convert string to 2d numpy array

Question

I am trying to convert 'b' (a string in which the column entries are separated by one delimiter and the the rows are separated by another delimiter) to 'a' (a 2d numpy array), like:

b='191.250\t0.00\t0\t1\n191.251\t0.00\t0\t1\n191.252\t0.00\t0\t1\n'
a=numpy.array([[191.25,0,0,1],[191.251,0,0,1],[191.252,0,0,1]])

The way I do it is (using my knowledge that there are 4 columns in 'a'):

a=numpy.array(filter(None,re.split('[\n\t]+',b)),dtype=float).reshape(-1,4)

Is there a better way?

• OMRY VOLK: what is wrong with your way?
• A B: Nothing, my way works, I just cobbled it together from stuff I found online. I was just wondering if there was a "more correct" way of doing it, a way to do it without using regular expressions, and/or a way to do it without knowing a priori the number of columns.

Answer 1

Instead of splitting and filtering, you could use np.fromstring:

>>> np.fromstring(b, sep='\t').reshape(-1, 4)
array([[ 191.25 , 0. , 0. , 1. ],
 [ 191.251, 0. , 0. , 1. ],
 [ 191.252, 0. , 0. , 1. ]])

This always returns a 1D array so reshaping is necessary.

Alternatively, to avoid reshaping, if you already have a string of bytes (as strings are in Python 2), you could use np.genfromtxt (with the help of the standard library's io module):

>>> import io
>>> np.genfromtxt(io.BytesIO(b))
array([[ 191.25 , 0. , 0. , 1. ],
 [ 191.251, 0. , 0. , 1. ],
 [ 191.252, 0. , 0. , 1. ]])

genfromtxt handles missing values, as well as offering much more control over how the final array is created.

Answer 2

Here's what I did to get the result you're looking for:

import numpy as np
b='191.250\t0.00\t0\t1\n191.251\t0.00\t0\t1\n191.252\t0.00\t0\t1\n'
a = np.array([[float(j) for j in i.split('\t')] for i in b.splitlines()])