NumPy: convert decimals to fractions

Question

I compute the reverse of matrix A, for instance,

import numpy as np

A = np.diag([1, 2, 3])
A_inv = np.linalg.pinv(A)
print(A_inv)

I got,

[[ 1.          0.          0.        ]
 [ 0.          0.5         0.        ]
 [ 0.          0.          0.33333333]]

But, I want this,

[[ 1.          0.          0. ]
 [ 0.          1/2         0. ]
 [ 0.          0.          1/3]]

I tried np.set_printoptions,

import fractions
np.set_printoptions(formatter={'all':lambda x: str(fractions.Fraction(x))})
print(A_inv)

but I got this,

[[1 0 0]
 [0 1/2 0]
 [0 0 6004799503160661/18014398509481984]]

How do I convert decimals to fractions in NumPy?

Answer 1

This is a floating point issue - recall that 2/3 is not exactly 2/3 in Pythons representation.

The Fraction class has a built in method, limit_denominator(), to take care of this:

import fractions
np.set_printoptions(formatter={'all':lambda x: str(fractions.Fraction(x).limit_denominator())})
print(A_inv)

Which gives the desired answer:

[[1 0 0]
 [0 1/2 0]
 [0 0 1/3]]