Python - Convert numpy array to JSON

Question

I have a Python numpy array called myarray that looks like this..

[[148 362]
 [153 403]
 [163 443]
 [172 483]
 [186 521]
 [210 553]
 [239 581]
 [273 604]
 [314 611]
 [353 602]]

I want to create JSON that looks like this..

myjson = [
 {'section': '3',
 'x': '163',
 'y': '362',
 },
 {'section': '7',
 'x': '239',
 'y': '581',
 },
 {'section': '10',
 'x': '353',
 'y': '602',
 },
]

This represents the 3rd, 7th and 10th line in the original numpy array. Does anybody have an example of something similar being achieved?

• cs95: Who selects the arrays to be serialised?
• cs95: Can you please let me know? This affects the answer you get. When should the filtering be done? Before or during serialisation? I am not sure. Or is this just an example, and you want all rows with indices in this manner?

Answer 1

To add on to Andrii's answer, I believe you can also unpack the arrays like this to make it slightly cleaner:

[{'section': i+1, 'x': x, 'y': y} for i, [x, y] in enumerate(myarray) if i in [2, 6, 9]]

• fightstarr20: When I try and serialise that I get an error message - TypeError: Object of type 'int64' is not JSON serializable - I am assuming it is because it is numypy values instead of python int
• andnik: @fightstarr20 , you're right. I believe solution with pd.DataFrame is better, but if you want to make this work you need to wrap x and y in int(x) and int(y).

Answer 2

If your input array is arr I believe you want something like:

[{'section': i+1, 'x': x[0], 'y': x[1]} for i, x in enumerate(arr) if i in [2, 6, 9]]

[2, 6, 9] are your [3, 7, 10] positions, only starting from 0.

Answer 3

pandas presents a convenient solution:

import pandas as pd
import numpy as np
df=pd.DataFrame(myarray, columns=["x", "y"])
df["Section"]=df.index
df.to_json(orient="records")

this yields:

 '[{"x":148,"y":362,"Section":0},{"x":153,"y":403,"Section":1},
{"x":163,"y":443,"Section":2},{"x":172,"y":483,"Section":3},
{"x":186,"y":521,"Section":4},{"x":210,"y":553,"Section":5},
{"x":239,"y":581,"Section":6},{"x":273,"y":604,"Section":7},
{"x":314,"y":611,"Section":8},{"x":353,"y":602,"Section":9}]'

The solution is a little bit different, but you can easily find your way in pure python to convert it into your structure.

Answer 4

I needed something similar but for a more general case. This replaces any numpy ndarray in a python dict, list, set or any mix of them recursively either in place or should return a numpyified copy.

from copy import deepcopy
import numpy as np
def unnumpyify(something, inplace=False):
 if inplace:
 new_something = something
 else:
 new_something = deepcopy(something)
 if type(new_something) == np.ndarray:
 new_something = new_something.tolist()
 if type(new_something) == dict:
 for k in new_something:
 if type(new_something[k]) == np.ndarray:
 new_something[k] = new_something[k].tolist()
 if type(new_something[k]) == dict:
 new_something[k] = unnumpyify(new_something[k])
 elif type(new_something[k]) == list:
 new_something[k] = unnumpyify(new_something[k])
 elif type(new_something) == list:
 for i,k in enumerate(new_something):
 if type(k) == np.ndarray:
 new_something[i] = k.tolist()
 if type(k) == dict:
 new_something[i] = unnumpyify(k)
 elif type(k) == list:
 new_something[i] = unnumpyify(k)
 elif type(new_something) == set:
 for k in new_something:
 if type(k) == np.ndarray:
 new_something.remove(k)
 new_k = k.tolist()
 new_something.add(new_k)
 if type(k) == dict:
 new_something.remove(k)
 new_k = unnumpyify(k)
 new_something.add(new_k)
 elif type(k) == list:
 new_something.remove(k)
 new_k = unnumpyify(k)
 new_something.add(new_k)
 return new_something