Fastest way to parse JSON strings into numpy arrays

Question

I have huge json objects containing 2D lists of coordinates that I need to transform into numpy arrays for processing.

However using json.loads followed with np.array() is too slow.

Is there a way to increase the speed of creation of numpy arrays from json?

import json
import numpy as np

json_input = '{"rings" : [[[-8081441.0, 5685214.0], [-8081446.0, 5685216.0], [-8081442.0, 5685219.0], [-8081440.0, 5685211.0], [-8081441.0, 5685214.0]]]}'

dict = json.loads(json_input)
numpy_2d_arrays = [np.array(ring) for ring in dict["rings"]]

I would take any solution whatsoever!

I'm getting a `json` decoder error: ` File "/usr/lib/python3.5,... json.decoder.JSONDecodeError: Expecting value: line 1 column 14 (char 13)` — hpaulj, Dec 09 '16 at 23:15
Other parsers are `eval` and `ast.literal_eval` (safer). On this small sample `json.loads` is noticeably faster. The `np.array` part takes even less time. — hpaulj, Dec 10 '16 at 01:32

score 5 · Answer 1 · edited May 23 '17 at 12:24

5

The simplest answer would just be:

numpy_2d_arrays = np.array(dict["rings"])

As this avoids explicitly looping over your array in python you would probably see a modest speedup. If you have control over the creation of json_input it would be better to write out as a serial array. A version is here.

edited May 23 '17 at 12:24

Community

1
1

answered Dec 09 '16 at 21:19

Daniel

18,441
7
56
74

score 3 · Answer 2 · answered Dec 11 '16 at 15:58

Since JSON syntax is really near to Python syntax, I suggest you to use ast.literal_eval. It may be faster…

import ast
import numpy as np

json_input = """{"rings" : [[[-8081441.0, 5685214.0],
                             [-8081446.0, 5685216.0],
                             [-8081442.0, 5685219.0],
                             [-8081440.0, 5685211.0],
                             [-8081441.0, 5685214.0]]]}"""

rings = ast.literal_eval(json_input)
numpy_2d_arrays = [np.array(ring) for ring in rings["rings"]]

Give it a try. And tell us.

Thank you but ast.literal_eval is slower than json.loads() with my data — Below the Radar, Dec 12 '16 at 15:24

score 0 · Answer 3 · answered Dec 09 '16 at 21:29

0

For this specific data, you could try this

import numpy as np

json_input = '{"rings" : [[(-8081441.0, 5685214.0), (-8081446.0, 5685216.0), (-8081442.0, 5685219.0), (-8081440.0, 5685211.0), (-8081441.0, 5685214.0)]]}'
i = json_input.find('[')
L = eval(json_input[i+1:-2])
print(np.array(L))

answered Dec 09 '16 at 21:29

Gribouillis

2,122
1
8
14

using eval is unsafe – ei-grad Jul 09 '20 at 09:39
@ei-grad It depends on the context. Python has always been made for consenting adults! – Gribouillis Aug 25 '20 at 09:56

Fastest way to parse JSON strings into numpy arrays

3 Answers3

Linked