带有dtype = int32的值列表转换为np.array对象的Dict

Question

I have a dictionary {0: [], 1: [], ...} . The length of every list is different.
How to convert such a dictionary to np.array object?
I want to get such a structure: array([[],[],..., dtype=int32)

Answer 1

have you tried this ?

np.array(list(d.values()))

ouput

array([], shape=(2, 0), dtype=float64)

Answer 2

Strangely, putting the dict into the numpy array works. Also putting the listed values into the array works.

import numpy as np

d = {0: [1, 2], 1: [3, 6, 7]}
arr = np.array(d)
print(arr)
li = d.values()
print(li)
arr2 = np.array(li)
print(arr2)

Answer 3

If the lists are of different lengths you could use a np.ma.MaskedArray :

import numpy as np

maxlen = max(len(lst) for lst in d.values())   # maximum length of all value-lists

arr = np.ma.MaskedArray(np.zeros((len(d), maxlen), dtype='int32'), 
                        mask=True, 
                        dtype='int32')

for line, lst in d.items():
    arr[line, :len(lst)] = lst

# In case the "keys" of your dictionary don't represent the "rows" you need to
# use another (more robust) approach:
# for idx, (line, lst) in enumerate(sorted(d.items())):
#     arr[idx, :len(lst)] = lst

For example:

d = {0: [], 1: [1], 2: [1, 2]}

Gives an arr like this:

masked_array(data =
 [[-- --]
 [1 --]
 [1 2]],
             mask =
 [[ True  True]
 [False  True]
 [False False]],
       fill_value = 999999)

It won't be the same as having empty lists but NumPy doesn't support ragged arrays so MaskedArray is one of the possible ways to "simulate" them. For most purposes they will behave like an "ragged array". :)