f4(x, y)]. You will notice the difference if you change p(x, y) with a more explicit p(q0=x, q1=y).

I agree that isn't the best option, and I will update the default in the next release. For now you can fix your problem by setting the following option at the top of you code:

chaospy.set_options(retain_names=True)