c-f-h / flamp Goto Github PK

Compare with lstsq.

import flamp
import numpy as np
A = np.array([[0, 1]]).T
b = np.array([2, 4])
print(np.linalg.lstsq(A, b, rcond=None)[0])
print(flamp.qr_solve(A, b))

Also try setting A = np.array([[0, 1], [1, 1]]).T.

For a complex numpy vector, e.g., a = (1+1j)*np.ones(2), the properties a.real and a.imag return vectors with the entry-wise real and imaginary parts of a. For a flamp vector, e.g., a = (1+1j)*flamp.ones(2), this is not working properly. a.real and a.imag return [mpc('1.0+1.0j',(216,216)) mpc('1.0+1.0j',(216,216))] and [0 0], respectivelly. As a workaround, calling real and imag on the vector entries works as expected, thus, [z.real for z in a] and [z.imag for z in a] return the proper vectors.

This snippet shows the difference

> flamp.linspace(1, 2, num=1)
array([mpfr('nan')], dtype=object)
> np.linspace(1, 2, num=1)
array([1.])