Numpy convert scalars to arrays

Question

I am evaluating arbitrary expressions in terms of an x array, such as 3*x**2 + 4. This normally results in an array with x's shape. However if the expression is just a constant, it returns a scalar. What is the best way to ensure it has x's shape without explicitly checking the shape? Multiplying by numpy.ones(x.shape) works, but I think that uses unnecessary computations.

Edit: To be clear, I don't just want it to be an array with size one, I want it to be the same shape and size as X. I'm evaluating a string using NumExpr which can contain an arbitrary function of x:

x = numpy.linspace(min, max, num)
y = numexpr.evaluate(expr, {'x': x}, {})

I want to get an array of y-values that could be plotted against x through matplotlib. Currently I am doing this, which works fine:

y = numpy.ones(x.size) * y

But I'm worried that this is wasteful for large sizes. Is there a better way?

Answer 1

See atleast_1d:

Convert inputs to arrays with at least one dimension.

>>> import numpy as np
>>> x = 42  # x is a scalar
>>> np.atleast_1d(x)
array([42])

>>> x_is_array = np.array(42)  # A zero dim array
>>> np.atleast_1d(x_is_array)
array([42])

>>> x_is_another_array = np.array([42])  # A 1d array
>>> np.atleast_1d(x_is_another_array)
array([42])

>>> np.atleast_1d(np.ones((3, 3)))  # Any other numpy array
array([[ 1.,  1.,  1.],
       [ 1.,  1.,  1.],
       [ 1.,  1.,  1.]])